Factorial ANOVA and ANCOVA Output Application in Criminology
Introduction
A researcher was interested in finding the best way to educate elementary-age children in mathematics. In particular, she hypothesized that fifth-grade girls do better in small class sizes while boys excel in larger classes. The researcher performed a pilot study throughout the school district where classroom sizes were reduced into three groups as follows: small classrooms containing no more than 10 children, medium classrooms having between 11 and 19 children, and large classrooms having 20 or more children. This paper is a write-up of the exploratory data analysis for the variables used in the study using SPSS. In addition, the paper presents a factorial ANOVA of the dataset as well as a hypothetical ANCOVA output for predicting criminal behavior.
Exploratory Data Analysis
The mean math score for girls studying in a class of 10 or fewer students was 93.80 with a standard deviation of 3.938. The minimum math score for this group was 88.00 whereas the maximum score was 98.00. The mean math score for boys in a class of 10 or fewer students was 92.70 with a standard deviation of 3.434. The minimum math score for this group was 87.00 whereas the maximum score was 99.00 (Table 1). It is therefore evident that girls performed slightly better in math than boys when studying a class of 10 or fewer students.
In a class of 11 to 19 children, the mean math score for girls was 88.50 with a standard deviation of 3.979. In the same class size, the mean math score for boys was 89.70 with a standard deviation of 2.406. The minimum score for girls was 82.00 whereas the minimum score for boys was 86.00. The maximum score on the other hand was 95.00 for girls and 93.00 for boys (Table 2). The mean math scores for this group, therefore, indicate that math performance for boys was better than that of girls.
The mean math score for girls in a class of 20 or more students was 79.20 with a standard deviation of 4.184 whereas the mean math score for boys in the same class size was 91.20 with a standard deviation of 3.225. The highest math score for girls in this group was 86.00 whereas the highest score for boys was 98.00. The lowest score for girls in this group was 72.00 whereas the lowest score for boys was 87.00 (Table 3). These results indicate that the math performance of boys was better in a class of 20 or more students compared to the performance of girls in the same class size.
In summary, the above descriptive statistics indicate that the performance of boys in math becomes better as the class size increases whereas the performance of girls in math deteriorates as the class size increases.
Factorial ANOVA
A factorial ANOVA was conducted at 95% CI and a significance level of.05 to determine the presence of the main effect and interaction between gender and classroom size on math performance. It was established that gender had a main effect on math performance since the F value for gender was significant, F (1, 54) = 19.056, p =.001 (Table 5). The F value was large enough to indicate that big differences in math performance existed across the genders. It was not necessary to perform posthoc tests for gender since gender has less than three groups which is the minimum number of groups needed for post hoc analysis.
The analysis also indicated that there was a main effect of classroom size on math performance for both boys and girls. This is indicated by a large F value which is significant, F (2, 54) = 25.311, p =.001. On conducting a post hoc test for the same, the LSD test showed significant differences (p <.05) between performances of classes of 10 or fewer students and those of 11 to 19 and 20 or more students. The Games-Howell test also confirmed significant differences in math scores between classes of 10 or fewer students and 11-19 students and 20 or more students. However, the Games-Howell test indicated non-significant differences in math performance for between classes of 11-19 students and 20 or more students (p =.086) (Table 6).
The above factorial ANOVA test also indicated an interaction between gender and classroom size and the interaction was significant, F (2, 54) = 19.102, p =.001. The R squared for this model (interaction) was.666 (Table 5) indicating that both gender and classroom size contributed to 66.6 percent of performance in math.
From the descriptive statistics and the factorial ANOVA, it is evident that math performance for children is significantly affected by the classroom size. In particular, the performance of girls is better than that of boys in classrooms with fewer students. This is supported by the fact that the mean performance for girls was better than that of boys in classrooms of 10 or fewer students and it worsened as the class size increased. On the other hand, the performance of boys bettered as the classroom size increased. The existence of the main effect of gender and classroom size as well as the interaction between the two variables also confirms this. As a result, the researcher’s hypothesis that girls would perform better than boys in classrooms with fewer students is supported.
Applying Analytical Strategies to an Area of Research Interest
My area of research interest is to understand criminal behavior to curb crime, terrorist attacks to be specific, which may occur within or outside our borders. I would like to predict the possibility of an individual getting involved in a crime based on the individual’s background as well as personality. Childhood delinquency and parent-child relationship will be important factors to consider in determining the likelihood of engaging in criminal behaviors. It is hypothesized that a high number of childhood delinquency cases as well as poor child-parent relationships increase the likelihood of an individual getting involved in crime.
Based on my area of interest, age would be a good example of a predictor variable whereas appropriate dependent variables would be a frequency of involvement in crime, prior delinquency as well as a parent-child relationship. A Hypothetical ANCOVA indicating that the age of the individual as well as prior delinquency (covariate) significantly predicts the frequency of involvement in crime (dependent variable) was conducted. The mock output table supporting this relationship is provided below:
Table 7: Hypothetical ANCOVA Output Table for Predicting Frequency/Likelihood of being caught in Crime using Age and Prior Delinquency.
Table 7 indicates that age and prior delinquency are significant predictors of an individual being involved in crime, F (1, 1) = 176.333, p =.048. The interaction of age and the covariate (prior delinquency) is very strong in determining the possibility of an individual being involved in crime as indicated by a large F value which is significant at the level of.05. The R squared value is.997 indicating that 99.7 percent of the frequency/likelihood of being caught or involved in crime is a factor of age and prior delinquency of the individual.
Reference
Field, A P. (2009). Discovering statistics using SPSS, Third edition. San Diego, CA: Sage Publications Ltd.
Appendix
Table 1: Descriptive Statistics for Math Performance in Classrooms of 10 or Less Students.
Table 2: Descriptive Statistics for Math Performance in Classrooms of 11-19 Students.
Table 3: Descriptive Statistics for Math Performance in Classrooms of 20 or More Students.
Table 4: Between Subject Factors for Classroom Size and Gender.
Table 5: Between-Subjects Effects for Classroom and Gender.
Table 6: Post-hoc Tests for Classroom Size.