Impact of Child’s Special Health Care Needs on Employment ... · SAS Enterprise Miner Client 12.3 version was used for model development. Among all 41 variables, 2 are interval,

#analyticsx

Ram Poudel1, Vivek Manikandan Damodaran1

1. MS in Business Analytics, Spears School of Business, Oklahoma State University

ABSTRACT

In the United States more than 10 million children have special health care needs. Maternal and Child Health Bureau defines Children with Special Health Care Needs (CHSCN) as those who have one or more chronic physical, developmental, behavioral, or emotional conditions and therefore require extra health and related services beyond that required by normal children. Children with special health care needs are likely to require a high amount of parental and family care that affects parents’ employment significantly. Family members often struggle to pay for health care needs of such child and wrestle with balancing employment and caregiving. This study aims at analyzing indicator variables that cause those family members to shrink their working hours or stop from working. A model was developed based on the response provided by family members to The National Survey of Children with Special Health Care Needs (NS-CSHCN). The final sample consists of 40,242 CSHCN— approximately 750 in each state and the District of Columbia. About one quarter of total respondents said they had to cut back work hours or stop working because of the responsibilities of caring special health care needs child. Spatial distribution of children with special health care needs was illustrated using SAS Visual Analytics. Various machine learning and regression models using SAS Enterprise Miner were assessed to find out the key indicators impacting employment of the family members. Logistic regression was found the best performing model to predict whether family members’ employment was affected or not with a total accuracy of 83.36%. The sensitivity and specificity of selected model were 43.56% and 94.08% respectively. The important covariates that explained model’s predictive power are age of a child, language spoken at home, gender of a child, poverty level, family’s financial burden, and activity limitations of a child because of disabilities.

METHODS

Data Collection

This study is a analysis of the data drawn from 2009-2010 National Survey of Children with Special Health Care Needs (NS- CSHCN). 2009-2010 National Survey of Children with Special Health Care Needs (NS- CSHCN) is a nationwide telephone survey sponsored by the United States Department of Health and Human Services, Health Resources Services Administration, Maternal and Child Health Bureau(MCHB), and conducted by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention, using State and Local Area Integrated Telephone Survey (SLAITS) technology for sampling and administration. The Data Resource Center for Child and Adolescent Health (DRC) creates and maintains indicator and outcome variables of Maternal and Child Health Bureau. The population, disability prevalence, and employment data are from US Census, United States Department of Labor, and Cornell University respectively.

METHODS CONTINUED

Study Population

Using CSHCN Screener, the 2009-2010 NS-CSHCN interview sample was obtained by screening 372,698 children (0-17 years old) living in 196,159 households nationwide. Out of 372,698 children screened, only 59,941 were identified with special health care needs. Approximately 750 in each state and the District of Columbia, ranging from 751 in District of Columbia to 878 in Texas, a total of 40,242 CSHCN were selected for in-depth interviews. This study examines the subset of CSHCN whose family members had responded to the question whether they had cut back and/or stopped working because of child’s health needs. After removing those children whose responses were unknown or missing, the total number of children reduced to 33,939 for further analysis.

Variable Selection and Model Development

SAS Enterprise Miner Client 12.3 version was used for model development. Among all 41 variables, 2 are interval, 21 are binary and 18 are nominal variables. Target is a binary indicator variable which ask yes or no for “ Impact on Family Work Life: CSHCN whose family members had cutback and/or stopped working because of child’s health needs.

Fig 1. Diagram of SAS Enterprise Miner Project

Impact of Child’s Special Health Care Needs on Employment of Parents and Family Members: A Datamining Approach Using SAS EM and SAS VA

#analyticsx


RESULTS

Fig 3. State-level correlation between Disability Prevalence and Employment Rate of Individuals with Disabilities in the USA in 2012

RESULTS CONTINUED


Ram Poudel1, Vivek Damodaran1

Fig 2. State-Level Disability Prevalence (%) and Employment Rate (%) of Individuals with Disabilities of 18-64 Years in the United States of America, 2012

Table 1. Disability Prevalence (%) and Employment Rate (%) of Individuals with Disabilities, SSI Recipients, and Per Capita Income in the United States of America, 2012

Employment Rate(%)D

isab

ility

Pre

vale

nce

(%)

Observations 51Parameters 2Error DF 49MSE 25.96R- Square 0.32Adj. R-Square 0.30

Most of the Southern States have higher prevalence of disabilities and low employment rate of individuals with disabilities. States in Midwest have opposite pattern: low prevalence and higher employment rate.

#analyticsx

RESULTS CONTINUED




Fig.3 Percentage of Race of CSHCN Whose Family Members’ Work Has Been Impacted Because of Their Disabilities

Comparing to their general population employment of Hispanic and other races seems to be more affected.

Fig.4 Percentage of Federal Poverty Level of CSHCN Whose Family Members’ Work Has Been Impacted Because of Their Disabilities

People who are in top of the poverty level are the most affected. Possible reason could be people who have higher income have jobs and now due to disabilities their jobs have been impacted.

Fig.5 Percentage of Education Level of CSHCN Family Members Whose Work Has Been Impacted Because of Their Disabilities

People with education level greater than high school are highly affected.

Fig.6 Percentage of Insurance Type of CSHCN Whose Family Members’ Work Has Been Impacted Because of Their Disabilities

More CSHCN, whose family members’ work has been impacted, have private insurance.

#analyticsx

RESULTS CONTINUED

Fig 3. Fit Statistics of All Nine Models from Model Comparison Node Fig 4. ROC Curves of All Nine Models from Model Comparison Node




From all nine models developed, the model of Logistic Regression with backward selection has been selected as the best model with the lowest validation misclassification rate of 16.64%.

The selected Logistic Regression with backward selection has

Validation ROC Index = 0.85

Accuracy = 83.36%Sensitivity = 43.56%

Specificity = 94.08%

#analyticsx

REFERENCES

ACKNOWLEDGEMENTS: ACKNOWLEDGEMENTS: We thank Dr. Goutam Chakraborty, Professor of Department of Marketing, Founder of MS Business Analytics & SAS and OSU Data Mining Certificate Program for his support throughout the project.

CONCLUSIONS

1. Age of a child*: As child grows older there is less effects on a target: work life of family members. (Odds Ratio: 0.977)

2. Language spoken at home*: If English is a language spoken at home, impacts on target is less (odds ratios: English vs. Hispanic- 3.07, English vs. Non-Hispanic- 1.29)

3. Gender of Child*: Male Child is positively associated with impacts on work life of family members. (Odds Ratio: 1.16)

4. Poverty level of a family: As the income level goes up the odds of impact on work life of family members goes down.

(Odds ratios: 0-99% FPL vs. 100-199% FPL; 0. 94, 0-99% FPL vs. 200-299% FPL; 0. 89, 0-99% FPL vs. 300-399% FPL; 0.83)

5. Family’s financial burden: Families with less financial problems have less impacts on work life(Odds Ratio: 0.58 ).

6. Activity limitations of a child because of disabilities: It has positive impacts on work life of family(Odds Ratio: 1.23)

7. Family structure: Families with biological two parents have less impact on their work life than others (Odds Ratios: biological two parents vs. stepfamily,;0.1.18, biological two parents vs. mother only; 1.23, biological two parents vs. other family structure; 1.29)

8. Type of Insurance: If child is currently insured, the impacts on family members’ work life is inverse (Odds ratio: 0.95).




RESULTS CONTINUEDImportant Covariates that Explained Model’s Predictive Power * = Significant at α level of 0.05

Most of the Southern States of the country where per capita income is less have more population with disabilities but low employment rate.

A large proportion of children with disabilities and other special needs are in low-income families whose family members had impacts on their jobs.

Researches show low-income children are more likely to have learning disabilities than higher-income children.

A work-life balance for employees, especially those with special needs children, is essential to the health and productivity of any organization.

Children in poverty are more prone than other children to have a chronic condition that limits activities.

Children of minorities’ members had to limit or cut back their jobs that push them more to poverty which makes them more vulnerable.

CONCLUSIONS CONTINUED

Our model has pretty decent accuracy to predict whose members’ employment would be affected. However, we will work further to improve model’s sensitivity.

In future, we will do more analysis to assess the detail effects of important covariates like age, gender, poverty level, and type of insurance.

1. Child and Adolescent Health Measurement Initiative (CAHMI), “2009-2010 NS-CSHCN Indicator

and Outcome Variables SAS Codebook, Version 1,” 2012, Data Resource Center for Child and

Adolescent Health, www.childhealthdata.org.

2. S. Lee, M. Sills, and G. Oh. (June 20, 2002). Disabilities among Children and Mothers in Low-Income Families (Research in Brief, IWPR Publication #D449). Washington, DC: Institute for Women’s Policy Research.

3. S. J. Heymann and A. Earle. (Winter 2000). “Low-Income Parents: How Do Working Conditions Affect Their Opportunity to Help School-Age Children At Risk?” American Educational Research Journal, Vol. 37, No. 4. pp. 833-848.

4. Casserly, Meghan. "The Top Five Reasons Employees Will Quit In 2013." Forbes. 2 1 2013:

Web. 16 Apr. 2013.

5. U.S. Census Bureau; Census 2000, Summary File 1, Matrix P7; generated by Joe Smith; using American FactFinder; <http://factfinder2.census.gov>; (15 February 2012).

6. "Effect on Family Life." . care.com. Web. 8 Apr 2013. <http://www.care.com/special-needseffect-on-family-life-p1167-q227936.html>.

Documents

Impact of Child’s Special Health Care Needs on Employment ... · SAS Enterprise Miner Client 12.3 version was used for model development. Among all 41 variables, 2 are interval,