1. Two-Way Interaction Between Categorical and Continuous Variables

Data: K:\nd.edu\user22\yli\Public\593sp01\Data\ordgss77.dta
         Source: Long, J. Scott. 1997. Regression Models for Categorical and Limited Dependent Variables.
         Content: Data from 1977 General Social Survey on education and occupational prestige.

Task:
You are interested in the effects of age (AGE), education (ED), sex (MALE), and race (WHITE) on occupational prestige (PRST).  Previous studies suggest that age, as a proxy of labor market experience, and years of schooling have positive effects on occupational prestige.  In trying to make new contribution to this research area, you are interested in examining:

1) the effects of sex and race on occupational prestige;
2) whether the effect of education differ between men and women; and
3) whether the effect of education differ between whites and non-whites.

In Stata, use regression techniques to answer the questions above. Discuss your findings by

1) interpreting the b coefficients and their significance;
2) testing whether the interaction model is better than the model without the interaction term -- i.e. does the interaction term explains significantly more variation in the dependent variable?
3) graphing the interaction effects of race and education, and/or sex and education (whichever is significant) on occupational prestige.

Data: K:\nd.edu\user22\yli\Public\593sp01\Data\Agresti\crime.dta
        Source: Agresti and Finlay, 1997. Statistical Methods for the Social Sciences. Chapter 9.
        Content: Data from Statistical Abstract of the United States for the 50 states and the District of Columbia on crime rates and state demographic characteristics.

Task:
Analyze the determinants of crime rate. Drawing from past research, you hypothesize that percentage of white population (WHITE), percentage of the population living in the metropolitan areas (METRO) and percentage of families living below the poverty level (POVERTY) are important predictors of violent crime rate (VIOLENT, defined as "the number of murders, forcible rapes, robberies, and aggravated assaults per 100,000 people in the population"). You also suspect that there are interaction effects between metropolitan-ness and poverty level, which past research failed to address. You therefore estimate the following two models:

Model 1: violent = a + b1 * metro + b2 * poverty + b3 * white

Model 2: violent = a + b1 * metro + b2 * poverty + b3 * white + b4 * metro * poverty

Test and compare the two models. Discuss your findings by:

1) comparing and interpreting the b coefficients in the two models -- keep in mind of the meaningful ranges of each independent variable;
2) testing whether the interaction model is better than the model without the interaction term -- i.e. does the interaction term explains significantly more variation in the dependent variable?
3) graphing the interaction effects of metropolitan-ness and poverty level on crime rate;