48
Working Paper: Understanding Rural Teacher Retention and the Role of Community Amenities Luke C. Miller* * University of Virginia 405 Emmet Street, P.O. Box 400277 Charlottesville, VA 22904 [email protected] Updated 6 September 2012. Center on Education Policy and Workforce Competitiveness University of Virginia PO Box 400879 Charlottesville, VA 22904 CEPWC working papers are available for comment and discussion only. ey have not been peer-reviewed. Do not cite or quote without author permission. I am tremendously grateful to the research team of Don Boyd, Hamilton Lankford, Susanna Loeb, and James Wyckoff for providing me with access to the New York teacher data. Daphna Bassok, Robert Costrell, Pam Grossman, Eric Hanushek, Susanna Loeb, and Sean Reardon provided useful feedback on earlier versions of this analysis as did seminar participants at the American Education Finance Association annual conference, Abt Associates, SRI International, and the Urban Institute. All errors are attributable to the author. CEPWC Working Paper Series No. 1. October 2012. Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia Two oft-cited assertions about rural teacher labor markets are examined in this study: first, rural schools have a harder time retaining teachers than non-rural schools and, second, their efforts to retain teachers are complicated by relatively poor community amenities. Administrative data on first-time teachers in New York State between 1994 and 2002 are combined with data from a large number other sources capturing community amenities such as geographic isola- tion, access to medical services, access to family networks, availability of shopping, and socio- economic health. The results support both claims and highlight important heterogeneity among rural communities. Center on Education Policy and Workforce Competitiveness

Working Paper: Understanding Rural Teacher Retention and

  • Upload
    others

  • View
    13

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Working Paper: Understanding Rural Teacher Retention and

Working Paper:Understanding Rural Teacher Retention and

the Role of Community AmenitiesLuke C. Miller*

* University of Virginia405 Emmet Street, P.O. Box 400277

Charlottesville, VA [email protected]

Updated 6 September 2012.

Center on Education Policy and Workforce Competitiveness

University of Virginia PO Box 400879

Charlottesville, VA 22904

CEPWC working papers are available for comment and discussion only. They have not been peer-reviewed. Do not cite or quote without author permission.

I am tremendously grateful to the research team of Don Boyd, Hamilton Lankford, Susanna Loeb, and James

Wyckoff for providing me with access to the New York teacher data. Daphna Bassok, Robert Costrell, Pam Grossman, Eric Hanushek, Susanna Loeb, and Sean Reardon provided useful feedback on earlier versions of

this analysis as did seminar participants at the American Education Finance Association annual conference, Abt Associates, SRI International, and the Urban Institute. All errors are attributable to the author.

CEPWC Working Paper Series No. 1. October 2012.Available at http://curry.virginia.edu/research/centers/cepwc/publications.

Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Two oft-cited assertions about rural teacher labor markets are examined in this study: first, rural schools have a harder time retaining teachers than non-rural schools and, second, their efforts to retain teachers are complicated by relatively poor community amenities. Administrative data on first-time teachers in New York State between 1994 and 2002 are combined with data from a large number other sources capturing community amenities such as geographic isola-tion, access to medical services, access to family networks, availability of shopping, and socio-economic health. The results support both claims and highlight important heterogeneity among rural communities.

Center on Education Policy and Workforce Competitiveness

Page 2: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

1 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

UNDERSTANDING RURAL TEACHER RETENTION AND THE ROLE OF COMMUNITY AMENITIES By Luke C. Miller, University of Virginia

1. Introduction

In surveys of rural administrators and teachers, relatively poor community amenities are

often cited as significant barriers to attracting and retaining highly skilled teachers (e.g.,

Schwartzbeck et al., 2003; NASBE, 2004; Hammer et al., 2005). Given the current policy

environment’s dual emphasis on teacher quality and student performance, it is imperative for

policymakers and rural administrators to reevaluate and strengthen teacher retention policies. A

recent overview of rural education in America found gaps in key student achievement outcomes

between rural and suburban students (Provasnik et al., 2007). Fewer rural students met the 2005 4th

and 8th grade NAEP proficiency standards in reading and mathematics than suburban students.

Rural students are also less likely to graduate (75 versus 79 percent) and to enroll in college (27

versus 37 percent) than non-rural students. The prevailing logic, prominently embodied in No Child

Left Behind, holds that improving teacher quality will help close these student achievement gaps.

This paper provides useful information to policy makers who seek to direct resources

intended to improve teacher retention to the neediest rural schools and communities. A host of

policies at the federal and state levels acknowledge the connection between quality teachers and

student performance. Generally, these policies provide financial incentives for teachers to accept

positions in specific hard-to-staff schools (e.g., high poverty or low performing) or in critical-

shortage subjects (e.g., mathematics, science, and special education). The federal government

operates Perkins and Stafford student loan forgiveness programs for teachers in eligible positions.

The financial incentives at the state level are most frequently education loan forgiveness or salary

bonuses; however some provide housing or tuition assistance. A review of state teacher policies

Page 3: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

2 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

found that almost every state has some form of teacher retention incentive policy on the books

(Loeb & Miller, 2006). In eleven states, these policies target rural schools specifically.

Framed by hedonic wage theory and building on exploratory studies suggesting community

characteristics play a role in teacher career paths, I ask: how are characteristics of rural schools and

communities related to the transfer and quit decisions of rural teachers, on average, and how do these relationships vary

systematically across teachers of different subjects? Communities are characterized by the amenities they

offer, including access to other communities and professional networks, shopping, family networks,

housing, and other employment opportunities. I take New York as a case study. Although not

viewed as a rural state, a significant portion of America’s rural students are educated within its

borders. More than half of all rural students in this country attend school in just 12 States which are

among the most populous and urban in the nation (Johnson & Strange, 2007). New York has the

country’s eighth largest rural student population. Therefore, if we were to ignore rural education in

non-rural states like New York, we would be ignoring the context in which many rural students are

educated.

Rural schools are not a homogeneous group, and the teacher retention challenges they face

are varied. Using data collected from more than two dozen sources, my results highlight observable

and measurable characteristics of rural schools and communities that predict lower retention rates.

Of particular interest, the results demonstrate the influence of richer community amenities in teacher

retention, even after accounting for differences in salary, non-wage job attributes, and opportunity

costs.

This paper is laid out in five sections. I briefly review the teacher retention literature to

highlight key predictive factors to be included in my analysis. The second section describes the data

collected. Here I detail how I differentiate between rural and non-rural schools and how I map

amenities onto school communities. In the third section, I summarize the hedonic wage theory of

Page 4: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

3 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

job selection which guides my analytic strategy. I present my results in the fourth section and discuss

policy implications in the final section.

2. Factors Predicting Teacher Retention

There is a rather large and growing body of research dedicated to exploring factors that

predict teacher retention. Retention is not consistently defined across these studies. Some examine

the predictive factors of retaining teachers within specific schools (e.g., Boyd et al., 2005b; Hanushek

et al., 2004; Ingersoll, 2002; Murnane, 1981; Scafidi et al., 2007) and others focus on retention within

specific districts (e.g., Baugh & Stone, 1982; Gritz & Theobald, 1996; Imazeki, 2005; Mont & Rees,

1996; Murnane, 1981; Murnane, 1984; Ondrich et al., 2008; Theobald, 1990). Still others consider

retention as remaining within a specific state’s teacher labor force (e.g., Feng, 2005; Grissmer &

Kirby, 1992; Murnane & Olsen, 1989 and 1990; Murnane et al., 1988) or within a nation’s teacher

labor force (e.g., Dolton & van der Klaauw, 1999; Stinebrickner, 1998, 1999, 2001, and 2002).

Collectively, they demonstrate the power of wage, opportunity costs, non-wage attributes,

and teacher characteristics in predicting teacher retention. Teachers are more likely to remain

teaching when they earn higher salaries (Grissmer & Kirby, 1992; Murnane & Olsen, 1989 & 1990)

and are more likely to transfer as salaries in other districts increase relative to their own (Baugh &

Stone, 1982; Imazeki, 2005; Ondrich et al., 2008). Expectations of higher future wages in teaching

also predict higher retention (Imazeki, 2005; Stinebrickner, 2001). Teachers are more likely to leave

teaching if they face higher opportunity costs (i.e., higher wages in non-teaching positions) (Dolton

& van der Klaauw, 1999; Murnane & Olsen, 1989 & 1990; Ondrich et al., 2008).

These studies demonstrate that the neediest schools have particularly acute problems with

teacher retention. Teachers are more likely to leave schools with low student achievement, high

percentages of minority students, high poverty levels, and greater student discipline problems (Boyd

et al., 2005b; Feng, 2005; Hanushek et al., 2004; Scafidi et al., 2007).

Page 5: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

4 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Specific characteristics of teachers, such as the subject they teach and their educational

background also predict retention rates. Teachers of specific subjects such as mathematics, science,

and special education have been found to have lower retention levels (Grissmer & Kirby, 1992;

Murnane et al., 1988). Additionally, Mont & Rees (1996) find that teaching outside of certified

subject area increases the probability of the teacher leaving. Higher ability teachers are more likely to

transfer and quit teaching (Boyd et al., 2005b; Stinebrickner, 2001).

Prior studies highlight key factors to include in any analysis of teacher retention; however,

they provide only minimal insight on how the teacher retention picture in rural schools may differ

from that in non-rural schools. In some, the sole focus is non-rural teachers (e.g., Boyd et al., 2005b;

Murnane, 1981 and 1984; Ondrich et al., 2008). Average retention differences between rural and

non-rural teachers, where measured at all, enter these analyses as a single covariate with mixed

results. Baugh & Stone (1982) include a dummy variable to indicate if the teacher’s district is located

within a metropolitan statistical area (MSA) and find teachers in an MSA (i.e., non-rural teachers) are

more likely than teachers outside an MSA (i.e., rural teachers) to change occupations in 1977-78, but

found no difference in the 1974-75 sample. Gritz & Theobald (1990) find that teachers were less

likely to remain in schools located more than 30 miles from an urban area (i.e., rural teachers) than

teachers located within 30 miles (i.e., non-rural teachers); however, this relationship was significant

for female teachers only. Imazeki’s (2005) analysis of Wisconsin teachers found rural teachers were

no more or less likely to remain in their current district than teachers in other non-rural districts.

Mont & Rees (1996) find that teachers in New York districts with a lower percentage of students

living in an urban area (i.e., more rural districts) are more likely to leave their district; however, the

relationship is not significant. In their study of Missouri teacher retention, Podgursky et al. (2004)

include an indicator of whether or not the school was located in a rural community but do not

present the coefficients in the paper.

Page 6: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

5 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

The literature focusing specifically on rural teacher retention is both smaller in number and

much less rigorous in the analytic techniques employed. Whereas almost all the studies described

above use multivariate techniques to model teacher retention as a function of wage, non-wage

attributes, opportunity costs, and teacher characteristics, the rural-focused studies rely on descriptive

statistics of survey data from administrators and teachers to highlight potential causes of low teacher

retention in rural schools.

One common theme evident in this literature and absent from the general teacher retention

literature discussed above is the role of community amenities in aiding/hindering teacher retention.

Rural communities are often lauded for the benefits of a rural lifestyle they offer such as a healthier,

quieter, safer lifestyle; a good place to raise children; a small, caring community; and an abundance

of clear, open spaces (Boylan et al., 1993). However, those benefits of the rural lifestyle come with

some substantial disadvantages which are theorized to pose barriers to retaining teachers in rural

classrooms. Isolation, be it geographical, social, or professional, frequently is mentioned as one such

disadvantage of the rural lifestyle (Boylan & McSwan, 1998; Boylan et al., 1993; Hammer et al., 2005;

Massey & Crosby, 1983; Murphy & Angleski, 1996/1997; Schwartzbeck et al., 2003). As jobs in the

traditional rural industries of farming, mining, and lumbering continue to disappear and the creation

of high-paying service or technical jobs in rural areas lags, rural labor markets in general have

become more disadvantaged. Consequently, couples are facing increasing difficulties securing stable,

satisfactory employment in rural communities (Boylan et al., 1993; Green, 2005; Schwartzbeck et al.,

2003). Rural communities also tend to provide their residents with fewer shopping venues (Murphy

& Angleski, 1996/1997), fewer cultural activities (Boylan et al., 1993), weaker access to health

services (Boylan, et. al, 1993; Kleinfield & McDiarmid, 1986), and a lack of adequate housing

(Schwartzbeck et al., 2003). However, retention rates are believed to be higher when teachers have a

stronger sense of connectedness to the community resulting from either having a strong family

Page 7: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

6 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

network within the community or having been made to feel welcome and a part of the community

by other community members (Bornfeld et al., 1997; Boylan & McSwan, 1998; McClure & Reeves,

2004; Murphy & Angleski, 1996/1997; Storey, 1993).

The present study marks the first time these two bodies of literature have been combined. It

models rural teacher retention decisions as a function of wages, non-wage attributes, opportunity

costs, teacher characteristics, and community amenities. My analysis is framed by the hedonic wage

theory of job selection.

3. Data and Methodology

I examine the effect of community amenities on the career paths of teachers in rural teacher

labor markets using data from New York State. My analysis focuses on full-time teachers who began

their careers in public schools outside New York City and Long Island between the 1993-1994 and

2001-2002 school years (1994 to 2002)—a 9-year period. This yields a sample of 22,698 teachers,

10,024 (45 percent) of which began their careers in rural schools (Table 1).

(Insert Table 1 about here)

These teachers are a subsample of teachers observed in administrative data compiled

annually by the New York State Department of Education on all teachers in all public New York

State schools. The data allows for individual teachers to be followed from year to year. The data

includes information on teacher characteristics, their wages, and the non-wage attributes of their

schools. Teacher characteristics include gender, age, ethnicity and race, competitiveness of the

undergraduate institution attended, educational attainment, performance on teacher certification

exams, teacher certification status, and subject(s) taught. All salaries were converted to 2004 real

dollars. Non-wage attributes include school level (elementary, middle, or high school), enrollment,

student-teacher ratios, and aggregate student body characteristics (percent minority, eligible for free

or reduced-price lunch, and limited English proficient).

Page 8: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

7 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

I calculate five measures of the opportunity costs of being a teacher. The first is the number

of schools within 20 miles is intended to be a proxy for the costs of transferring jobs (i.e., teachers

are less likely to incur the costs of changing residences when transferring to nearby schools).

Opportunity costs rise with the number of nearby schools. Following the approach taken by Imazeki

(2005), I calculate three within-region measures of relative teacher salaries. New York State

Department of Education groups New York’s 62 counties into 13 regions.1 The relative wage ratio

measures a teacher’s salary as a percentage of the regional average teacher salary.2 Higher values

indicate a teacher’s salary is larger relative to other teachers in his same region. The second and third

measures make use of a benchmark salary. The Master’s degree plus 10 years of experience

benchmark (MA+10 years) is the salary in the teacher’s district earned by his peers with a Master’s

degree and 10 years of district experience. This benchmark is included as a main effect but also as a

ratio to the regional average of the benchmark salary.3 The ratio represents the benchmark salary in

his current district as a percentage of the regional average salary for teachers with the same

qualifications. Both benchmark measures help capture the impact of a teacher’s expectations of

future earnings on career path decision. Opportunity costs decrease as each of these measures

increase. The final measure represents the salaries teachers could obtain in non-teaching position.

Using the 2000 Census Public Use Microdata Samples, I calculated the median annual wage earned

by college degree-holding workers aged 22 to 64 within the teacher’s county. Workers employed in

1 Open positions at other schools within the same region are a strong proxy for the other teaching jobs to which teachers consider transferring at the end of each school year. Between 1994 and 2001, 89 percent of observed teacher transfers were among schools within the same region. If same-district transfers are excluded (salaries are constant within districts), 57 percent of observed inter-district transfers were within the same region. Excluding teachers transferring from a New York City school where the district is contiguous with the region (hence inter-district transfers are across region), 72 percent of all inter-district transfers are within the same region. All relative salary measures are weighted by openings within the teacher’s subject area. 2 Regional average weighted by vacancies in a teacher’s subject. 3 Regional average weighted by vacancies in a teacher’s subject.

Page 9: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

8 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

the education sector are excluded from the calculations. Opportunity costs rise with non-teaching

wages.

I collected data on community amenities from extant data sources including data available on

the New York State and the National Center for Education Statistics (NCES) websites, the New

York State GIS Clearinghouse, the Economic Research Service (ERS) at the U.S. Department of

Agriculture, the U.S. Census Bureau, the U.S. Bureau of Labor Statistics, the U.S. Department of

Housing and Urban Development, and the U.S. Department of Commerce’s annual Zip Code

Business Patterns dataset. These data capture a broad set of the amenities a community offers its

residents. A particular strength of these data is the ability to calculate distances between schools and

community amenities rather than indicator variables for the presence of these amenities in a school’s

community. Below I review the process used to map amenities to schools.

3.1 Defining Rural

A challenge facing all rural research is the lack of a single definition of rural. The Rural

Policy Research Institute (2006) identified nine different rural definitions commonly used in

research. For this work, I triangulated three classification schemes to develop a six-point scale of

community type in New York State:

1. Non-metropolitan rural

2. Metropolitan rural

3. Suburban

4. Other Urban

5. Big Four City

6. New York City

The eight-category Johnson Locale Codes are included in the annual Common Core of Data

released by NCES. Using this scheme, I labeled schools as rural, suburban, or urban. Rural

Page 10: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

9 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

communities are not a homogenous group, and community characteristics are likely correlated with

how close the community is to an urbanized area. The nine-category county-level Beale Codes (or

urban-rural continuum codes) released by the U.S. Census Bureau allowed me to divide the rural

schools into rural schools in metropolitan counties and those in non-metropolitan counties. Finally,

the New York State Education Department’s annual Community Setting Codes separate the urban

category into schools in New York City, the Big Four Cities (i.e., Buffalo, Rochester, Syracuse, and

Yonkers), and other urban communities (e.g., Albany, Binghamton, Elmira, Glens Falls, Ithaca,

Middletown, Newburgh, Poughkeepsie, Rome, Schenectady, Troy, Utica, etc.).

Map 1 plots all public schools in New York State by their community type. The definition of

community type developed here visually aligns to how we generally view the rural-urban spectrum.

Suburban communities encircle urban centers. Outside suburban areas are the rural communities.

Non-metropolitan rural schools are concentrated in the Black River-St. Lawrence and Lake

Champlain-Lake George regions in the North Country and the three Southern Tier regions (west,

central, and east) along the border with Pennsylvania.

(Insert Map 1 about here)

During the 9-year period of this study, there were approximately 2.8 million students

educated by almost 180 thousand teachers in 42 hundred public schools administered by 705

districts. New York City Public Schools alone accounts roughly a third of all students, teachers, and

schools. Excluding New York City, a truly unique educational environment not only within the State

but also the nation, a substantial share of the New York State education system is situated in rural

communities. Approximately a third of students and teachers are in rural communities. Roughly 40

percent of schools are in rural communities and almost two-thirds of districts contain at least one

rural school.

3.2 Measuring Community Amenities

Page 11: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

10 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Guided by the exploratory work on rural teacher retention, I developed measures of the

following seven community amenities:

Geographic isolation: distance from the school to the nearest hub or primary airport4

Professional isolation: distance from the school to the nearest teacher education program5

Access to medical services: distance from the school to the nearest hospital

Access to family networks: distance from the school to the teacher’s hometown6

Availability of shopping: factor of distance-weighted sums of shopping establishments within

school’s community

Socio-economic health: factor of distance-weighted socio-economic indicators of the school’s

community

Adequate housing: distance-weighted average of fair-market rents for 2-bedroom apartments

within the school’s community

Principal component factor analysis was used to calculate the shopping and socio-economic health

factors. Shopping establishments were measured at the zip code level and includes hardware stores,

grocery stores, general merchandise stores, apparel stores, shoe stores, jewelry stores, drug stores,

book stores, sporting goods stores, and restaurants (alpha = 0.9896). The indicators of socio-

economic health are the community’s unemployment rate, percent of adults without a high school

4 Hub airports in New York are: Albany International, Buffalo-Niagara International, Greater Rochester International, John F. Kennedy International, LaGuardia, and Syracuse Hancock International. Primary airports in New York are: Binghamton Regional, Chautauqua County/Jamestown, Clinton County, Elmira/Corning Regional, Ithaca/Tompkins Regional, Long Island MacArthur, Oneida County, Stewart International, and Westchester County. 5 I included the 114 teacher education programs at non-proprietary four-year colleges and universities and graduate only institutions. 6 Information on a teacher’s hometown is the same information used in Boyd et al. (2005a). Either the zip code of the teacher’s high school or of the mailing address they provided when they applied to the SUNY system is used to indicate the teacher’s hometown. This information is missing for approximately a quarter of rural teachers in my sample.

Page 12: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

11 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

diploma, percent of adults with a college degree, median household income, youth poverty rates, and

average annual salaries (alpha = 0.8731).7

Extant data for these measures were available at three different levels—specific location or

address, zip code, and county—and each required a different process for mapping the amenities to a

school’s community, which I define as a 20-mile radius around the school.8 For the distance

measures, school and amenity addresses were geocoded and straight-line distances were calculated.

Data on shopping establishments and average salaries are all measured at the zip code level. To map

these to a school’s community, I calculated the distance from each school to the centroid of each zip

code within the school’s community. These distances were used to calculate the distance-weight

sums. These sums are more effective at capturing competition among goods and services providers

than would measures of the distance to the nearest establishment. Sums also may reflect quality to

the extent greater competition yields higher quality products. County-level data (e.g., the remaining

socio-economic health indicators and fair market rents) was first assigned to individual zip codes and

then mapped to school community’s following the process for zip code level data.9 For schools

within 20 miles of New York State borders, I include data from those neighboring state zip codes

(i.e., Vermont, Massachusetts, Connecticut, New Jersey, and Pennsylvania) in the measures of

community amenities.

As Table 2 shows, there are significant differences in community amenities across

community type. Rural schools have the poorest set of amenities. They are farther on average from

7 Median household income pertains to the income earned by residents of the school’s community whereas average salaries pertain to the salaries paid out by establishments within the school’s community. 8 20 miles was chosen as available evidence suggests this distance is a good measure of the geographic size (or “localness”) of teacher labor markets. Boyd et al. (2005a) found that 65 percent of beginning teachers in New York took a first job in a school within 15 miles of their hometown. Using nationally-representative data, Reininger (2006) finds that teachers are much more likely to work within 20-miles of their hometown eight years after graduating than are workers in almost forty other professions. 9 In New York, not all zip codes are fully contained within counties. The website www.melissadata.com indicates the percentage of a zip code’s addresses that are within each county. For those zip codes crossing county boundaries, the distance-weighted data were multiplied by these percentages.

Page 13: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

12 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

an airport, teacher education program, and hospital. They also have the fewest shopping

opportunities and the poorest socio-economic health. For example, non-metropolitan rural schools

are on average 37 miles from an airport versus 12 miles for suburban schools and the shopping

opportunities are on average one standard deviation below the state average whereas suburban

communities are 0.6 of a standard deviation above the state average. Teachers in these rural schools

are also 14 miles farther from their hometowns on average than teachers in suburban schools (45

versus 31 miles). However, these rural communities have cheaper rents than non-rural communities

($588 in non-metropolitan rural communities versus $789 in suburban communities) which may

help them retain teachers at least to the extent that the rents reflect factors beyond the poorer

amenities available in rural areas and to the extent the amenities I measure here reflect the amenities

set reflected in rents.10 Distance to the nearest airport, hospital, teacher education program, and

teacher’s hometown and fair market rents are hypothesized to have positive relationships with

teacher attrition while the shopping and socio-economic health amenities are hypothesized to have a

negative relationship.

(Insert Table 2 about here)

Before proceeding, my hypothesis that teachers prefer cheaper rents deserves additional

discussion. Rents are a function of the demand for and supply of rental units. Demand for rental

units in turn is partly a function of the market value of the amenities available in the local

community. Therefore, communities with richer amenities such as closer proximity to an airport,

more numerous shopping venues, and better socio-economic health should be expected to have

higher rents. This correlation is evident in Table 2. However, it’s a safe assumption that given two

communities with identical amenities, teachers on average will prefer the community with cheaper

rents. A model relating rents to teacher retention therefore must capture the value of these

10 Throughout this analysis, all dollar statistics such as rent, salaries, and incomes have been converted to 2004 dollars.

Page 14: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

13 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

amenities. Otherwise, the coefficient on rents would suggest teachers have preferences for more

expensive rents when in fact the coefficient is measuring their preferences for the richer amenities

reflected in the higher rents. To the extent that the variation in the amenities I measure capture the

variation in the amenity set factored into rents, my estimated coefficient on rents reflects teacher

preferences for rents net of their preferences over other amenities.11

Geospatial plots of the variation among rural schools on these seven community amenities

reveal a common pattern. With the exception of distance from hometown and monthly rents,

amenities are highest for rural schools located closest to major travel routes such as I-90 connecting

Albany and Buffalo. Thus, rural schools located in the North Country (northwest corner) and

Southern Tier regions (along border with Pennsylvania) tend to have the poorest amenities.

Accordingly, rents are lowest in these regions and increase with proximity to major interstate

highways. Map 2 plots the hometowns of beginning teachers who grew up in or near New York

State. A clear majority of beginning teachers grew up in New York. The density of dots aligns with

population density. As Map 3 demonstrates, the pattern of average distance from these hometowns

and their initial school is more random than for the other amenities such as distance to the nearest

airport.

(Insert Map 2 about here)

(Insert Map 3 about here)

3.3 School-Level Retention Model

I conduct to sets of analyses on teacher retention. In the first, I examine teacher retention

across community type in order to place New York State’s rural teacher retention in the broader

11 This also requires that teachers have the same preferences over amenities as other individuals with demand for rental units. Teacher demand for rental units is only one component of the overall demand. Therefore, if teacher preferences differ from the preferences of the average individual, my models will fail to capture the contribution to rents of renter preferences for amenities. I have no reason to believe that teachers have systematically different preferences for amenities than the average individual.

Page 15: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

14 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

state-wide context. Here I am interested in testing the hypothesis that rural teacher retention differs

from non-rural teacher retention. In these models, I incorporate the standard set of retention

factors—teacher characteristics, salary, school characteristics, and opportunity costs. I delve deeper

into rural teacher retention in the second set of analyses. Here, I restrict my sample to rural teachers

and include community amenities in the models. In these models, I test the hypothesis that teacher

retention is lower in rural communities with poorer community amenities. Both sets of analyses are

guided by the same model of school-level teacher retention.

Hedonic wage theory (Rosen, 1974; Mortensen, 1990; Hwang et al., 1998) is frequently used

by economists to model the labor decisions of workers. In brief, when faced with a set of job offers,

workers select the job offer that maximizes their utility and provides them with the greatest value.

Both utility and value are determined by the wage rate and the non-wage job attributes (e.g., working

environment, tasks performed, colleagues, etc.). Assuming there is a job offer with a value exceeding

their reservation wage, the worker will accept the job with the highest value. Otherwise, they will

choose no offer and be unemployed. Workers seek to maximize their utility both when selecting

their first job and when deciding whether or not to switch jobs. This analysis extends this theory to

include in the workers’ job offer valuation the amenities of the community in which the jobs are

located. I examine this valuation process by estimating a series of discrete-time competing-risk

hazard models. I detail both the theory and the model in turn below.

The concept of implicit prices for non-wage job attributes (and, by extension, community

amenities) is at the heart of hedonic wage theory. Each job offer is characterized by a wage (wj), a n-

dimensional vector (xj) of non-wage attributes and a n′-dimensional vector (zj) of amenities provided

by the school’s community. Implicit prices convert teacher preferences over these attributes and

amenities to monetary values so that they can be combined with the wage rate to determine the

value or utility of the job offer (1).

Page 16: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

15 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

(1) ),,|,,( mmmjjjmjmj qpAzxwvvU

The value (vmj) the jth job offer provides the mth teacher is determined by the wage rate, the non-job

attributes, and the community amenities given the teacher’s characteristics (Am) that affect his

valuation of the jth job offer. The teacher-specific implicit price vectors (pm and qm) reflect the his

preferences over attributes and amenities, respectively. Teachers will prefer the job offer that

provides the maximum value.

With respect to this retention analysis, a teacher’s career decision is determined by equation

2. It details the decision rules the teacher uses to select which job offer to accept among all available

to him (Jm). In the simplest sense, a teacher makes one of three career decisions at the end of each

academic term: remain at current school j; switch to a different school k; or leave the New York

teacher labor force. The teacher will be retained, R, at his current school if it continues to provide

the greatest value. Should he have a job offer from another school that he values more (i.e., provides

him more satisfaction than his current job), he will not be retained in his current school. Instead, he

will accept the maximum value offer and transfer, T, to the new school. He will leave, L, the teacher

labor force if none of the teaching job offers available to him exceed his reservation value (vm*).

In order to examine this annual valuation process, I estimate a series of discrete-time

competing-risk hazard models of the transfer and quit decisions of first-time teachers. I follow these

teachers until they separate from their initial school and thus observe at most one transition

[1] R if vmj ≥ vmk and vmj ≥ vm* for all k ≠ j Jm

(2) D = [2] T if vmk > vmj and vmk ≥ vm* for some k ≠ j Jm

[3] L if vmj < vm* for all j Jm

Page 17: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

16 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

outcome for each teacher.12 The decision to initiate a teaching spell at a second school is not random

and including these later spells would introduce a form of selection bias into my results.

I define the risk set to include the following three career decision alternatives (d) explained in

equation 2. At the end of each academic term, the teacher revalues of each of his three career path

decision alternatives (d=1, 2, 3). He selects the highest-value alternative. The conditional probability

that teacher m chooses to remain in his current school (d = 1) or elects to make a job transition (d =

2 or 3) at time t (given he has not yet made a transition) is:

(3)

)~)((exp1

)~)(exp(

))((exp

))(exp()|Pr(

3

2

3

1

d

mt

d

d

d

mt

d

d

mt

d

d

d

mt

d

ddd

mt

vt

vt

vt

vttTtT

Td represents the year of experience at the end of which the teacher is observed either transferring

schools (d=2) or leaving the teacher labor force (d=3).13 The notation 1~mt

d

mt

d

mt vvv

for d, d′ = 2 or

3 in the second expression indicates that the probability of a job transition is determined by the

attributes of alternative job opportunities to those of the current job. )(td is a function used to

determine how the probability of choosing a particular job decision changes with the number of

years (t) the teacher has been in his teaching spell. It allows for duration dependence. I assume the

following non-parametric form for the baseline hazard:

12 Data accuracy checks were conducted to identify and impute teacher records in cases where teachers were erroneously absent from the annual administrative datasets. For example, assume I observe a teacher in period t, but not the following year (period t + 1) even though he was still teaching in a New York public school but for whatever reason does not appear in the dataset. I would (erroneously) conclude that he left the New York teacher labor force at the end of the first period. But now assume I observe him in the data two years later (period t + 2). I would (erroneously) conclude that he returned to the classroom after a one-year hiatus. In all cases where the data suggest a teacher left teaching, I conducted data checks by looking two years out. If I observe the teacher returning to the classroom and his years of teaching experience increased by two years rather than one, I imputed the missing records. This helps prevent my attributing teacher career path decisions to these data inconsistencies. 13

When a teacher appears in the dataset in one year but not the following year, I assume he left the teacher labor force. The data do not allow me to differentiate between several career path decisions, all of which are coded as leaving in my analysis. He could have taken a non-teaching job, decided to teach at a private school in New York, or to take a teaching job in another state. It is also possible that a teacher returns to a public school classroom but on a part-time basis. The data allow me to observe this occurrence; however, part-time teachers comprise less than 5 percent of all teachers during the observation period and only a portion of these were previously full-time teachers.

Page 18: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

17 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

(4) )(...)2()1()( 21 ntItItIt d

n

ddd

where I is an indicator function equal to 1 if its argument is true. The number of terms in equation 4

depends on how many years a teacher is observed making a career path decision.14

The value at time t of a given decision alternative (d) to the mth teacher takes the following

form:

(5) d

t

d

mt

d

mt

d

mt

dd

mt czxv 3210

Here, the value of the alternative is a linear function of the teacher’s characteristics (xmt), the

characteristics of the initial job (zmt), and the characteristics (i.e., opportunity costs) of the alternative

job opportunities available to the teacher (cmt). The subscript t acknowledges that some

characteristics of the teacher, initial job and opportunity costs are time-variant. d

t is a calendar year

fixed effect. The scalar d

0 and vectors d

1 , d

2 , and d

3 are estimated.

In order to model transfer and quit decisions as one-sided, I assume all separations are

teacher-driven.15 This likely holds for teachers who have received tenure which is granted after

teaching for three years in the same district. Although the law (NY CLS Educ §§3014 and 3020)

allows superintendents and principals greater flexibility in dismissing probationary teachers, available

research suggests teachers are rarely involuntarily terminated. Data from Florida’s 2006-2007 teacher

exit interview show less than 5 percent of exiting teachers had their employment involuntarily

terminated (FLDOE, 2008). This percentage was likely lower prior to the strengthening of state

accountability systems which post-date most of the time period under study here.

14 For key covariates, I test if their relationship to teacher separation probabilities remains constant as the teacher gains teaching experience by interacting them with a continuous time trend variable that takes on integer values from 1 to 9 indicating depending on how many years of experience the teacher has at the time he is observed making a career path decision. Higher-order interactions are also explored to maximize model fit. 15 During this period, several schools closed. Teacher observations in these schools in these years are not included in the analysis has all these teachers did not have the option of remaining in their current position.

Page 19: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

18 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Observed school separations are also unlikely to have been the result of involuntary

reductions in force as the number of teachers employed increased between 1994 and 2002 in rural

and non-rural schools (Table 3). Overall, the sum full-time equivalent (FTE) across all teachers

increased by 16.7 percent in upstate New York with the metropolitan rural and suburban schools

experiencing the largest increases (20.6 percent each) and non-metropolitan rural schools the

smallest increase (8.5 percent). The size of entering beginning teacher cohorts, the focus of this

analysis, more than doubled during this time period. Cohorts increased among teachers of all

subjects with the exception of occupational education teachers which decreased by 5.9 percent

overall (not shown in table).

(Insert Table 3 about here)

Therefore, if a teacher is observed remaining at his current school, the analysis assumes the

teacher’s job offer set contained no job offers providing greater utility than their current position.16

A teacher moving to another school reveals that utility derived from the new position exceeded that

of the old position. Finally, when a teacher exits the teacher labor market, it signals that he faced no

teaching job offer that exceeded his reservation wage.

3.4 Limitations

Despite the wealth of data collected for this study, there are two key sources of bias for

which I may not have sufficiently controlled and which prevent me interpreting the results as causal

rather than correlational. The first is self-selection bias. There could be something unobserved about

the teachers related to their initial decision to accept jobs in rural schools and/or in poor amenity

rural communities that is also correlated with a higher propensity to separate. For example, it could

16

This analysis does not distinguish between teachers who remain in their same teaching post and those who change

posts within the same school. For example, teachers who move from teaching 2nd grade to teaching 3rd grade are considered to have remained in their current job. All movements are assumed to be voluntary decisions on the part of the teacher. However, within-school movements are often involuntary decisions made by school administrators in response changing enrollment patterns.

Page 20: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

19 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

be that beginning teachers do not want to be rural teachers yet rural positions were the only jobs for

which they received employment offers. They would have preferred teaching at another school, but

accept the rural teaching offer because they prefer to have any job versus being unemployed.

Consequently, they begin their careers more committed than other teachers to searching and getting

a job in another school. If more teachers in rural schools versus non-rural schools begin their careers

with this mentality, rural beginning teachers are predisposed to lower retention rates than teachers in

non-rural schools.

Another concern is bias introduced in the estimates from an omitted variable which is both

correlated with teachers’ labor market decisions and key covariates such as community amenities.

Marital status and fertility behavior are two sources of potential bias as they are in all but a handful

of studies. Stinebrickner (1998, 2001, 2002) finds lower retention rates for married teachers, higher

retention rates for teachers who are parents, and higher attrition rates for female teachers when they

have a newborn. I do not have any information on either marital status or fertility behavior; except

to the extent these are correlated with age. However, failure to include these in my model will only

bias my community type or community amenity estimates if marital status and fertility behavior

differ systematically between rural and non-rural schools or with different levels of amenities which

is not readily obvious.

4. Results

There are sizeable differences across community type in observed teacher career path

decisions. Figure 1 plots the percent of beginning teachers deciding to remain at their initial school

after each year of teaching by community type. At the extremes, suburban schools retain the most

teachers, and the schools in the Big Four Cities retain the fewest. However, non-metropolitan rural

schools also have substantial difficulties with teacher retention, retaining the second lowest percent

of beginning teachers. To characterize these difficulties, it helps to compare teacher retention in

Page 21: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

20 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

rural schools to that in suburban schools where retention rates are the highest. After their first year

of teaching, non-metropolitan rural schools have retained almost 2 percentage points fewer teachers

than suburban schools. This retention gap grows over the next three years as a significantly greater

percentage of non-metropolitan rural teachers compared to suburban teachers choose to leave their

initial schools. At the end of the fourth year, almost 7 percentage points more of the initial cohort

hired in the non-metropolitan rural schools have either transferred to another school or quit

teaching than is the case among the initial cohort hired in the suburban schools.

(Insert Figure 1 about here)

Non-metropolitan rural schools also have a harder time retaining teachers than rural schools

in metropolitan counties. At the end of the fourth year after the cohort of beginning teachers were

hired, non-metropolitan rural schools have needed to replace almost 4 percentage points more of

their beginning teachers than rural schools in closer proximity to urban areas.

There are also substantial differences in the retention of critical-shortage subject teachers

(i.e., mathematics, science, and special education) and multiple-subject teachers across community

type (see Table 4). At the end of third year after the beginning teachers were hired, non-

metropolitan rural schools compared to suburban schools have had to replace almost 15 percentage

points more of their mathematics teachers, five percentage points more of their science teachers, 4

percentage points more of their special education teachers, and 6 percentage points more of their

multiple-subject teachers. In fact, there are also substantial retention differences between rural

schools in metropolitan and non-metropolitan counties particularly for science and multiple-subject

teachers. Again, at the end of the third year after the beginning teachers were hired, non-

metropolitan rural schools compared to metropolitan rural schools have had to replace 6 percentage

points more of their science teachers and 8 percentage point mores of their multiple-subject

Page 22: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

21 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

teachers. The differences in retention rates between metropolitan rural and suburban schools are less

severe.

(Insert Table 4 about here)

The question remains, however, as to the extent to which these differences can be explained

by measureable differences across community type in teacher characteristics, wages, non-wage job

attributes, and opportunity costs or by unmeasured differences in the community types themselves.

Community type means (see Table 5) suggest these predictors of teacher retention work against each

other. For example, starting salaries are the lowest for rural teachers suggesting lower retention rates

relative to other community types. Yet, rural teachers face the lowest opportunity costs which theory

suggests should drive rural retention rates higher than in other community types. Teachers should

find rural schools more attractive than suburban schools on some non-wage job attributes such as

class size but less attractive with respect to others such as percent of students eligible for free or

reduced-price lunch.

(Insert Table 5 about here)

Rural beginning teachers also differ from their non-rural peers. They are more likely to be

male and non-white than teachers whose first job is in a non-rural school. A beginning rural teacher

is more likely to teach high needs subjects such as math and science (though beginning suburban

teachers slightly higher) and more likely to teach multiple subjects than non-rural first-time teachers.

Rural teachers are less likely to be teaching out-of-field (i.e., teaching subjects in which they are not

even provisionally certified) than peers in urban and Big Four Cities but more likely than their

suburban peers.

Below I present results from two sets of analyses. In the first, I test the hypothesis that rural

schools have a harder time retaining teachers even after accounting for differences in key

determinants of their retention decision. I then delve deeper into the rural labor market to test the

Page 23: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

22 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

hypothesis that rural schools with relatively poor amenities experience greater difficulties retaining

teachers than rural schools with a richer amenity set to offer teachers.

4.1 Predicting Retention Across Community Type

I estimate a series of four models to test the robustness of the relationships between

community type and teacher retention to the inclusion of more and more covariates. I begin with

the base model which I refer to as the problem description model. It includes the time trend (i.e.,

years of teaching experience), indicators of school community type where suburban communities are

the reference, and interactions between community type and the time trend. The interaction terms

capture the differential annual attrition rates across community type noted above. In the second

model, I add a full array of teacher characteristics including subject(s) taught. I add salary and non-

wage job attributes in the third model, and opportunity costs in the fourth model.

Coefficients on the community type covariates from the four models of beginning teacher

retention across community type are presented in Table 6 (full results are available from the author).

Each model includes the nine teaching experience indicator variables, a continuous time variable

interacted with community type (non-metropolitan rural, metropolitan rural, other urban, and Big

Four Cities), quadratic and cubic non-metropolitan-by-time interactions, quadratic Big Four Cities-

by-time interactions, and indicator variables for the four community types.17

(Insert Table 6 about here)

In order to better understand the effect of accounting for additional predictors on the

beginning teacher retention gap between suburban schools and schools in other communities, I

convert the estimated coefficient (i.e., log odds ratios) to predicted retention probabilities for each

community type. Figure 2 displays the differences in the predicted retention rates between suburban

17 This specification of the time trend was selected after iteratively adding the community type-by-time interactions and estimating a Wald test of joint significant (i.e., transfer vs. stay and quit vs. stay). Terms that failed the Wald test (i.e., were insignificant) were dropped and added higher-order terms were added when the Wald test was rejected.

Page 24: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

23 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

and non-metropolitan rural schools. For example, consider two cohorts of beginning teachers. One

cohort began their teaching careers in suburban schools and the other in non-metropolitan rural

schools. If three years later at the end of the school year we observe the schools at which each

cohort was initially employed, the problem description model predicts we would find that roughly 6

percentage points fewer members of the non-metropolitan rural cohort compared to the suburban

cohort had taught at their initial school in each of the preceding three years and had committed to

return to that same school for a fourth year. If we returned seven years later, we would find roughly

4 percentage points fewer of the original group of non-metropolitan rural cohort members

(according to the problem description model) agreeing to teach at the same school for an eighth

consecutive year than we would observe among the suburban cohort.

(Insert Figure 2 about here)

With respect to a specific cohort of first-time teachers, non-metropolitan rural schools are

predicted to retain fewer teachers than suburban schools after each of the first five years after the

cohort was initially hired. It is not until school characteristics (which includes salaries) are added to

the model that the survival functions for non-metropolitan rural and suburban schools cross (i.e.,

the difference between the two is zero). This occurs between the fifth and sixth years of the

teachers’ careers. This suggests differences in the salaries and non-wage job attributes of non-

metropolitan rural and suburban schools are major contributing factors to differences in retention

rates.

Metropolitan rural schools are predicted to have retained fewer teachers than suburban

schools at the end of each year regardless of the set of covariates included. These results, similar to

Page 25: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

24 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

non-metropolitan rural schools, demonstrate that much of the difference in retention rates from

suburban schools is driven by differences in salary and non-wage attributes.18

Turning attention to these other factors of teacher career paths, the results suggest they

influence labor market decisions in similar ways in rural and non-rural settings. Wald Tests

consistently indicated the relationship between a predictor (e.g., salary, opportunity costs, subject

taught, etc.) and a teacher’s decision to transfer or quit did not significantly differ across community

type. For example, the relative difficulty of retaining a math teacher versus a non-math teacher, for

example, is the same in a rural non-metropolitan school as a suburban school as a Big Four City

school.

Schools have a particularly challenging time retaining special education and multiple-subject

teachers. The full model predicts that special education teachers are 37 percent more likely

)05.;1(exp 3080 p to transfer from their initial school and 9 percent more likely to quit

)10.;1(exp 0830 p . Multiple-subject teachers are 30 percent more likely to transfer and 28 percent

more likely to quit (both p<.05). These are particular concerning for rural schools where special

education and multiple-subject teachers comprise a greater share of beginning teachers. I tested if

the relationship between subject taught and teacher retention varied with teaching experience. All

the interactions between subject taught and the continuous time variable failed to reject Wald Test

suggesting the relationship remains constant with experience.

Salary and non-wage job attributes such as enrollment and student demographics are all

significant predictors of teacher career paths. Teachers earning higher salaries are more likely to

remain at their initial school. A ten percent increase in teacher salaries is associated with a 29 percent

18 A pattern similar to that in rural non-metropolitan schools is observed in urban schools. In the full model, retention rates in urban schools exceed that in suburban schools in the 6th year and later. And while holding salary, non-wage job attributes, and opportunity costs constant between the Big Four Cities and suburban schools, the additional share of teachers leaving inner city schools relative to suburban schools grows monotonically with years of experience.

Page 26: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

25 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

decrease in the probability of quitting the teacher labor force (p<.01). Higher salaries are also

associated with a lower probability of transferring away from their initial school, though the

relationship is not significant. Additionally, beginning teachers at smaller schools are statistically

more likely to transfer and to quit (p<.01). Higher concentrations of minority students are associated

with a higher likelihood of quitting (p<.01), and higher rates of student poverty predict higher

transfer rates (p<.05).

Whereas the relationship between teacher salaries and their labor market decisions conforms

to theory and the literature, opportunity costs are less consistent with the theory. The coefficients

for non-teaching salaries suggest teachers’ are more likely to separate from their current school when

non-teacher college-educators workers in the same region earn higher wages. A ten percent increase

is associated with a 7 percent greater likelihood of transferring away from the initial school (p<0.01)

and a 5 percent greater likelihood of quitting (p<.05). This is less than a fourth of the teacher salary-

retention elasticity indicating teachers are much more responsive to their own wages than the wages

they could earn through outside employment.

Results for the relative teacher salary measures tend to contradict theory more than affirm it.

Coefficients for the relative wage ratio (i.e., teacher salary to average salary of all teachers within the

same region) indicate teachers who earn more relative to other teachers at nearby schools are less

likely to transfer (conforming to theory; insignificant) but more likely to quit (contradicting theory;

p<.01). Higher expected wages, as measured by the MA+10 years benchmark salary, are associated

with lower retention rates rather than higher retention rates as predicted. The estimated elasticity

indicates a ten percent increase is correlated with an 18 percent greater probability of quitting

teaching (p<0.01). The benchmark wage ratio (i.e., current district benchmark salary relative to the

average benchmark salary of nearby districts) is associated with higher transfer probabilities

(contrary to theory; insignificant) and lower quit probabilities (conforming to theory; p<.01). If the

Page 27: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

26 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

higher benchmark salaries indicate districts are backloading salary increases, these results could

indicate teachers heavily discount the future. Finally, having more schools within 20 miles is

associated with higher separation probabilities, which confirms the theory (i.e., more local schools

lowers transfer costs and increases transfer rates). This relationship, though statistically significant, is

very small.

4.2 Predicting Retention within Rural Schools

The above analysis of both rural and non-rural schools demonstrates persistent retention

gaps between rural and suburban schools, especially in the early years of a teacher’s career, even after

accounting for teacher characteristics, salary, non-wage job attributes, and opportunity costs.

Additionally, it showed non-metropolitan rural schools have a harder time retaining teachers than

rural schools in metropolitan counties. Might differences in amenities among rural communities

explain the variation in teacher retention at rural schools?

Again, I estimate four models. The problem description model includes the time trend and

seven measures of community amenities (i.e., distance from an airport, hospital, teacher education

program and hometown, shopping, socio-economic health, and 2-bedroom rents). The second

model adds teacher characteristics, the third adds information on salaries and non-wage job

attributes, and the fourth adds the opportunity cost measures. Selected coefficients from these

models are presented in Table 7.

(Insert Table 7 about here)

Distance to hometown has the most robust association with teacher retention. Teachers are

much more likely to remain at their initial school the closer it is to their hometown. The farther

from home they are, the more likely they are to both transfer schools and quit teaching. Holding all

other variables at the rural school mean, approximately 3.5 percent more teachers are retained each

Page 28: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

27 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

year when they are working 10 miles from their hometown versus 40 miles from their hometown

(70th versus 30th percentile).

Economic amenities such as shopping, socio-economic health, and rents also predict teacher

retention. Teachers are less likely to transfer from and quit schools in communities with greater

access to shopping, less likely to transfer away from schools in communities with better socio-

economic health, and more likely to transfer from and quit schools in communities with higher

rents. While these three are not all significant in each model, the directions of these relationships are

consistent.

Interpreting individual odds ratios on the amenity measures is complicated by moderately

high correlations between some of them. Take, for example, the economic amenity set of shopping,

socio-economic health, and rent. Communities with greater access to shopping are also communities

with better socio-economic health (r=0.68) and higher rents (r=0.55). Communities with better

socio-economic health have higher rents (r=0.78). Although high correlations present interpretation

challenges, they are desirable from a theoretical viewpoint. It makes sense that shopping

establishments would be concentrated in communities with a population with greater ability to

spend (i.e., better socio-economic health). And it makes sense that landlords view this greater ability

to spend as an opportunity to raise rents.

To view how the retention rates vary as the values of these amenities change, I estimated

predicted probabilities as the economic amenity set moved together from the poorest values to the

richest values (i.e., least shopping, lowest socio-economic health, and highest rents to most

shopping, highest socio-economic health, and lowest rents). Results of a Wald Test reject the null

hypothesis that the odds ratios for these three amenity measures are simultaneously equal to zero.

As more factors are held constant at the rural school mean, improvements in the economic

amenity set predict substantially higher retention rates, especially once the effects of salary and non-

Page 29: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

28 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

wage job attributes are taken account (see Figure 3). Compare, for example, schools in communities

at the 70th percentile (i.e., shopping 0.4 standard deviations below the state mean, socio-economic

health 0.5 standard deviations above the state mean, and 2-bedroom rents of $569 a month) to

schools in the 30th percentile (i.e., shopping 1.1 standard deviations below the state mean, socio-

economic health 0.6 standard deviations below the state mean, and 2-bedroom rents of $669 a

month). Without controlling for any other factors, schools in the richer amenity community relative

to schools in the poorer amenity community retain about 2 percent more beginning teachers each

year (see Figure 4). This predicted retention advantage grows to 5 percent annually once all other

factors are held constant.

(Insert Figure 3 about here)

(Insert Figure 4 about here)

The relationship between community amenities and teacher retention rates is even more

pronounced when a school’s entire amenity set (i.e., all seven measures) is simulated to move from

the poorest to the richest values (see Figure 8). Predicted retention rates from the problem

description model indicate that schools with the richer amenities retain almost 5 percent more

beginning teachers annually. Accounting for the full set of controls, the retention advantage of the

richer amenity schools grows to just over 8 percent annually.

Failure to account for differences in salaries and opportunity costs across rural teachers not

only downward biases the correlations between community amenities and rural teacher retention but

also fails to reveal an important aspect of rural teacher retention. Rural teachers earning higher

salaries are more likely to be retained than teachers earning lower salaries. Echoing the results from

the across community type model, teacher salaries are shown to have an association with teacher

quit decisions (p<.01) and not transfer decisions once opportunity costs are included in the model.

Opportunity costs, on the other hand, are shown to have the opposite pattern—i.e., an association

Page 30: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

29 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

with transfer decisions but not quit decisions. Rural teachers earning higher salaries relative to the

local average are less likely to transfer than those earning relatively lower salaries (p<.05) as are

teachers working in districts with larger benchmark salaries (p<.01).

I also estimated a version of the problem description model that included interactions

between the amenities and the time trend. Wald test results all failed to reject the null hypothesis that

all were simultaneously equal to zero. This indicates that the relation between amenities and annual

attrition rates may not vary as teachers gain more experience. Additionally, Wald tests on the odds

ratios for the interaction between amenities and subject taught also failed to reject the null

hypothesis that all were simultaneously equal to zero. This indicates that the relationship between

amenities and retention does not vary across subject taught.

Once the sample is restricted to rural teachers, some of the relationships measured in the

across-community type models become more pronounced. Rural schools have particular challenges

retaining critical-shortage subject and multiple-subject teachers relative to teachers of other subjects.

Holding the full set of retention factors constant at the rural school mean, mathematics teachers are

19 percent more likely to transfer from their initial school (p<.05), special education teachers are 56

percent more likely to transfer (p<.01), and multiple-subject are 18 percent more likely to transfer

(p<.05) and 33 percent more likely to quit (p<.01). Their ability to retain science teachers does not

significantly differ from their ability to retain other non-critical-shortage subject and non-multiple

subject teachers.

Several school characteristics are significantly related to rural teacher retention. Teaching in

elementary schools and larger student bodies are all associated with higher retention rates. Student

body characteristics such as percent minority and percent eligible for free or reduced lunch are

negatively associated with teacher retention (i.e., more minority and poor students are associated

with a higher odds ratio of transferring schools and/or quitting teaching) though the coefficients are

Page 31: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

30 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

not significant at conventional levels. Student-teacher ratios are not significantly associated with

rural teacher retention.

4.3 School Quality Robustness Check

The influence of school quality on teacher career path decisions as been highlighted in

previous studies. If this influence differs systematically with community amenities (e.g., school

quality decreases as the distance between the school and the nearest airport increase, etc.), my

estimates of the relationship between community amenities and teacher retention will be biased.

Similarly, if the influence differs across community type, my estimates of retention rates in rural

versus non-rural settings are also biased. Although my models include several important indicators

correlated to school quality such as the percent minority, the percent limited English proficient, and

the percent eligible for free or reduced lunch, I do not include a more direct measure such as student

academic achievement. I do not have access to longitudinal student-level academic achievement

such as scores on standardized state-level examinations, but was able to collect aggregate test scores

for 4th and 8th graders in mathematics and reading via the National Longitudinal School-Level State

Assessment Score Database. These data span four of the nine years analyzed herein (1999-2002).

I re-estimated all the models described above—i.e., the series of four models for both

retention across community types and retention within rural schools—but include the average

school-level test score as a school characteristic. I standardized test scores across all schools within

each year. These models are estimated once with 4th grade test scores and again with 8th grade test

scores. I view these data as indicators of the academic performance of all students in the school

rather than just the grades tested. Consequently, I include all teachers in a school with test score data

in these models. And since there are no data on high school student achievement, I must exclude

high school teachers from these analyses.

Page 32: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

31 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

The inclusion of student achievement does not change the key relationships discussed

above—i.e., retention across community type and across rural schools with varying amounts of

amenities. The estimated coefficients of the relationship between student achievement and teacher

retention are mostly insignificant (26 of the 32 estimates) and only two of six significant results align

with the hypothesis that low performing schools have greater difficulty retaining teachers. For

example, teachers at schools with higher average 8th grade reading achievement have a greater

likelihood of quitting teaching than teachers at poorer performing schools. Furthermore, the

relationship between community type and teacher retention and between community amenities and

rural teacher retention remains when these student achievement measures are included.

5. Policy Implications and Conclusions

This analysis provides evidence that rural schools have less success retaining beginning

teachers than their do suburban counterparts. It also shows rural schools have a harder time

retaining critical-shortage subject and multiple-subject teachers than teachers of other subjects.

These findings point to the need to direct additional resources toward policies aimed at improving

rural teacher retention. However, the results also show retention differences among rural schools with

those in non-metropolitan counties having lower retention rates than rural schools in metropolitan

counties which are closer to urbanized areas and more likely to be part of districts that include

suburban schools.

My results highlight observable and measurable characteristics of rural schools and their

communities that predict lower retention rates. Community amenities, heretofore not included in

models of teacher retention, are shown to have significant predictive power. Schools located in

communities with richer community amenities have higher teacher retention even after accounting

for differences in salary, non-wage job attributes, and opportunity costs.

Page 33: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

32 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Echoing previous findings of Boyd et al. (2005a and 2005b), teachers are shown to have

preferences for proximity to their hometown. Teachers at rural schools farther from their

hometown are more likely to transfer and quit. This suggests that “Grow Your Own” programs may

prove particularly beneficial in rural schools. These programs provide financial assistance to either

area high school graduates or current education paraprofessionals to earn a teaching credential and

return to teach in the community.

Schools in rural communities with richer economic amenities (i.e., shopping, socio-economic

health, and rents) have an easier time retaining teachers. One policy implication could be an

increased focus on rural community development (Beaulieu & Gibbs, 2005). Admittedly, community

development is a longer term policy goal. However, in the more short-term, policymakers can work

to ensure the expansion of broadband internet access to rural communities as a means of lessening

the negative consequences of poorer amenities. Sixty percent of rural households have broadband

Internet service compared with 70 percent of non-rural Americans (NTIA, 2011). Recent years have

seen an explosion in the amount of online shopping and online social networking available to users.

Connecting more rural teachers at faster speeds to these services will not only increase their access

to goods and services but will also diminish the sense of isolation felt by many in rural communities.

The results also highlight the role of salaries in closing retention gaps between rural and

suburban schools. Once salaries are added to the models, the differences are substantially reduced.

The fact salaries are lower in rural schools is frequently justified on the basis that cost-of-living in

rural communities is lower than in non-rural communities. Many states build geographic variation in

the cost-of-living into their funding formula for teacher salary support. The skills needed to teach

are quite portable across community type, and it appears teachers are not fully discounting the

higher non-rural salaries to reflect the higher cost-of-living. Or perhaps they are but are also

discounting the rural salaries for the psychological costs of living in a rural community. These

Page 34: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

33 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

psychological costs relate to the perceived disutility of working in a rural school unrelated to costs of

goods and services. For example, in a society that places emphasis on being successful in the global

marketplace, urban and not rural communities are frequently seen as the place to accomplish that

(Edmondson, 2003). States ought to adjust their funding formula to include additional salary

supports for rural schools to help them retain teachers.

If policymakers are serious about ensuring all students meet academic performance

proficiency standards, they need to be serious about providing quality teachers for all students.

Currently, rural students are serving as proving grounds for new teachers who leave rural schools at

higher rates after gaining valuable classroom experience. Policymakers must demonstrate a greater

willingness to target incentives to specific subsets of teachers and schools than is now the case (Loeb

& Miller, 2006). Policymakers must target additional resources toward the neediest rural schools.

Page 35: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

34 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Table 1. Sample of Beginning Teachers by Cohort and Community Type, 1994 to 2002

Cohort

Non-Metropolitan

Rural

Metropolitan Rural

Suburban Other Urban Big Four Cities

N % N % N % N % N %

1993-94 383 8.3 390 7.2 549 7.5 201 7.2 183 7.1 1994-95 399 8.6 395 7.3 498 6.8 205 7.3 175 6.8 1995-96 377 8.1 497 9.2 632 8.7 190 6.8 153 5.9 1996-97 407 8.8 549 10.2 730 10.0 218 7.8 249 9.6 1997-98 483 10.4 559 10.4 770 10.6 273 9.8 183 7.1 1998-99 609 13.1 678 12.6 927 12.7 366 13.1 301 11.6

1999-2000 641 13.8 748 13.9 995 13.7 453 16.2 436 16.8 2000-01 682 14.7 760 14.1 1,082 14.8 466 16.7 449 17.3 2001-02 658 14.2 809 15.0 1,107 15.2 420 15.0 463 17.9

TOTAL 4,639 100 5,385 100 7,290 100 2,792 100 2,592 100

Table 2 Average School Community Amenities by Community Type, 1994 to 2002

Airport (miles)

Teacher Education Program (miles)

Hospital (miles)

Home-town

(miles)

Shopping (z-score)

Socio-Economic

Health (z-score)

Fair Market Rent

(2004$)

Non-Metropolitan Rural 36.7 21.6 7.9 45.3 -1.0 -1.12 588.19 Metropolitan Rural 20.5 14.3 9.1 38.7 -0.4 0.2 722.39 Suburban 11.5 6.8 4.1 31.2 0.6 0.6 789.38 Other Urban 13.4 8.6 1.4 28.5 0.5 0.1 729.80 Big Four Cities 6.7 2.0 1.2 36.4 1.2 0.3 715.82

Statewide 19.4 11.9 5.6 36.4 0.0 0.0 711.96

Table 3 Percentage Change in Employed Teacher Labor Force by Community Type, 1994 to 2002

Total FTE

of All Teachers

Number of… FTE of Specific Subjects Number of Multiple-Subject

Teachers

New Hires

New Teachers

Math Science Special

Education

Non-Metropolitan Rural 8.5 38.2 71.5 6.9 18.1 52.0 16.5 Metropolitan Rural 20.6 56.8 105.1 23.8 30.1 57.0 39.5 Suburban 20.6 49.5 101.8 19.8 27.2 47.9 33.0 Other Urban 11.3 39.1 111.4 3.2 19.2 34.8 31.1 Big Four Cities 19.6 58.3 158.5 2.7 25.9 38.4 20.8

Statewide 16.7 48.8 102.9 14.1 24.9 47.0 28.8

Page 36: Working Paper: Understanding Rural Teacher Retention and

Miller – Rural Teacher Retention

35 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Table 4 Survival Rates over First Five Years by Subject Taught by Community Type, 1994 to 2002

Years of Experience

Mathematics

Non-Metropolitan

Rural (%)

Metropolitan Rural (%)

Suburban (%)

Other Urban (%)

Big Four Cities (%)

1 72.51 70.30 76.40 69.54 56.86 2 57.13 58.89 66.76 54.08 46.15 3 45.31 48.18 59.78 49.17 37.42 4 41.50 40.95 52.49 38.99 24.95 5 36.88 36.18 43.14 32.11 24.95

Years of Experience

Science

Non-Metropolitan

Rural (%)

Metropolitan Rural (%)

Suburban (%)

Other Urban (%)

Big Four Cities (%)

1 71.94 74.74 73.66 72.81 63.37 2 58.74 63.08 62.04 59.29 44.12 3 48.62 55.02 53.72 54.35 37.26 4 41.81 47.83 48.47 48.42 28.32 5 39.16 40.51 44.32 45.57 20.23

Years of Experience

Special Education

Non-Metropolitan

Rural (%)

Metropolitan Rural (%)

Suburban (%)

Other Urban (%)

Big Four Cities (%)

1 69.13 70.00 70.00 70.92 65.53 2 54.21 57.95 57.19 58.94 51.87 3 45.64 47.86 49.69 48.12 45.02 4 38.08 38.29 41.02 39.75 36.38 5 31.32 31.35 33.90 32.77 27.44

Years of Experience

Multiple-Subject

Non-Metropolitan

Rural (%)

Metropolitan Rural (%)

Suburban (%)

Other Urban (%)

Big Four Cities (%)

1 70.55 71.76 69.24 70.65 62.89 2 51.81 59.04 56.22 57.32 49.65 3 42.93 50.46 48.79 56.33 41.09 4 33.67 42.12 43.22 41.92 31.09 5 27.70 38.40 36.12 28.82 29.26

Note. “Elementary” is one subject a teacher could teach; thus, elementary teachers are not necessarily multiple-subject teachers.

Page 37: Working Paper: Understanding Rural Teacher Retention and

36 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Table 5 Means and Standard Deviations of Beginning Teachers Characteristics, Starting Salaries, Non-Job Attributes and Opportunity Costs by Community Type, 1994 to 2002

Non-Metropolitan Rural

Metropolitan Rural Suburban Other Urban Big Four Cities

N Mean (S.D.) N Mean (S.D.) N Mean (S.D.) N Mean (S.D.) N Mean (S.D.)

Teacher Characteristics

Female (%) 4,639 69.6 5,385 70.3 7,290 72.3 2,792 74.5 2,591 72.1

Age 4,639 29.0 (6.8) 5,380 29.3 (6.9) 7,288 28.6 (6.6) 2,790 29.5 (7.3) 2,590 30.5 (8.1)

Minority (%) 4,571 3.4 5,291 4.5 7,180 6.0 2,739 10.8 2,522 21.9

{Missing} 4,639 1.5 5,385 1.8 7,290 1.5 2,792 1.9 2,592 2.7

Failed Certification Exam (%) a 4,386 4.3 5,147 4.4 7,034 4.5 2,622 7.9 2,352 10.3

{Missing} 4,639 5.5 5,385 4.4 7,290 3.5 2,792 6.1 2,592 9.3

Graduate Degree (%) 4,631 26.4 5,379 31.9 7,274 36.7 2,787 31.8 2,582 29.4

Subject Taught (%)

Math 4,639 8.8 5,385 8.1 7,290 7.5 2,792 6.0 2,592 6.3

Science 4,639 9.4 5,385 9.5 7,290 10.1 2,792 8.5 2,592 6.9

Special Education 4,639 16.1 5,385 14.9 7,290 13.5 2,792 15.3 2,592 17.9

Multiple Subject 4,639 13.2 5,385 10.6 7,290 8.9 2,792 7.3 2,592 8.1

Competitiveness of Undergraduate Institution (%)

Most 4,364 10.8 4,923 14.8 6,734 17.3 2,499 12.1 2,329 16.0

Least 4,364 2.8 4,923 6.3 6,734 6.9 2,499 5.4 2,329 9.2

{Missing} 4,639 5.9 5,385 8.6 7,290 7.6 2,792 10.5 2,592 10.2

Type of Appointment (%)

Tenured 4,618 0.4 5,365 0.5 7,254 0.5 2,779 0.6 2,572 0.5

Probationary 4,618 84.8 5,365 79.3 7,254 77.0 2,779 80.8 2,572 70.4

Substitute 4,618 14.8 5,365 20.0 7,254 22.5 2,779 18.6 2,572 29.1

Degree to Which Permanently Certified in All Subjects Taught (%)

Not Even Provisionally Certified in Any 4,636 8.4 5,380 7.2 7,287 6.4 2,791 11.4 2,102 17.4

Provisionally in Some, Permanently in None 4,636 89.9 5,380 91.0 7,287 91.8 2,791 86.6 2,102 80.6

Permanently in Some, but Not All 4,636 0.2 5,380 0.1 7,287 0.2 2,791 0.0 2,102 0.1

Permanently in All 4,636 1.6 5,380 1.7 7,287 1.6 2,791 2.0 2,102 2.0

(Continued on next page)

Page 38: Working Paper: Understanding Rural Teacher Retention and

37 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Table 5 (cont.) Means and Standard Deviations of Beginning Teachers Characteristics, Starting Salaries, Non-Job Attributes and Opportunity Costs by Community Type, 1994 to 2002

Non-Metropolitan Rural

Metropolitan Rural Suburban Other Urban Big Four Cities

N Mean (S.D.) N Mean (S.D.) N Mean (S.D.) N Mean (S.D.) N Mean (S.D.)

Bachelor’s 3,408 32,461 (2,667)

3,663 34,278 (3,685)

4,608 34,959 (3,655)

1,901 34,311 (3,840)

1,822 35,643 (3,692)

Master’s 1,223 35,044 (3,226)

1,716 38,171 (4,911)

2,666 39,471 (5,563)

886 38,521 (5,416)

760 38,610 (4,608)

Non-Wage Job Attributes (School Characteristics)

Elementary (%) 2,634 40.6 2,879 47.6 3,739 52.0 1,365 57.7 1,085 69.7

Middle (%) 2,634 13.1 2,879 16.6 3,739 22.2 1,365 22.7 1,085 11.3

High (%) 2,634 34.4 2,879 32.9 3,739 24.4 1,365 19.3 1,085 18.0

Other (%) 2,634 12.0 2,879 3.0 3,739 1.4 1,365 0.3 1,085 1.0

Enrollment (100s) 2,634 5.0 (2.0) 2,879 6.3 (3.2) 3,739 6.9 (3.6) 1,365 7.4 (4.6) 1,085 7.1 (3.4)

% Minority 2,634 5.3 (9.2) 2,879 5.8 (7.7) 3,739 13.1 (18.1) 1,365 32.2 (25.4) 1,084 67.9 (17.2)

% Eligible for Free/ Reduced Lunch 2,634 35.6 (14.4) 2,879 22.1 (14.6) 3,739 17.7 (16.5) 1,365 46.7 (22.3) 1,083 73.8 (19.0)

% Limited English Proficient 2,634 0.4 (1.5) 2,879 0.5 (1.3) 3,739 1.6 (4.5) 1,365 3.5 (6.0) 1,084 7.9 (11.9)

Student-Teacher Ratio 2,634 13.8 (2.5) 2,879 14.8 (2.6) 3,739 15.1 (2.7) 1,365 14.7 (2.5) 1,085 14.4 (3.0)

Opportunity Costs

Number of schools within 20 miles 4,639 35.2

(18.2) 5,385

107.0

(62.5) 7,290

181.7

(81.5) 2,792

131.0

(75.4) 2,592

212.2

(39.6)

Relative wage ratio b 4,639 0.661

(0.070) 5,385

0.636 (0.079)

7,290 0.645

(0.086) 2,792

0.648 (0.089)

2,592 0.661

(0.076)

MA+10 years benchmark ($2004) c 4,639 46,416

(6,874) 5,385

52,755 (11,013)

7,290 53,393

(11,031) 2,792

51,889 (11,640)

2,592 50,766 (8,189)

Relative MA+10 benchmark ratio d 4,639 0.974

(0.081) 5,385

0.969 (0.105)

7,290 0.966

(0.104) 2,792

0.966 (0.106)

2,592 0.961

(0.095)

Median non-teaching wage ($2004) e 4,639 40,034 (3,817)

5,385 46,001 (6,762)

7,290 48,576 (8,059)

2,792 47,174 (7,933)

2,592 46,957 (7,721)

Notes. Teacher samples are beginning teachers in their first year of teaching experience. School sample summarized in this table are all schools in each year they hired a beginning teacher. Each school appears once per year in sample. Salary sample sizes are number of beginning teachers. a Indicates that the teacher failed either the NTE General Knowledge exam or the L beral Arts and Sciences Test on their first attempt.

b Measured as a percent of average salary in region (regional average weighted by vacancies in a teacher’s subject).

c Average salary within district

d Measured as a percent of the average MA+10 years salary in region (regional average weighted by vacancies in a teacher’s subject).

e County-level measure of wages earned by college-educated workers aged 22-64 years in non-teaching position.

Page 39: Working Paper: Understanding Rural Teacher Retention and

38 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Table 6 Selected Estimated Coefficients of Beginning Teacher Attrition from Initial School Across Community Type, 1994 to 2002

Problem

Description

+ Teacher Characteristics

+ School Characteristics + Opportunity

Costs

Transfer

vs Stay

Quit

vs Stay

Transfer

vs Stay

Quit

vs Stay

Transfer

vs Stay

Quit

vs Stay

Transfer

vs Stay

Quit

vs Stay

Community Type

Non-Metropolitan Rural

-0 345 -0 364 -0 148 -0 110 -0 314 -0 167 -0 267 -0 156

(0 171)* (0 193)+ (0 172) (0 197) (0 175)+ (0 199) (0 177) (0 200)

Metropolitan Rural -0 059 0 139 -0 026 0 180 -0 058 0 183 -0 057 0 158

(0 063) (0 070)* (0 064) (0 072)* (0 065) (0 072)* (0 066) (0 073)*

Other Urban 0 031 0 285 0 090 0 285 -0 005 0 270 0 012 0 279

(0 087) (0 097)** (0 088) (0 099)** (0 091) (0 109)* (0 091) (0 107)**

Big Four Cities 0 600 0 787 0 559 0 458 0 448 0 404 0 507 0 512

(0 156)** (0 192)** (0 177)** (0 152)** (0 202)* (0 184)* (0 204)* (0 183)**

Time Trend Interacted with Community Type

Non-Metro Rural*TIME

0 537 0 681 0 393 0 517 0 392 0 522 0 389 0 509

(0 192)** (0 207)** (0 193)* (0 212)* (0 193)* (0 213)* (0 193)* (0 212)*

{squared} -0 122 -0 219 -0 092 -0 187 -0 097 -0 192 -0 097 -0 188

(0 055)* (0 057)** (0 055)+ (0 058)** (0 055)+ (0 058)** (0 055)+ (0 058)**

{cubic} 0 008 0 017 0 006 0 015 0 007 0 016 0 007 0 015

(0 004)+ (0 004)** (0 004) (0 004)** (0 004) (0 004)** (0 004) (0 004)**

Metropolitan Rural*TIME

0 054 -0 023 0 046 -0 035 0 038 -0 041 0 034 -0 036

(0 025)* (0 022) (0 025)+ (0 022) (0 025) (0 022)+ (0 025) (0 022)

Other Urban* TIME 0 020 -0 081 0 010 -0 091 0 010 -0 095 0 006 -0 092

(0 032) (0 032)* (0 032) (0 033)** (0 032) (0 034)** (0 032) (0 033)**

Big Four Cities*TIME -0 159 -0 283 -0 186 -0 188 -0 206 -0 206 -0 206 -0 214

(0 101) (0 125)* (0 120) (0 120) (0 120)+ (0 120)+ (0 119)+ (0 118)+

{squared} 0 031 0 027 0 035 0 020 0 037 0 022 0 036 0 023

(0 012)** (0 014)+ (0 013)** (0 015) (0 013)** (0 015) (0 013)** (0 015)

Log Pseudo-Likelihood -37,463 7 -34,997 0 -34,738 1 -34,708 0

Teacher-Year Observations

63,687 62,414 62,398 62,398

Time Trend YES YES YES YES

Teacher Characteristics YES YES YES

Salary YES YES

Non-Wage Job Attributes

YES YES

Opportunity Costs YES

Year Effects YES YES

** p<0.01; * p<0.05; + p<0.10 Note. Robust standard errors appear in parentheses.

Page 40: Working Paper: Understanding Rural Teacher Retention and

39 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

TABLE 7 Selected Estimated Coefficients of Rural Beginning Teacher Attrition from Initial School, 1994 to 2002

Problem Description + Teacher

Characteristics + School Characteristics + Opportunity Costs

Transfer vs Stay

Quit vs

Stay

Transfer

vs Stay

Quit vs

Stay Transfer vs Stay

Quit vs

Stay Transfer vs Stay

Quit vs

Stay

Community Amenities

ln(airport) -0 004 -0 088 0 003 -0 098 0 009 -0 089 0 011 -0 077

(0 045) (0 047)+ (0 046) (0 049)* (0 048) (0 051)+ (0 050) (0 052)

ln(hospital) -0 017 -0 020 -0 018 -0 017 0 002 -0 002 -0 007 -0 007

(0 022) (0 025) (0 023) (0 025) (0 023) (0 026) (0 024) (0 027)

ln(teacher education Program)

-0 012 0 038 0 004 0 045 -0 002 0 045 0 000 0 052

(0 031) (0 031) (0 032) (0 033) (0 031) (0 033) (0 031) (0 033)

ln(hometown) 0 166 0 171 0 189 0 175 0 190 0 169 0 192 0 169

(0 015)** (0 018)** (0 015)** (0 019)** (0 015)** (0 019)** (0 015)** (0 019)**

Shopping -0 175 -0 112 -0 173 -0 121 -0 034 -0 047 -0 082 -0 084

(0 057)** (0 061)+ (0 059)** (0 064)+ (0 067) (0 073) (0 080) (0 087)

Socio-economic health -0 039 0 053 -0 054 0 014 -0 089 0 017 -0 138 -0 025

(0 043) (0 046) (0 043) (0 048) (0 045)* (0 050) (0 055)* (0 058)

Fair market rents 0 098 0 205 0 234 0 321 0 978 0 579 1 572 0 794

(0 199) (0 210) (0 205) (0 222) (0 240)** (0 263)* (0 392)** (0 418)+

Log Pseudo-Likelihood -16,704 2 -15932 2 -15,792 8 -15,783 2

Teacher-Year Observations

28,444 28,367 28,367 28,367

Time Trend YES YES YES YES

Teacher Characteristics YES YES YES

Salary YES YES

Non-Wage Job Attributes

YES YES

Opportunity Costs YES

Year Effects YES YES

** p<0.01; * p<0.05; + p<0.10 Note. Robust standard errors appear in parentheses.

Page 41: Working Paper: Understanding Rural Teacher Retention and
Page 42: Working Paper: Understanding Rural Teacher Retention and
Page 43: Working Paper: Understanding Rural Teacher Retention and

42 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Map 1. Community Type of Public Schools within New York State

Page 44: Working Paper: Understanding Rural Teacher Retention and

43 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Map 2. Hometowns of Teachers who Grew up in or near New York State, 1994-2002

Map 3. Rural Schools by Distance to Hometown Deciles, 1994 to 2002

Page 45: Working Paper: Understanding Rural Teacher Retention and

44 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

REFERENCES

Baugh, W.H., & Stone, J.A. (1982). Mobility and wage equilibrium in the educator labor market.

Economics of Education Review, 2(3), 253-274. Beaulieu, L.J., & Gibbs, R. (2005). The Role of Education: Promoting the Economic and Social Vitality of

Rural America. Mississippi State, MS: Southern Rural Development Center. Bornfield, G., Hall, N., Hall, P., & Hoover, J.H. (1997). Leaving rural special education positions:

It’s a matter of roots. Rural Special Education Quarterly, 16(1), 30-37. Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2005a). The draw of home: How teachers’

preference for proximity disadvantages urban schools. Journal of Policy Analysis and Management, 24(1), 113-132.

Boyd, D., Lankford, H., Loeb, S., & Wyckoff, J. (2005b). Explaining the short careers of high

achieving teachers in schools with low performing students. American Economic Review, 95(2), 166-171.

Boylan, C., & McSwan, D. (1998). Long-staying rural teachers: Who are they? Australian Journal of

Education, 42(1), 49-65. Boylan, C., Sinclair, R., Smith, A., Squires, D., Edwards, J., Jacob, A., O’Malley, D., & Nolan, B.

(1993). “Retaining teachers in rural schools: Satisfaction, commitment, and lifestyles.” In Rural Education Issues: An Australian Perspective, Eds. C. Boylan & M. Alston, pp. 111-130. Amsterdam: North-Holland.

Dolton, P., & van der Klaauw, W. (1999). The turnover of teachers: A competing risk explanation.

The Review of Economics and Statistics, 81(3), 543-552. Edmondson, J. (2003). Prairie Town: Redefining Rural Life in the Age of Globalization. New York:

Rowman & Littlefield. Feng, L. (2005, November 9). Hire Today, Gone Tomorrow: The Determinants of Attrition among Public

School Teachers. Retrieved from http://garnet.acns.fsu.edu/~lff6254/job_market_paper_10.pdf.

Florida Department of Education (2008, April). Teacher Exit Interview Data, 2006-07. Statistical

Brief, Series 2008-10B: 1-16. Grissmer, D.W., & Kirby, S.N. (1992). Patterns of Attrition among Indiana Teachers, 1965-1987. Santa

Monica: RAND. Gritz, R.M., & Theobald, N.D. (1996). The effects of school district spending priorities on length of

stay in teaching. The Journal of Human Resources, 31(3), 477-512.

Page 46: Working Paper: Understanding Rural Teacher Retention and

45 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Hammer, P.C., Hughes, G., McClure, C., Reeves, C., & Salgado, D. (2005, December). Rural Teacher Recruitment and Retention Practices: A Review of the Research Literature, National Survey of Rural Superintendents, and Case Studies of Programs in Virginia. Nashville, TN: Edvantia.

Hanushek, E.A., Hain, J.F., & Rivkin, S.G. (2004). Why public schools lose teachers. The Journal of

Human Resources, 39(2), 326-354. Hwang, H., Mortensen, D.T., & Reed, W.R. (1998). Hedonic wages and labor market search. Journal

of Labor Economics, 16(4), 815-847. Imazeki, J. (2005). Teacher salaries and teacher attrition. Economics of Education Review, 24(4), 431-449. Ingersoll, R.M. (2002). The teacher shortage: A case of wrong diagnosis and wrong prescription.

NASSP Bulletin, 86(631), 16-31. Johnson, J., & Strange, M. (2007, October). Why Rural Matters 2007: The Realities of Rural Education

Growth. Washington, DC: Rural School and Community Trust. Kleinfield, J., & McDiarmid, G.W. (1986). The satisfaction of Alaska’s isolated rural teachers with

their work life. Research in Rural Education, 3(3), 117-120. Loeb, S., & Miller, L.C. (2006, December). A Review of State Teacher Policies: What are they, What are their

effects, and What are their implications for school finance? Stanford, CA: Institute for Research on Education Policy & Practice. Retrieved from http://irepp.stanford.edu.

Massey, S., & Crosby, J. (1983). Special problems, special opportunities: Preparing teachers for rural

schools. Phi Delta Kappan, 65(4), 265-269. McClure, C.T., Redfield, D., & Hammer, P.C. (2003, December). Recruiting and Retaining High Quality

Teachers in Rural Areas. Charleston, WV: AEL. Mont, D., & Rees, D.I. (1996). The influence of classroom characteristics on high school teacher

turnover. Economic Inquiry, 34, 152-167. Mortensen, D.T. (1990). “Equilibrium wage distributions: A synthesis.” In Panel Data and Labor

Market Studies, Ed. J. Hartog, G. Ridder, and J. Theeuwes, pp. 279-296. Amsterdam: North-Holland.

Murnane, R.J. (1981). Teacher mobility revised. The Journal of Human Resources, 16(1), 3-19. Murnane, R.J. (1984). Selection and survival in the teacher labor market. The Review of Economics and

Statistics, 66(3), 513-518. Murnane, R.J., & Olsen, R.J. (1989). The effects of salaries and opportunity costs on duration in

teaching: Evidence from Michigan. The Review of Economics and Statistics, 71(2), 347-352.

Page 47: Working Paper: Understanding Rural Teacher Retention and

46 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Murnane, R.J., & Olsen, R.J. (1990). The effects of salaries and opportunity costs on length of stay in teaching: Evidence from North Carolina. The Journal of Human Resources, 25(1), 106-124.

Murnane, R.J., Singer, J.D., & Willett, J.D. (1988). The career paths of teachers: Implications for

teacher supply and methodological lessons for research. Educational Researcher, 17(6), 22-30. Murphy, P.J., & Angleski, K. (1996/1997). Rural teacher mobility: A report from British Columbia.

Rural Educator, 18(2), 5-11. National Association of State Boards of Education (2004). Teacher recruitment and retention: A

survey of the rural landscape. Policy Update, 12(11), 1-4. National Telecommunications and Information Administration (2011). Broadband Statistics Report:

Broadband Availability in Urban vs. Rural Areas. Washington, DC: U.S. Department of Commerce. Retrieved 20 February 2011 from http://www.broadbandmap.gov/download/reports/national-broadband-map-broadband-availability-in-rural-vs-urban-areas.pdf.

Ondrich, J., Pas, E., & Yinger, J. (2008). The determinants of teacher attrition in Upstate New York.

Public Finance Review, 36(1), 112-144. Podgurksy, M., Monroe, R., & Watson, D. (2004). The academic quality of public school teachers:

an analysis of entry and exit behavior. Economics of Education Review, 23(5), 507-518. Provasnik, S., KewalRamani, A., Coleman, M.M., Gilbertson, L., Herring, W., & Xie, Q. (2007).

Status of Rural Education in America (NCES 2007-040). Washington, DC: National Center for Education Statistics, Institute for Education Sciences, U.S. Department of Education.

Reininger, M. (2006). Teachers’ Location Preferences and the Implications for Schools with Different Student

Populations. Chicago, IL: Northwestern University. Rosen, S. (1974). Hedonic prices and implicit markets: Product differentiation in pure competition,

Journal of Political Economy, 82(Jan/Feb), 34-55. Rural Policy Research Institute (2006). Defining Rural: Definitions of rural areas in the U.S. Columbia,

MO: Author. Scafidi, B., Sjoquist, D.L. & Stinebrickner, T.R. (2007). Race, poverty, and teacher mobility.

Economics of Education Review, 26(2), 145-159. Schwartzbeck, T.D., Prince, C. D., Redfield, D., Morris, H., & Hammer, P.C. (2003). How are Rural

Districts Meeting the Teacher Quality Requirements of No Child Left Behind? Charleston, WV: Appalachia Educational Laboratory.

Stinebrickner, T.R. (1998). An empirical investigation of teacher attrition. Economics of Education

Review, 17(2), 127-136.

Page 48: Working Paper: Understanding Rural Teacher Retention and

47 CEPWC Working Paper Series No. 1. October 2012.

Available at http://curry.virginia.edu/research/centers/cepwc/publications. Curry School of Education | Frank Batten School of Leadership and Public Policy | University of Virginia

Stinebrickner, T.R. (1999). Estimation of a duration model in the presence of missing data. The Review of Economics and Statistics, 81(3), 529-542.

Stinebrickner, T.R. (2001). A dynamic model of teacher labor supply. Journal of Labor Economics, 19(1),

196-230. Stinebrickner, T.R. (2002). An analysis of occupational change and departure from the labor force:

Evidence of the reasons that teachers leave. The Journal of Human Resources, 37(1), 192-216. Storey, V.J. (1993). Factors, issues, and problems in the recruitment and retention of teachers for

rural schools. Journal of Research in Rural Education, 9(3), 160-169. Theobald, N.D. (1990). An examination of the influence of personal, professional, and school

district characteristics on public school teacher retention. Economics of Education Review, 9(3), 241-250.