Multiple Correlation/Regression Analysis



Here is the example provided by Gina. I have made some comments.

Multiple Correlation/Regression Analysis

• This is an example of a multiple correlation and regression analysis.

• Colwell, J. & Payne, J. (2000). Negative correlates of computer game play in adolescents. British Journal of Psychology, 91(3), 295- 311.

• The issue number should not be given here, since this journal is paginated by volume rather than by issue.

• The participants were 204 children (91 boys and 113 girls) at a comprehensive school in North London. Their ages ranged from 12-14 years old.

• The study analyses the hypothesis that playing computer games may be associated with increased social isolation, lowered self-esteem, and aggression among adolescents. The "dependent" variables were social isolation, self-esteem, and aggression, measured by questionnaires using Rosenberg's (1965) 7-item self-esteem scale, Dominick's (1984) 13-item aggression scale, and one question measuring amount of friends. These variables were correlated with game exposure.

• This was an observational study because nothing was manipulated.

• The correlations showed that social isolation and lowered self-esteem are not significantly correlated with exposure to computer games although aggressiveness is somewhat significant. A multiple linear regression analysis showed that sex and game exposure accounted for 10.5% of variance in predicting aggression.

• Author’s Summary…

Social isolation: The partial correlation between game exposure and number of

good friends does not reach significance (pr = -.10, p < .2).

Self-esteem: The partial correlation between game exposure and self-esteem,

is not significant (prr = -.05, p < .48).

Aggressiveness: The partial correlation between aggression and game exposure is significant (prr = .18, p < .03). Boys produced significantly higher total aggression scores than girls, t(176)= 3.83, p < .001, and there were sex differences in contributions to the correlation between total aggression scores and exposure to computer game play.

Multiple linear regression analysis: A standard multiple regression procedure was employed in order to ascertain the proportion of explained variance in aggression. The dependent variable was aggression and independent variables were game exposure, number of aggressive games, and sex. Frequency, duration, number of years of play, and total time spent playing games with aggressive content were not included as predictors due to problems with multicollinearity and singularity. Also the prime interest was in the effect of a 'total dose' of game play. Analyses of the data for boys and girls separately did not reach significance. The results for the whole sample are shown in Table 7.Two variables, sex and game exposure, each accounted for a proportion of the total of 10.5% explained variance, and both were significantly predictors of aggression.

Thought question: Does playing aggressive games cause increased aggressiveness, or is it that aggressive kids prefer games with aggressive content, or would a "third variable" explanation better fit these results?

The multicollinearity problem could have been resolved by conducting a principal components analysis on the measure of aggression and then using the orthogonal component from that analysis as the predictor variables. But that is a topic for next semester or the semester after than.

Thanks for a nicely presented example, Gina.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download