26
The Educational Influence of Dormitory Roommates Author(s): Robert L. Hall and Ben Willerman Source: Sociometry, Vol. 26, No. 3 (Sep., 1963), pp. 294-318 Published by: American Sociological Association Stable URL: http://www.jstor.org/stable/2786072 . Accessed: 25/06/2014 01:55 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to Sociometry. http://www.jstor.org This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AM All use subject to JSTOR Terms and Conditions

The Educational Influence of Dormitory Roommates

Embed Size (px)

Citation preview

Page 1: The Educational Influence of Dormitory Roommates

The Educational Influence of Dormitory RoommatesAuthor(s): Robert L. Hall and Ben WillermanSource: Sociometry, Vol. 26, No. 3 (Sep., 1963), pp. 294-318Published by: American Sociological AssociationStable URL: http://www.jstor.org/stable/2786072 .

Accessed: 25/06/2014 01:55

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access toSociometry.

http://www.jstor.org

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 2: The Educational Influence of Dormitory Roommates

The Educational Influence of Dormitory Roommates *

ROBERT L. HALL, National Science Foundation

BEN WILLERMAN, University of Minnesota

To study the effect of college roommates upon one another's grades, study habits, and other activities, dormitory roommates with various combinations of academic ability were experimentally formed. An analogous correlational study was conducted on roommates who had selected one another. Observed effects are mediated in various ways by the birth ordinal positions of the room- mates in their respective families. It appears that first-born students are more susceptible to influence and later-borns more influential. In the experiment, students with high-ability roommates obtain better grades than those with low-ability roommates only if the roommate is later-born. Among males, first- borns profit more than later-borns from sharing courses with their room- mate. Effects of the roommate on other educationally relevant activities are also examined.

This is a study of social influence in higher education. It is construed as having two facets: it is an inquiry into the influence exerted in two-person groups and some of the factors that mediate such influence; it is also an inquiry concerning higher education and some of the social psychological factors that may shape the impact of college upon the intellectual development of students.

This research focuses upon the effects that roommates (RMs) in college dormitories have upon each other's education. Academic performance, as re- flected in grades, is a special concern of the research. Many other effects are also of interest. The study considers possible effects of RMs upon study habits, informal learning in "bull sessions," interest and participation in cul- tural events (e.g., concerts, art exhibitions, special lectures), and extracur- ricular activity.

Preliminary interviews with students suggested that one important influ-

* This research was made possible by a grant from the Louis W. and Maud Hill Family Foundation, for which the authors express their profound gratitude. This study is part of a program of research on the social psychology of student life being conducted by the Office of the Dean of Students, University of Minnesota. The cooperation and ad- vice of the directors of the dormitories, James Schroeder and Myrtle Gagnon, and the head counselors, James Little and John Pringle, was essential to the project. Charles Buchwald and Sally Katz provided skillful research assistance during major portions of the research. The authors are indebted to Dr. Edward 0. Swanson for making available data on test scores, high schools, and high school records.

294

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 3: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 295

ence process in the setting is a tendency toward behavioral similarity. Most dormitory applicants in this study were freshmen from small towns and rural areas, facing their first experience with college, with living away from home, and with a large city. Under such conditions of uncertainty, forced into close interaction with a RM, a student finds the RM's example, encouragement, or reinforcement especially important. The interviews revealed that some RMs set an example of good study habits and reinforce serious study, while others encourage activities that compete with study, and some provide much dis- traction from study.

Besides the tendency toward behavioral similarity, other processes appeared from the interviews to be involved in determining the RM's effect upon academic performance. Some students directly help their RMs with studies, for example, by suggesting ways of solving problems or by reading and discussing the RM's work. Also, in "bull sessions," some students provide information or stimulate novel thought about matters covered in courses, thus possibly enhancing understanding.

It seemed likely that high-ability students differ systematically from low- ability ones in precisely those forms of behavior most likely to influence a RM's academic performance. That is, high students, on the average, probably more often help their RMs with studies, provide the stimulation of novel ideas, set a good example in study habits, etc. Hence, on the basis of a tendency toward behavioral similarity and also differences in help and stimu- lation, it is logical to expect students with high-ability RMs to perform better academically than those with low-ability RMs, if other important influences are controlled.

The planning of this study was guided in large part by an effort to test Whether students' academic performance is differentially affected by RMs of different levels of ability and an effort to clarify the processes by which this might come about. This was tested by an experiment in which RMs were assigned in a systematic way, and also by a survey of a separate sample of students who had chosen each other as RMs. Portions of the study were aimed at detecting tendencies toward behavioral similarity of RMs in their educationally relevant activities, regardless of ability levels, and clarifying the circumstances under which these tendencies occur, both for experimental RMs and for mutually chosen ones.

RESEARCH PROCEDURES

The Experimental Assembly of Roommates. Two newly opened dormitories were included in this study. The students in Dormitory A were male, mostly freshmen studying liberal arts, science, or engineering. Dormitory A con-

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 4: The Educational Influence of Dormitory Roommates

296 SOCIOMETRY

tained several "houses," each consisting of approximately 25 double rooms, one counselor's room, and one small common lounge room. The residents of Dormitory B were also largely freshmen, with male students in one wing and female students in another wing, and with an administrative office and a large lounge shared by the two wings. The males were studying agricultural subjects, and the females were studying home economics, with few excep- tions. Dormitory B contained several "houses," each consisting of 14 to 24 double rooms, a small lounge, and one counselor's room.

In order to determine the effect upon students of their RMs' academic ability, RMs were assigned so as to create various combinations of ability. The relatively small number of cases available and the need for making rapid assignments to rooms made it impossible to use more than one index of academic ability in making assignments. Past experience and research had shown that a student's percentile rank in his high school graduating class was the best single predictor of his academic performance in college. High school percentile rank (HSR) was, therefore, the index used for the experi- mental pairing of RMs.

Prior to the opening of the dormitories in the fall, certain information concerning each applicant to the two dormitories was obtained. This infor- mation included his college, class, age, major field of study, name of RM requested (if any), HSR, and when available, score on the American Council on Education Psychological Examination (ACE)-a test of scholastic apti- tude.

The assignment of RMs was accomplished in several steps. First, all mutual requests for particular RMs were honored. These were to be analyzed sepa- rately from the experimentally assigned cases, using correlational methods devised to be analogous to the experiment. Second, all cases were divided into six categories:

(1) freshmen in Dorm A (2) upperclassmen in Dorm A (3) male freshmen in Dorm B (4) male upperclassmen in Dorm B (5) female freshmen in Dorm B (6) female upperclassmen in Dorm B

Most female upperclassmen requested particular RMs, and there were too few left in category 6 for experimental assignment. Hence, there were five categories of experimental RMs and six categories of mutually chosen RMs. Some students were then excluded from the experimental assembly because essential information was lacking. Also, all freshmen who were over 21 or veterans of military service, all foreign students, and all graduate students were arbitrarily excluded. Each remaining student was assigned to a RM

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 5: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 297

and to a matched case, both in the same dormitory category as the subject. The exact procedure was as follows.

(1) All the students in a category were put into rank order on HSR, and ties were resolved randomly.

(2) If the total number of cases within the category was not an exact multiple of 8, cases were eliminated to reduce the number to an exact multi- ple. First cases eliminated were those few who made a restrictive request for a particular type of RM, such as "a non-smoker" or a "Baptist." Then further eliminations, if necessary, were random. Exact multiples of 8 were required for the later steps.

(3) The two highest students in the rank-order of HSR were designated a matched pair, the next two, another matched pair, and so on.

(4) Those students above the median HSR of their category were desig- nated "high-achieving" or simply "high" students, and those below the median as "low-achieving" or "low" students.

(5) From each matched pair, one student was randomly selected to have a "high" RM and the other a "low" RM. Hence, all students within a dormitory category were divided into four conditions with equal numbers of cases: H(H)-high student with high RM; H(L)-high student with low RM; L(H)-low student with high RM; L(L)-low student with low RM.

(6) The H(H) students in the highest and the second highest matched pairs were assigned as RMs of each other, and so on down the HSR scale through all H(H) and L(L) students.

(7) The H(L) student in the highest matched pair was made RM of the L(H) student in the highest matched pair below the median; the second highest H (L) was assigned to the second L (H) below the median; and so on.

The net effect of this procedure was as follows. Each student was matched in HSR with another student. One member of each matched pair had a RM of equivalent HSR. The second member of each high matched pair had a low RM, about 50 per cent lower in HSR within his category. The second mem- ber of each low matched pair had a high RM, about 50 per cent higher in HSR within his category. There were 192 students so assigned, with 48 in each of the four experimental conditions. The 48 in each condition were divided among the five dormitory categories as follows: freshmen in Dorm A, 16; upperclassmen in Dorm A, 6; male freshmen in Dorm B, 8; male up- perclassmen in Dorm B, 8; female freshmen in Dorm B, 10. Hence, the 96 RM pairs were distributed among three conditions: 24 HH pairs, 48 HL pairs, and 24 LL pairs.

Data Collection. Several classes of information were desired concerning each student: his study habits, the type of personal relationship between him

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 6: The Educational Influence of Dormitory Roommates

298 SOCIOMETRY

and his RM, the informal discussions, concerts, art exhibitions, or other cultural and extracurricular activities he participated in, and his grades in courses. This information came from several sources: a questionnaire com- pleted by each student, interviews with a sample of students, interviews and report forms from residence counselors, and administrative records.

A questionnaire was devised to obtain most of the information about relations between RMs, study habits, and time spent in various kinds of activi- ties. It was pretested several times on small groups of students from dormi- tories not in this study, and it was revised after each pretest. Early in Decem- ber, after the residents had lived in the dormitories for over two months, and one week before final exams, the questionnaire was administered to all residents of both dormitories. Of 293 residents in Dormitory A, all but two completed the questionnaire, and every one of 233 residents in Dormitory B completed it.

Early in winter quarter, 22 students who had been included in the experi- ment were interviewed as an aid in interpreting data from the questionnaires. Also early in winter quarter, each residence counselor in both dormitories was interviewed. In this interview the counselor rated each RM pair on the closeness of friendship between them, and he rated each individual on time spent studying and time spent in serious intellectual discussions.

Other information collected concerning each student included changes of RM, with reasons reported to the counselor, departure from the University, any complaints or incidents indicating friction between RMs, major field of study, and academic grade transcripts for fall and winter quarters.

Statistical Procedure. In assessing the effects of the experimental assign- ment, it was generally possible to compare matched pairs of observations. In many statistical analyses described below, the analysis of variance model was the same. In these instances the two tests of significance desired were: (1) the experimental effect, i.e., differences between matched pairs of observations in two experimental conditions; and (2) differences among the five dormitory categories in the magnitude of the experimental effect. For these two tests the following variance components were extracted: (1) between experimental treatments; (2) between matched pairs; (3) interaction, 1 X 2. Then the in- teraction term was further partitioned into two sources: (3a) interaction, dormitory category X experimental treatment; (3b) residual. Differences be- tween dormitory categories in magnitude of experimental effect were tested by F ratio: the mean square for source 3a divided by the mean square for source 3b. Over a number of such tests, there were generally not significant differences among dormitory categories. Therefore, the mean square for source 3 was taken as the appropriate error term for testing the experimen-

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 7: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 299

tal effect (source 1). This general model was used in many instances reported below.

The distributions of HSR and ACE among students applying to the dor- mitories were severely skewed. For all subsequent statistical analyses, both HSR and ACE were transformed to T scores, based on a norm population of over 2,600 incoming freshmen in those parts of the university that were represented in this study. Hence, the scores for high school achievement and aptitude used in later analyses were approximately normal in distribution.

In the results reported below, the number of cases varies from one analysis to another for several reasons. Some dormitory applicants included in the design failed to show up. Some insisted on a newly found friend as RM. During the year some left the university for scholastic or personal reasons. Some moved out of the dormitory to other housing. Some changed RMs within the dormitory by request, as will be discussed later. Finally, a particular datum needed for one analysis was occasionally not available for a few students.

RESULTS

Overview of Results. The full results of the research involve a number of intricate statistical analyses on several different samples of students, and various conclusions are drawn with various degrees of confidence. In order to simplify the exposition, the main conclusions are summarized here very briefly to provide a framework for fuller presentation of the evidence. The more detailed presentation will provide the reader with an opportunity to challenge these (admittedly oversimplified) summary statements.

First, among the experimentally assigned students, the "high-achieving" ones (those high in HSR) did, as expected, study more diligently and obtain better grades than the "low-achieving," and these differences were apparent to their RMs. Nevertheless, they did not have an over-all differential effect upon the academic performance of their RMs, when other important influences were controlled. The high students were generally regarded as more desir- able RMs, and pairs including high students were more likely to remain together as RMs.

The initial hypotheses about the effects of RMs were much too simple, although not entirely wrong. There were especially two matters not controlled in the initial design of the study that appeared to be of crucial importance to the relationship between RMs. One was the extent to which RMs took courses in common. The other was the birth ordinal position of the RMs in their respective families, which appeared to be correlated strongly with

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 8: The Educational Influence of Dormitory Roommates

300 SOCIOMETRY

the exercise of and susceptibility to social influence in this setting. The results described here for experimental cases refer only to freshmen, since there were few experimental upperclassmen, and results on them were inconclusive.

Those freshmen who took a number of courses in common with their RMs performed better academically than those who shared few courses with their RMs. Birth order mediated exertion of social influence in such a way that (1) first-born students were more susceptible to influence by a RM, and (2) later-born students had more effect upon their RMs. To illustrate this mediation by birth order, first-borns benefited more from having over- lapping courses with their RMs than did later-borns. Also, the originally predicted differential effect of high and low RMs upon academic perform- ance appears to be present only when the RM is later-born.

Turning attention to those who chose each other as RMs, rather than being experimentally assigned, they are not independent in their academic performance. The typical outcome is that they sink or swim together: either both do better or both worse than expected. Once again, however, birth order mediates this effect. The effect occurs only if at least one member of the pair is later-born; when both are first-born, there is no clear evidence of RMs' influence upon academic performance.

Tests were also made to determine whether RMs became more alike in other kinds of educationally relevant activities, such as cultural activities and extracurricular participation. All RMs (both experimentally assigned and mutually chosen) tended to be alike in frequency of attendance at plays and in their extent of joining extracurricular organizations. There was no evidence of their being similar in other cultural activities such as concerts, art exhibitions, and special lectures.

Interpretation of Experimental Conditions. In the experimental part of this research, matched pairs of students were assigned to two different "treat- ments." One treatment was sharing a room with a high-ability RM; the other treatment was sharing a room with a low-ability RM. To help understand the experimental conditions, several types of evidence were collected as to how, on the average, high RMs did differ from low RMs as stimulus persons.

Students with high HSR were, as expected, clearly better than low stu- dents in their academic performance. The mean difference in grade point average (GPA) between highs and lows was about .5 and was statistically sig- nificant. Judging from students' ratings of their own academic ability relative to their RMs' ability, this difference was clearly evident to the students. Various ratings by students and counselors showed the following additional differences between the high-ability and low-ability students in the expected direction: high students studied more, had more interest in studies, more

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 9: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 301

often set a good example of studying hard, more often encouraged their RMs to study, and less often distracted their RMs from studying.

To the extent that students of high and low ability differ in their fields of study, the assignment of RMs on the basis of HSR might result in differences between conditions in the number of courses shared with the RM, in addition to the differences in ability and study habits described above. To check this possibility, the number of course credits that each RM pair had in common was counted. Both identical courses and courses very similar in subject matter were counted as course overlap. RMs who were upperclassmen, regardless of experimental condition, rarely had courses in common, and their course overlap was not further examined. All freshmen were divided approxi- mately at the median into those with high overlap (8 or more credits) and those with low.

Among freshman RMs who remained together at least one quarter, HH pairs had the most course overlap, and HL pairs the least. Of 17 HH pairs, 13 had high overlap. Of 31 I-L pairs, 12 had high overlap. Of 12 LL pairs, 6 had high overlap. By Chi 2 on this 2 X 3 table, the difference is significant at the .05 level, as is the difference in a 2 X 2 table for HH vs. HL pairs. Hence the intended experimental treatment and course overlap with RMs were confounded among high students, and some effort to determine their separate effects was necessary.

It was also expected that high and low RMs would differ in the stimula- tion they provided concerning cultural and intellectual matters in general, apart from course work and studying. In general, this was not the case on the measures used here, which included reports (both by the students them- selves and by counselors) concerning the students' interest in and attendance at cultural activities and their participation in extracurricular activities.'

In summary, the experimentally assigned RMs with high HSR differed as experimental "treatments" in the expected ways in matters closely related to studies, but not in more general cultural and intellectual interests nor in extracurricular activities. However, the experimental treatments had another, unintended effect, too: students in the different conditions differed in the extent to which they shared courses with their RMs.

Academic Performance in Relation to Roommate's Ability and Course

1 One item on which students compared themselves with their RMs concerned interest in music, literature, and art. On this item, regardless of experimental condition, there was a strong, statistically significant tendency for students to describe themselves as more in- terested than their roommates. This result suggests the existence of a norm in this setting opposing open expression of cultural interests, resulting in pluralistic ignorance of other stu- dents' cultural interests.

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 10: The Educational Influence of Dormitory Roommates

302 SOCIOMETRY

Overlap. The experimental hypothesis, as noted above, was that students with high-ability RMs would perform better academically than students (of matched ability) with low-ability RMs. The hypothesis was tested separately for high students by comparing the grade point average (GPA) of H (H) with that of matched H(L) cases, and for low students by comparing L(H) with matched L (L) cases. The analysis of variance design described above was used to test differences between experimental conditions and differences among the five dormitory categories in the size of the experimental effect.

Table 1 presents the mean GPA of students in each condition, including only RM pairs that remained together at least until December 1 (over 2 months).

TABLE 1

Mean Grade Point Averages, by Experimental Condition

High Students Low Students

n* H(H) H1(L) Difference n* L(H) L(L) Difference

Freshmen only Fall GPA 31 2.49 2.35 0.14 22 1.91 1.89 0.02 Winter GPA 21 2.61 2.40 0.21 13 1.85 2.08 -0.23

All cases Fall GPA 39 2.39 2.25 0.14 32 1.86 1.95 -0.09

Cases for covariance (Freshmen) Fall GPA 25 2.56 2.28 0.28 19 1.89 1.84 0.05 ACE 25 56.2 59.9 -3.7 19 48.3 50.0 -1.7 * n equals number of matched pairs.

The mean fall GPA of H(H) students was higher than the mean GPA of matched H (L) students, but the difference was not statistically significant, and the differences among dormitory categories in the experimental effect were not significant.

The loss of cases was greater among low students than among high ones. Table 1 shows very small and inconsistent differences between L(H) and L(L) students. These differences were not significant for freshmen nor for all cases, and there were no differences among dormitory categories in the effect.

Since it had not been possible to match for ACE test score in original as- signments, analyses of covariance were conducted to control ACE score. These analyses required dropping additional matched pairs in which ACE score was not available for both cases. For the analysis of high students there were only 3 pairs of upperclassmen, and for low students only 5 pairs of up- perclassmen. Hence, analyses were conducted on freshmen only. Table 1

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 11: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 303

shows mean GPA and mean ACE, by experimental condition, for the cases that could be used in the covariance analysis. These tests showed the H(H) students significantly higher in GPA than H (L) students (F =4.53, df = 1.23, p<.05), but still no significant difference between L(H) and L(L).

It should be noted, however, that these analyses of covariance involved not only controlling for ACE scores, but also dropping matched pairs that did not have complete data. From Table 1 it is apparent that the cases lost were less favorable to the hypothesis than those retained, since the difference for cases used is greater than than the difference for all freshmen.

The corresponding means of winter quarter GPA for surviving matched pairs of freshmen are also reported in Table 1. The loss of cases was severe- so severe among upperclassmen that they were not even included. Neither the differences between H(H) and H(L) nor L(H) and L(L) are significant for freshmen.

It appears that, among high students, the experimental conditions may have resulted in a significant difference in fall quarter GPA. However, since course overlap with the RM was confounded with RM's ability level, further analysis was needed to clarify its effects.

In order to test the effects of course overlap and for other tests in which matching of pairs on HSR could not be used, another method was devised to control HSR in examining academic performance. Each student's grade point average (GPA) was first computed for fall and winter quarters sepa- rately. Then GPA was correlated with HSR within each dormitory category so that the correlations could be used as a basis for adjusting the grade aver- ages. The correlations are reported in Table 2. These correlations were

TABLE 2

Correlations of High School Rank with Grade Point Average

Experimental Cases Mutually Chosen Cases Dormitory Fall Winter Fall Winter

Category r n r n r n r n

1 .31* 52 .22 38 .65** 38 .40* 28 2 .13 18 .19 9 .36* 38 .44* 34 3 .45* 30 .46* 24 .48* 26 .59** 25 4 .15 28 -.28 16 .54* 17 .42 17 5 .76** 38 .80** 32 .81** 18 .83** 17 6 - - - - .32 29 .50** 26

Significance of differences p<.01 p<.01 n.s. n.s.

Average r .43** .43** .53** .53**

*P<.05; **p<.o1.

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 12: The Educational Influence of Dormitory Roommates

304 SOCIOMETRY

transformed to z, and within each column of the table they were averaged over the dormitory categories and differences among the correlations tested for significance.

In fall quarter all the correlations for freshmen (categories 1, 3, 5) are significant, but the correlations for upperclass experimental cases (categories 2 and 4) are small and non-significant. Also, there are significant differences in correlation among dormitory categories for experimentals. Hence, the two upperclass categories were excluded from the adjustment procedure. Among mutuals in fall quarter, there were no significant differences in correla- tion among the six dormitory categories. Also, the average r and five of the six separate ones were significant. Among mutuals all dormitory categories were included in the adjustment procedure.

Within each category the equation for the linear regression of GPA and HSR was computed and a predicted GPA for each student, based on his HSR, was obtained from this equation. Each student's adjusted GPA was his obtained GPA minus his predicted GPA. Hence, each adjusted GPA ex- presses, in effect, the student's deviation above or below the mean GPA of cases with the same HSR in the same dormitory, sex, and class. The adjusted GPA is interpreted as an index of academic performance in college relative to earlier performance in high school, i.e., an index of change in academic performance concomitant with entry into college.

In analyzing the extent of course overlap between RMs, as noted above, all freshmen RMs were divided into two categories of course overlap as near the median as possible. Among freshmen experimental cases the mean ad- justed GPA was computed separately for the two categories of overlap and four experimental conditions. These means are reported in Table 3. The mean of all 62 students with high course overlap was higher than the mean of 58 with low overlap. This difference was significant whether treated as individual observations (F - 5.25; df = 1, 118; p<.O25) or as means of RM pairs (F = 5.97; df = 1, 58; p<.O25).

TABLE 3

Mean Adjusted GPA of Freshmen, by Experimental Condition and Course Overlap

H(H) H(L) L(H) L(L) Total Course overlap n M n M n M n M n M

High course overlap 26 .11 12 .23 12 .17 12 .06 62 .14

Low course overlap 8 -.25 19 -.28 19 .07 12 -.12 58 -.13

Total 34 .03 31 -.08 31 .11 24 -.03 120 .01

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 13: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 305

Since course overlap and experimental condition were confounded, the next effort was to determine whether one or both contributed to differences in aca- demic performance. It should be noted in Table 3 that the mean difference in adjusted GPA between high and low course overlap is consistently in the same direction within all four of the experimental conditions. On the other hand, the differences between the H(H) and H(L) experimental conditions within conditions of course overlap are small and inconsistent. Hence, it is probable that the total difference in the predicted direction between H(H) and H(L) conditions is attributable primarily to the difference in course over- lap rather than HSR level of the RM.2

In an effort to establish the specific ways in which course overlap affected the academic achievement of experimental RMs, a further analysis of the data was done. As might be expected, RMs with high course overlap received more help from one another in their studies (p<.01) and discussed studies more often with one another (.10>p>.05) than did RMs with low course over- lap. High overlap RMs have more knowledge of one another's grades on examinations (p< .01), but they report no greater competition with one another than do RMs with low course overlap.

Birth Order as a Mediator of the Roommate's Influence. Recent research in social psychology 3 has suggested that first-born children are more socially dependent than later-born children and that this difference persists into adult- hood. That is, first-born more often than later-born persons seek the help of others or the company of others when faced with a difficult or ambiguous situation. If the first-born and later-born students in this sample differ in this respect, it might be expected that first-born students would seek more help in coping with their studies. The success of first-borns in academic work might be more contingent upon circumstances that make it easy to discuss studies and obtain help. Also, first-borns might be more prone to follow the example set by a RM in study habits, and later-borns might be more con- sistent in setting an example. Also, to the extent that such differences between first-borns and later-borns in social dependence obtain in this sample, high course overlap would help the performance of first-born students more than later-born ones.

The effect of birth order in mediating the effect of the RM's ability was

2 This result is a clear illustration of a major methodological problem of certain types of field experiments in which the intended experimental variable is systematically corre- lated with other potentially effective variables. This problem is acutely present in studies such as the present one in which an experimental manipulation is based upon people's characteristics or background.

3 Stanley Schachter, The Psychology of Affiliation, Stanford, California: Stanford Uni- versity Press, 1959, Chapters 5 and 6.

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 14: The Educational Influence of Dormitory Roommates

306 SOCIOMETRY

explored among freshmen experimental cases by computing mean adjusted GPAs separately for each combination of birth order and course overlap. Since the amount of course overlap was confounded with the ability level of the RM, the means were computed separately for students with high RM -the H(H) and L(H) cases-and those with low RM-the H(L) and L(L) cases. Table 4 reports mean adjusted GPA, classified into high and low course overlap, high and low RM, first-born and later-born subjects, and first-born and later-born RM, for a total of 2 X 2 X 2 X 2 = 16 cells and the various marginal combinations of these. For simplicity in referring to Table 4, the columns are labelled with letters A through I, and the rows with numbers 1 through 9.

From examining the two-by-two table classified by RM's HSR level and RM's birth order (G7, G8, H7, H8) it appears that academic performance is correlated with the RM's HSR level when the RM is later-born but not when he is first-born. The difference for later-born RMs (G8 vs. H8) is sig- nificant (t = 2.44; df = 59; p<.02), while that for first-born RMs (G7 vs. H7) is not. Also, the difference between G8 and H8 is significantly greater than the difference between G7 and H7 (t = 2.35; df = 116; p = .02).

Could this apparent mediating effect of the RM's birth order be a result of confounding the RM's HSR level with course overlap and with birth order of the subject? Two observations argue against the possibility. First, the same pattern of means found in the 2 X 2 table (G7, G8, H7, H8) is found also, with perfect consistency in spite of small n's, in each of four 2 X 2 tables that are homogeneous in overlap and S's birth order. That is, in Table 4, A2>B2, A4>B4, C2>D2, and C4>D4, and also (A2-B2)>(A1-B1), (A4-B4)>(A3-B3), (C2-D2)>(C1-D1, and (C4-D4)>(C3-D3). Sec- ond, the mediating effect its still significant (p<.05) if pairs with high over- lap and pairs with low overlap are treated as two different populations, and t tests are performed separately on the table (A7, A8, B7, B8) and the table (C7, C8, D7, D8), and then combined.4

Could this apparent effect be an artifact of differences between sexes or between dormitories, which are disproportionately represented in the cells of the 2 X 2 table? Apparently not. When the same 2 X 2 classification is made separately for subgroups of the same sex and living in the same dor- mitory, the same pattern of means obtains for males and females separately and for each dormitory. It appears, then, that later-born RMs do influence students' academic performance more than first-born RMs do.

Turning attention to the possible mediation of effects of course overlap

4 Two-tailed probability, computed by Fisher's method. See R. A. Fisher, Statistical Methods for Research Workers, Edinburgh and London: Oliver and Boyd, 1946 (10th edition), pp. 99-101.

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 15: The Educational Influence of Dormitory Roommates

TABLE 4

Mean Adjusted GPAs of Freshman Experimental Cases 4 High Low

Course Overlap Course Overlap Course Overlap Roommate High RM Low RM High RM Low RM High Low High Low Overall Z Birth order of: A B C D E F G H I )

Subject Roommate n M n M n M n M n M n M n M n M n M o 1. First First 9 -.26 5 .44 5 -.36 9 -.28 14 -.01 14 -.31 14 -.29 14 -.02 28 -.16 t 2. First Later 11 .37 7 -.11 3 -.17 10 -.44 18 .18 13 -.38 14 .26 17 -.31 31 -.05 Pd 3 . Later First 12 .05 6 .20 10 .08 3 .27 18 .10 13 .12 22 .06 9 .22 31 .11 8 4. Later Later 6 .42 6 .15 9 .10 9 -.08 12 .28 18 .01 15 .23 15 .01 30 .12 y 5. First Combined 20 .09 12 .12 8 -.29 19 -.36 32 .10 27 -.34 28 -.02 31 -.18 59 -.10 6. Later Combined 18 .17 12 .18 19 .09 12 .01 30 .17 31 .06 37 .13 24 .09 61 .11 7. Combined First 21 -.08 11 .31 15 -.07 12 -.14 32 .05 27 -.10 36 -.08 23 .07 59 -.02 8^ Combined Later 17 .39 13 .01 12 .03 19 -.27 30 .22 31 -.15 29 .24 32 -.16 61 .03 T 9. Overall 38 .13 24 .15 27 -.02 31 -.22 62 .14 58 -.13 65 .07 55 -.06 120 .01 u

0 4

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 16: The Educational Influence of Dormitory Roommates

308 SOCIOMETRY

by birth order, it appears in Table 4 that first-born subjects do considerably better with high course overlap (B5) than with low course overlap (F5), while overlap makes little difference for later-born subjects (E6 vs. F6). The difference, E5 vs. F5, is significant (t 2.43; df = 57; p<.02), while E6 vs. F6 is not, but the difference between these two differences is not significant. An examination of this 2 X 2 table separately for each sex and dormitory reveals that this apparent mediation holds true for males but does not hold true for females, though females generally perform better with high course overlap. The means for males only corresponding to E5, F5, E6, and F6 in Table 4 are respectively .25, -.38, .14, .09. The difference between .25 and -.38 is significant (t - 2.71; df - 39; p = .01), and the difference between .14 and .09 is not. Furthermore, the difference between these differences closely approaches significance (t 1.90; df -78; p approximately .06). On the possibility that these findings for males could result from confounding course overlap with the RM's HSR level and birth order, similar comparisons of means were made within the four 2 X 2 tables that are homogeneous on both these variables. As above, the mean differences, with small n's, are con- sistent in each case; i.e., in each case the adjusted GPA of first-borns is higher with high course overlap than with low, and the difference be- tween high and low course overlap is greater for first-borns than for later- borns. It appears, then, that high course overlap with the RM benefits first- born students more than later-borns.

Birth order was not controlled experimentally but appears after the fact to be an important mediator of the RM's influence upon academic perform- ance. It is important to try to determine whether some well known correlate of birth order, such as size of family or socio-economic level of parents, might account for the results. Each of the 2 X 2 tables in which mediation by birth order was observed (E5, E6, F5, F6 and G7, G8, H7, H8) was tabulated separately within relatively homogeneous groupings of subjects on family size and on father's education (as an index of socio-economic level). Subjects were divided into three categories of paternal education: grammar school only, at least some high school, and at least some college. Within each category the 2 X 2 breakdown was repeated for males and females separately and combined. This resulted in small and disproportionate numbers of cases per cell, and no inferential statistics were used to try to make probability state- ments about separate tables. If the relevant differences among means were the same in each homogeneous 2 X 2 table, this was regarded as adequate evidence that the over-all effect was not attributable to the controlled variable.

The mediation of the convergence effect by RM's birth order remained the same for both males and females in tables controlling family size and in tables controlling father's education. The mediation of the course-overlap

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 17: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 309

effect by subject's birth order remained consistent for males throughout. It was noted above that females generally did not show any mediation by birth order. It appears that the safest tentative conclusion is that birth order, rather than family size or socio-economic level, is the variable that mediates the effects of RMs upon each other's academic performance.

Various questionnaire items about interaction between RMs were tabulated separately for each combination of birth order and course overlap. Unfortu- nately, these analyses showed no differences in amount or type of interaction which would illuminate the specific ways in which first-born male RMs were more affected by course overlap than later-born RMs, or the ways in which later-born RMs of both sexes exerted more influence.

Academic Performance of Mutually Chosen Roommates. Roughly analogous tests of the effects of the RM's ability level and course overlap were devised for the students who had chosen each other as RMs. First the mean adjusted GPA of 46 freshman mutuals who had high course overlap with their RMs was compared with the mean of 34 with low overlap. In contrast with ex- perimental cases the difference between the means was only .02 and non- significant. When the difference between high and low course overlap was ex- amined separately for different birth order combinations, no differences were significant. However, the numbers of cases per cell were small, and no firm conclusion can be drawn about the interaction between course overlap and birth order.

Next, to test whether mutual roommates who differed in HSR tended to converge in academic performance, each pair was classified into the higher RM and the lower one in HSR. Pairs in which the RMs differed in HSR by only three points or less (on the T scale) were eliminated, leaving 60 RM pairs. If there were a tendency to converge, the adjusted GPA of the RM lower in HSR should be higher than the adjusted GPA of the RM higher in HSR. A t test for correlated observations revealed no significant difference. There is no evidence of convergence in the academic performance of mutuals.

It should be noted that among mutuals, unlike experimentally assigned cases, the size of the HSR difference between RMs was quite variable. Hence, a second test was conducted to determine whether there was a stronger tendency for convergence in performance between RMs who were very different in HSR than between those relatively similar in HSR. A correla- tional test of this question was conducted separately for high students and low ones. Without describing technical details, it can be stated that there was no evidence of greater convergence with large HSR discrepancies than with small ones.

Similarity Between Roommates in Study Habits and Grades. Another hypothesis concerned the tendency toward behavioral similarity of RMs,

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 18: The Educational Influence of Dormitory Roommates

310 SOCIOMETRY

regardless of ability level. As a consequence of close interaction, RMs were expected to become more similar in their study habits and educational inter- ests-and therefore in their academic achievement and participation in vari- ous kinds of activities. Interviews with students revealed a number of specific instances of behavior that might contribute to this effect. For example, some students reported attending concerts or special lectures because the RM did, joining a particular extracurricular organization in which the RM was active, and developing an interest in music because of RM's playing records and dis- cussing them. Concerning study habits there were reports of studying late because the RM was studying, exchanging lecture notes, and studying more because the RM closed the door and turned away casual visitors.

Statistical tests of similarity between RMs were conducted separately for experimental and mutual RMs. Before and after measures were not obtained, and, therefore, influence had to be inferred from the existence of similarity after one quarter. If only the mutual RMs tended to be similar on a particular measure, it was interpreted as probably a result of the way RMs selected each other. If experimental RMs were similar, the similarity was regarded as probably the result of influence exerted through some of the kinds of behavior cited from the interviews.

The first statistical tests concern similarity between RMs in time spent studying. Intra-class correlations between RMs were computed on the coun- selors' ratings of time studying and on the students' self-reports of time studying. Since there were known differences between dormitory categories in time studying, these correlations were computed separately within each category. For experimental cases the correlations gave no evidence of similarity. For mutuals, on the other hand, as Table 5 shows, the intra-

TABLE 5

Similarity Between Mutual Roommates Intra-class Correlations

Time Studying Number Hours

Dormitory Counselor's Self in in Cultural Category n Rating Report Organizations Organizations Participation

1 19 .45*a .51*a 47* .42 .03 2 21 .74** .60** .26 .10 .38* 3 13 .64** .32 .33 .06 -.06 4 8 .45 .17a .30 .22 .03 5 9 -.02 .16 -.37 -.40 -.09 6 14 .17 .26 .50* .10 .56*

n=number of pairs of roommates * p<-05; ** p<.01. a Based on one less than indicated n.

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 19: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 311

class correlations were rather consistently positive and were significant within several dormitory categories on both measures.

The hypothesis that RMs achieve similarly, relative to previous achieve- ment in high school, was tested by computing intra-class correlations be- tween adjusted GPAs of RMs, with results given in column D of Table 6. The correlations were computed separately for experimental freshmen,

TABLE 6

Similarity of Academic Performance of Roommates: Intra-class Correlations by Birth Order and Course Overlap

Birth Order

A B C D Both One First, Both All Pairs

First-Born One Later Later-Born Combined Course Overlap ri n ri n ri n ri n

Experimental 1. High -.04 7 .23 18 .39 6 .17 31 2. Low -.59 7 -.19 13 -.46 9 -.35 29 Freshmen 3. Combined -.32 14 .04 31 -.10 15 -.08 60

Mutual 4. High .10 9 -.03 9 .52 5 .16 23 5. Low a 2 .73** 8 .64* 7 .68** 17 Freshmen 6. Combined .14 11 .42** 17 .66** 12 .46** 40

Mutual 7. Combined -.10 7 .64** 18 .68** 13 .47** 38 Upperclassmen

n=number of pairs of roommates. a=not computed because of small n. *p<-05; **p<.01.

mutual freshmen, and mutual upperclassmen. Experimental upperclassmen were not included because, as explained above, adjusted GPAs could not be computed. Correlations were not computed separately for each dormitory category because the adjustment of GPA had assured no mean difference between categories and hence no spurious inflation of the correlation. A corre- lation could be generated, however, by the courses that RMs take in com- mon, as an artifact of different grading standards in different courses. If this were true, the correlation should be higher between RMs with high course overlap than with low overlap. Hence, the correlations were also computed separately by course overlap conditions.

Among experimental freshmen the correlation was not significant for all pairs combined (D3 in Table 6) nor in high or low overlap condi- tions (D1 and D2). For mutual freshmen the correlation was .46 (D6) and for upperclassmen, .47 (D7), both significant. Comparing high and low overlap conditions for mutual freshmen (D4 and D5), the correlation is higher with low course overlap, thus ruling out the possibility that the corre-

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 20: The Educational Influence of Dormitory Roommates

312 SOCIOMETRY

lation is an artifact of grading standards. Among experimentals the academic achievement of RMs was not correlated. Among mutuals it was correlated; both tended to overachieve or both to underachieve. It was shown above (Table 5) that mutuals were similar in time studying, which may help account for their similarity in achievement. In order to test whether mutual RMs also were similar in ability, intra-class correlations of HSR and ACE scores were computed within each dormitory category. The HSRs of mutual RMs are not significantly correlated in any category. Correlations between RMs' ACE scores are inconsistent and puzzling. In only two of the six categories is the intra-class correlation significant, and these two are in opposite directions. In category 2 (Dorm A upperclassmen), with 16 pairs, ri- .58 (p<.01), while in category 5 (Dorm B freshman girls) with 8 pairs, ri -.73 (p<.5).

Since experimental RMs are not correlated in achievement and mutuals are, it is probably a result of the method of selecting RMs, not of influence. Since some categories of mutuals are similar in time studying and some pos- sibly in ACE, a clear conclusion concerning the basis of choice is not possible.

Table 6 also reports the intra-class correlations between RMs in adjusted GPA separately for each combination of birth orders. Among experimentals the correlations are not significant for any combination of birth order and course overlap. The only consistent pattern appears to be that the correla- tion is higher with high course overlap, and this may be an artifact of grad- ing standards. Among mutuals both birth order and course overlap appear to make a difference. RMs are not correlated when both are first-born, among either freshmen (A6) or upperclassmen (A7), and they are significantly corre- lated in the other two birth order combinations (B6, B7, C6, C7). The corre- lation is higher with low course overlap (D5) than with high (D4), the differ- ence between these correlations being significant (p<.05), and this direc- tion of difference holds in different birth order combinations (B4 vs. B5, and C4 vs. C5).

The reasons for the rather striking difference associated with course over- lap are not clear. Probably it has its origin in the basis for selection of a RM. A possible guess, though risky without further evidence, is that students tended to select RMs who were similar to themselves either in study habits or in courses taken. Thus, those who had little course overlap would have to be chosen on similarity of study habits.

From Table 6 it appears at first glance that the very low correlations be- tween mutual RMs both of whom are first-born might be simply a result of the accidental fact that most of them have high course overlap. Closer examina- tion makes this seem unlikely. First, the birth order difference is striking among upperclass mutuals (A7, B7, C7). Examining these pairs to see how

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 21: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 313

they would have been classified if upperclassmen were broken down on course overlap, it turns out that all 7 first-born pairs have low overlap, while 2 later-born pairs and 5 mixed pairs have high overlap. The low overlap in first-born pairs would be expected to raise their correlation relative to other birth order combinations. As a further check, the data from cells A5 and A7 were combined to obtain 9 first-born pairs with low overlap, and ri = -.13 for these pairs in contrast to correlations over .6 and significant for pairs in other birth orders (B5, C5, B7, C7). Hence, it appears that mutual RMs are similar in academic performance only if they are not both first-born. Discussion of this observation will be deferred.

Since the results obtained on experimentals and mutuals are often strikingly different, some comment is required on possible reasons for such differences. First, the two populations differ in that experimental cases did not request any particular RM. This may possibly mean that experimentals differ in such characteristics as extroversion, urban or rural origins, social class, etc. The upperclass experimentals appeared to be particularly selected, judging from interviews and informal observations, in being either transfers from another college, or rather asocial, or little motivated to perform as students. Second, mutual RMs were nearly always congenial and friendly with each other, while a substantial minority of experimentals showed indifference or even antipathy.

It was possible that greater congeniality might account for generally higher intra-class correlations between mutual RMs than between experi- mentals. As a test of this, the intra-class correlations were repeated on only those experimental pairs high in congeniality (the top 2/3, approximately). The correlations did not change substantially, suggesting that congeniality does not account for the differences in results between experimentals and mutuals.

Additional analyses were conducted to determine whether some results based on the intra-class correlation of adjusted GPA might be artifacts of high school differences. The adjusted GPA is a derived measure, based on a correlation between HSR and GPA, and the students in the study came from many different high schools. Hence, differences between high schools in average ability of students might result in higher correlations between students from the same high school than between students from different ones. It is also possible that percentile rank in a large high school class might be a more reliable measure or more valid for prediction than HSR in a small school. Hence, there might be higher correlations for students from large high schools. Analyses were conducted to determine whether either of these factors was confounded with course overlap and birth order in such a way as to over-represent pairs from the same high school or students from large high schools in cells where a large intra-class correlation was found. In brief,

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 22: The Educational Influence of Dormitory Roommates

314 SOCIOMETRY

these potential artifacts do not account for any of the reported differences between birth orders or conditions of course overlap. The over-all tendency for more positive correlations among mutuals than among experimentals may, however, be attributable to the fact that over 50 per cent of freshman mutual pairs of RMs attended the same high school, while no experimental RM pair did.

Similarity Between Roommates in Cultural and Extracurricular Activities. In the questionnaire, information was obtained on the number of times during fall quarter that each student attended concerts, plays, university convoca- tions, art exhibitions, and special lectures. Also, each student named the formal student organizations to which he belonged and estimated the number of hours per week spent in each.

Examining frequencies of attendance at each of the five types of cultural activity separately, it was noted that the majority of students reported no at- tendance, and the distributions were badly skewed. For a simple test of simi- larity between RMs on each of these separate measures, each distribution was dichotomized into no attendance vs. attendance at least once, and Chi-squared tests were performed. The relative frequencies of attendance for individuals in each dormitory category were computed separately, and they differed strik- ingly among categories. Within each category and for each type of activity separately, the expected frequency of having both RMs attend, assuming they did so independently, was obtained by squaring the individual proba- bility of attendance and multiplying by the total number of RM pairs. Simi- larly expected frequencies of having only one RM attend, and neither RM attend, were obtained. The two expected frequencies for "agreement" between RMs were added over dormitory categories. A Chi-squared test with one df for each measure was then conducted to determine whether RMs were more frequently alike (in attending or not attending) than would be expected on the assumption of independence between RMs. The whole procedure was fol- lowed separately for experimental and mutual cases, so that there were 10 Chi-squares.

Roommates were significantly more often alike than expected in attendance at plays, among both experimental cases (p< .05) and mutual cases (p<.05). In concerts and art exhibitions both experimental and mutual cases showed somewhat more similarity than expected, but not significantly so. In convo- cations and special lectures, frequencies were very close to expected with no tendency to similarity of RMs.

The frequencies of attendance at all of the five types of cultural activities named above were added together for each individual in order to obtain an indication of total cultural participation. Also, two scores of participation in extracurricular activity were obtained: the number of formal organizations

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 23: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 315

belonged to, and the estimated hours per week spent in activities of formal organizations. For experimental cases the distributions of these three scores were extremely skewed, and similarity between RMs was tested by Chi- squared. Within each dormitory category each measure was dichotomized as near the median as possible. Expected frequencies of having both RMs on the same side of the median and of having one above and one below the median were computed as above. Then Chi-squared with one df was used to test simi- larity between RMs. The same three measures for mutual RMs were distrib- uted approximately symmetrically, and similarity between RMs was tested by computing an intra-class correlation between RMs within each dormitory category.

It appears that RMs tend to be more similar than expected by chance in the number of formal organizations they belong to. Among experimentals there is more similarity than expected among both freshmen and upperclassmen, and the differences are significant by Chi-squared (p<.02) for all students combined and for freshmen separately. Among mutuals, five of the six cor- relations are positive, two of them significantly so (p<.05) as shown in Table 5, above. The exception is dormitory category 5, which has a negative correlation (not significant), based on only nine pairs of roommates. There is no evidence of greater than chance similarity between RMs in hours spent in organizations. The reports of hours spent are probably less reliable than reports of what organizations they belong to.

In total cultural participation, the experimental RMs are no more similar than expected, nor are mutual freshmen (categories 1, 3, and 5). The upper- class mutuals (categories 2, 4, and 6) show some evidence of similarity in frequency of attendance at all cultural events. Their similarity is probably a basis for selection of each other, rather than an indication of influence, since it does not appear among experimental cases.

Personal Relations Between Roommates. Another potential effect of the ex- perimental pairing of RMs was in the stability and strength of friendship between them and the types of interaction that occur. During the academic year following the assembly of RMs, all instances in which pairs of RMs sep- arated were determined from dormitory records. Of all 96 experimental pairs assembled, 15 out of 24 HH pairs, 23 out of 48 HL pairs, and 6 out of 24 LL pairs remained together the entire academic year. The respective percentages were 62%, 48%, and 25%. By a Chi-squared test on this 3 x 2 table, these differences were significant at the .05 level.

It should be noted that these differences in stability among the three types of pairs resulted, in part, from reasons other than incompatibility. A few assigned pairs changed RMs without ever living together. Some students dropped out of school, and some moved to another type of residence, such as

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 24: The Educational Influence of Dormitory Roommates

316 SOCIOMETRY

a fraternity. The only clear-cut evidence of break-ups caused by incompati- bility was that provided by counselors' records (kept only during the fall quarter) of complaints about RMs and the subsequent disposition of them. Six pairs (1 HH, 2 HL, 3 LL) broke up during fall quarter for reasons of incompatibility, as reported to counselors.

To estimate differences in compatibility over the entire three quarters, it seemed reasonable to regard break-ups in which both RMs subsequently re- mained in the dormitory as instances of incompatibility. Combining these with the known instances of breaking up for incompatibility during the first quarter shows 2 out of 24 HH pairs, 5 out of 48 HL pairs, and 7 out of 24 LL pairs incompatible. The percentages of break-ups due to incompatibility are re- spectively 8%,o 10%,o and 29%o. Paralleling the previous finding, the LL pairs are least compatible, but the HH and HL pairs differ very little. Theoretical frequencies of break-up are too small to permit a Chi-squared test on this 3 x 2 table, but each of the three pairs of conditions was compared by Fisher's exact test for a 2 x 2 table, and no differences were significant. The two-tailed probabilities were: HH vs. HL, p>.5; HL vs. LL, p=.10; HH vs. LL, p =.14.

So few mutual pairs of RMs separated during the year that no statistical analysis was attempted.

A number of other measures of compatibility and closeness of friendship were taken from the questionnaire and counselors' ratings. Among experi- mental cases, these measures were analyzed for differences among experimental conditions by the analysis of variance with matching on HSR. The measures included ratings of the desire to change RMs, closeness of the relationship between RMs, frequency of leisure-time activities together with RM, frequency of eating meals with RM, and admiration and respect for the RM. The major relatively consistent finding is that, among experimental cases, the high-achiev- ing students are more satisfied and compatible with high-achieving RMs than with low-achieving RMs. This may be in part attributable to differences in distraction from study and help in studies described above. It appears to be true, but with less clarity and consistency, that low-achieving students are also more satisfied with high-achieving RMs than with low RMs.

Roughly analogous analyses were conducted on mutual RMs to determine whether the size of the difference between RMs in HSR was associated with differences in personal relations. Without describing technical details, it can be reported that there was evidence of such a relationship on only one of the relevant questions in the questionnaire. The more similar in HSR, the more the mutuals reported that their RMs discussed and advised about personal affairs. In general, on questions of compatibility and closeness of friendship,

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 25: The Educational Influence of Dormitory Roommates

INFLUENCE OF ROOMMATES 317

mutual RMs showed higher means and less variance than the experimental RMs.

DISCUSSION

This study examined several classes of student behavior that presumably produce different educational consequences, including studying for classes, discussion, extracurricular activities, cultural activities. One hypothesis was that roommates (RMs) would influence each other toward uniform behavior, i.e., towards a similar extent and type of participation in each class of activity. In fact, similarities between those roommates who had been experimentally assigned were rare after one quarter. RMs were more alike than chance ex- pectancy in attendance at plays and in belonging to student organizations- and this was true for both experimentally assigned and mutually selected RMs. The mutual RMs were also alike in time spent studying, over-all participa- tion in cultural activities (upperclassmen), and academic achievement. The similarities that occur only between mutual RMs are presumably the result of their methods of selecting a RM, rather than of mutual influence.

As expected, students of high academic ability, as indicated by high school percentile rank (HSR), are more likely than low students to study diligently, according to the reports of both RMs and counselors. Also, as expected, the high students get better grades. The major hypothesis for experimental test was that students living with high RMs would study more and obtain better grades than matched students living with low RMs. A comparison of students experimentally assigned to high RMs with matched students assigned to low RMs showed that they perceived important differences (in the sense of de- scribing the two types of RMs differently). For example, high RMs were seen as setting a better example in study, providing more encouragement and praise for study, being more desirable as RMs, more stimulating, and less of a dis- traction from studies. Though RMs of different ability levels showed dis- criminable differences as stimulus persons, there was no evidence that they had a differential over-all effect on grades, when proper controls were made.

An exploratory search for particular circumstances under which the RM's ability level might affect grades turned up some evidence that later-born RMs may have the expected effect. It appears that later-borns exert more influence on their RMs, but it is not clear just how this comes about. Among experi- mentally assigned pairs, a later-born RM's influence appears to be exerted in the direction originally predicted on the basis of HSR. Among those who chose each other as RMs, there is a tendency for both RMs to perform better or both worse than would be expected from their HSR, provided at least one of them is later-born. However, the direction of the influence (whether for

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions

Page 26: The Educational Influence of Dormitory Roommates

318 SOCIOMETRY

better or worse performance) cannot be predicted in this study. In neither instance is it clear how social interaction with a later-born RM differs from that with a first-born RM so as to bring about differences in influence. In both instances, however, the presence of a later-born RM appears to be a necessary condition for interpersonal influence on academic achievement.

Apparently two variables that were not experimentally controlled had con- siderably more effect than expected. These were the extent of course overlap between RMs, and the birth order of subjects and RMs. In general, experi- mentally assigned pairs of RMs who shared many courses did better than those who did not share courses. Furthermore, among males the beneficial effect of sharing courses was especially strong for first-born students.

The significance of birth order as a mediator of influence in interpersonal relations appears consistent with various evidence cited by Schachter 5 that first-born persons are more socially dependent under certain circumstances than later-borns. There is no basis in this study for examining the etiology of this dependent behavior. This study can only reinforce Schachter's findings that suggest that important differences exist between the socialization of first- born and later-born children, and research on these differences in socialization would be useful for understanding susceptibility to social influence.

Though every effort was made to examine alternative explanations, there still remains the possibility that some correlated variable which could not be controlled in this study, rather than birth order, is responsible for the effects. For this reason it is important to cross validate the results in further studies which assign RMs experimentally on the basis of birth order and course over- lap, as well as HSR level, before putting these results to practical use.

5 Loc. cit.

This content downloaded from 185.44.78.143 on Wed, 25 Jun 2014 01:55:34 AMAll use subject to JSTOR Terms and Conditions