The data in this project were collected from a questionnaire answered by college students enrolled in summer session courses in 1997 at a predominately white mid-sized university located in North Carolina. I obtained permission to enter classrooms to administer a questionnaire by sending letters or e-mail to professors teaching classes during the summer. A total of ten classes participated with a total of 140 questionnaires collected. Participation was voluntary and the questionnaires were completed anonymously. Of the 140 questionnaires, only students who attended the university the previous fall or spring semesters and self-identified themselves as black or white were used. One hundred and twenty-six questionnaires qualified after this initial screening.

In the social sciences, the data commonly do not meet the assumptions needed to conduct statistical testing. When dealing with race, blacks are a clear minority and obtaining a large enough sample size for analysis is sometimes difficult. In this project, there was a large white sample but a small black sample. Many statistical tests prefer equal (or very similar ) sample sizes for each group. This poses a major problem when using data analysis procedures such as repeated measure ANOVA in satisfying the equal sample size assumption. Unequal sample sizes correlate directly with the problem of homogeneity of covariance and equality of error variance, which is where the variance in one situation (or group) is comparable to the variance in the next situation. To remedy this problem, a random sample of all the eligible white subjects (N=90) was taken to try and equalize the two sample sizes. The total sample came to consist of fifty-eight whites and thirty-six blacks (see Table 13). A majority of the sample was female (78 percent) and were seniors (50 percent). Most of the sample was single (81 percent) and lived on campus or within 25 miles of the university (91 percent). The average age was 24 years with a low of 18 and a high of 50.
 
 

The questionnaire was a three page document requiring fifteen minutes to complete (see appendix A). It asked race-related questions within different geographical contexts. Each set of questions was related to a particular topic and differed by changing the degree of proximity (ranging from campus to national levels).

In this study, answers were coded on a seven point Likert scale. The scale contained a neutral response (four) and three points each of a positive or negative opinion. For example, many of the questions used answers that could range from strongly agree to strongly disagree. Three of the points (1,2 or 3) indicate some level of agreement while the other three points (5,6 or 7) indicate some level of disagreement with the statement. Unlike dichotomous categories (i.e. agree/disagree), continuous variables can measure the degree to which a respondent agrees or disagrees.

All the questions used were adapted from existing surveys. The questions were not modified from their original format. The surveys were:

1) People for the American Way's Democracy's Next Generation II: A Study of American Youth on Race

2) Washington Post/Kaiser Foundation/Harvard University's The Four Americas: Government and Social Policy through the Eyes of America's Multi-racial and Multi-ethnic Society

3) Z. Smith Reynolds Foundation's Racial Attitudes of North Carolinians

4) ABC News: Nightline poll

5) Southern Regional Educational Board's studies on "minority" college students (Standley, 1978; Abraham & Jacobs, 1990).

These questions were used because in the existing studies they revealed differences in opinion between whites and blacks. This study retests the questions to see if the differences reappear and if they are relevant after controlling for degree of proximity.

Some research indicates question order sometimes has an effect on subject responses by inducing a bias toward the first item mentioned (Carlson et al. 1995; Willits and Saltiel 1995). Question order is when one type of question or item always precedes or follows another question. This effect might appear when surveys ask comparison or contrast questions. To minimize this possible occurrence, two versions of the questionnaire were distributed. One version asked subjects to rate their perceptions sequentially, first from a broad or abstract degree of proximity (the U.S.) to a specific one (the campus) while the other version reversed the order. This method helps minimize the bias that might set in; which is commonly seen with repeated measure designed studies (for a more detailed explanation, see below).

A two-way ANOVA with repeated measure on one factor was used to analyze the data. It can also be referred to as a mixed design. One of the two factors was race (which statistically is called the between factor) and the other was degree of proximity (the repeated factor) or statistically referred to as the within factor. The advantage of using a repeated measure design rather than a general factorial design is one does not have to use as many subjects. A regular two-way factorial design dictates every subject belong to only one treatment condition. In a repeated measure design, each subject receives each treatment condition. In this project, the treatment conditions were different degrees of proximity. Each person rated both the specific and general degree of proximity instead of just one degree of proximity. For example, if one had two treatments, a researcher would need twice as many subjects when using a general factorial design versus a repeated measure design. Repeated measure designed studies are common in the pharmaceutical field where companies want to observe the effects of different drugs on a person. A main advantage of a repeated measure design is that each person serves as his/her own control. The amount of variability found person to person in general factorial designs are much higher than if the same person is used constantly. For example, in a drug experiment, the effects of a drug might be different upon different people. Possible effects are age, weight, health etc. With repeated measure, these factors are removed because the same person is being measured which removes the variability found between people.

The main disadvantage with repeated measure designs are order and/or carry-over effects. Order ("question order") effects were minimized by using two different versions of the questionnaire. Carry-over effects - when the previous question influences the answer to the next question - are harder to control and generally require counterbalancing (randomization) and/or spacing of the questions. This is mainly a medical or educational (testing) problem so it was not a main concern for this study. An example of a carry-over effect problem is when people test food or perfume. A delay is needed so one can remove the effects of the first treatment before administering the second treatment. One must be sure the smell of the first perfume is removed so it does not influence the rating of the second perfume. As a control, one might use different arms to spray the perfumes or have people rinse out their mouths with water when testing foods. Another example is with pre and post tests. One needs to have a control group so one can observe if just seeing the test the second time improves one's score or if actual learning is taking place.

The statistical procedure used GLM (general linear model) and tested if the responses were the same across racial groups and degrees of proximity as well as if there was an interaction effect between race and degree of proximity. Main and interaction effects test for statistically significant differences based on the relationship between the variables. To assess statistical significance, alpha was set at the .05 level (p .05). The effect size of each variable will also be noted, as measured by R², to see if the independent variable contributed any proportion to the total amount of variance in the dependant variable.

Table of Contents

Chpt. 1

Chpt. 2

Chpt. 3

Chpt. 4

Chpt. 5

Chpt. 6

Chpt. 7