Learning about independent and dependent variables

Connie Teng inquired:

Q. How do we know the details of say the "bush job approval" scale in the vote92 file. I know the minimum/maximum, but which direction is it? For instance, I can say that the more strongly a person feels about increased minority aid, the more dislike one feels for Bush during the preelection. However, how do I know which directions on these scales are positive or negative?

A. Just as you did at the beginning of the class, run Frequencies on the "bush job approval" scale to see the coding categories and the number of cases in each category. (Time and again, this research paper assignment proves its worth as the most important learning tool in my statistics class.)

Q. Secondly, why do only some of the iv's print out for multiple regression? Do only the ones with significant r's and high f tests get printed out?

A. Sorry, but I don't know for sure what you mean by "iv's". If you mean independent variables, only those with significant t values (not r's and f tests) are included IN the regression equation under the stepwise procedure. That's true.

Q. Thirdly, do we need to be doing different kinds of procedures besides multiple regression(ie plot)?

A. At the beginning of the lecture on measures of dispersion, I noted that statistical analysis typically seeks to explain variation in a chosen dependent variable. Therefore, it is important to know (and report) the pattern of variation in your chosen dependent variable. So everyone should run Frequencies on your dependent variable to obtain its relevant univariate statistics. On that same run, ask for histogram. Copy the graph from the computer screen and paste it into your paper to demonstrate the variation that you are trying to explain.

Once you have identified independent variables that seem to explain a good deal of your dependent variable, create a scattergram to determine the nature of the relationship. If you are adventuresome, you can also compute the predicted value of the dependent variable from your regression analysis and plot it against the observed value using scattergram. (Use the constant and b-coefficients from the final regression equation.) The resulting bivariate r should equal the multiple R from the regression equation. This procedure was used in Jackman's article, when he correlated actual coup scores with estimated coup scores.