**
**

would use the menu commands to recode the variable, the logic is the same.]

**Q.** I finished reading the
chapter on **crosstabs** tonight, and I realized that in my
research, my dependent variable "Do you support/oppose death
penalty?" is a dichotomy...since it is neither interval nor ordinal,
can I really not use multiple regression ? Could I just convert it
into a dummy variable and use it that way?

A.This week in lecture, I said that using a dichotomous variable as a dependent variable in regression raised certain problems. Although the coefficients in the equation are unbiased estimates of the population parameters when used to explain a dependent variable, they are also inefficient and unstable. This means that one cannot test them adequately for significance. You could employ other techniques (that we haven't covered) to do this analysis, but those are beyond us now.

**Q.** My problem is that I really
feel like I have a great grasp on regression analysis, and I would be
more interested in doing that. We haven't
even gone over **crosstabs** in class, and it doesn't seem as
interesting to me when we have multiple independent
variables.

A.I'm glad that you feel this way, for understanding regression analysis is very important.crosstabsis a procedure to use mainly when the regression alternative is unavailable.

**Q. **What would your suggestions
be? Should I find a different dependent variable, one that is
continuous? Can I stick to death penalty and do regression ? Or must
I do **crosstabs **?

A.The first thing to do is to runFrequenciesto examine the variable codes and see whether you can do anything with the variable. Here is the table from my run of vote96:

V961197 96PO: Does R favor/oppose the death penalty? Valid Cum Value Label Value Frequency Percent Percent Percent 1. Favor 1 1168 68.1 78.7 78.7 5. Oppose 5 316 18.4 21.3 100.0 0. Inap, no post IW 0 180 10.5 Missing 8. DK 8 40 2.3 Missing 9. NA 9 10 .6 Missing ------- ------- ------- Total 1714 100.0 100.0

You will want to create a new variable, let's call it KILLEM, from V961197. Consider the logic of this command applied to the file vote96.

2 missing values v961197 (99). ^ Replaces all previous missing values codes. 3 compute killem = v961197. ^ Duplicates v961197 as the new variable, KILLEM. 4 recode killem (5=-1) (8= 0). ^ Recodes KILLEM into a three category ordinal scale. 5 missing values killem (9). ^ Establishes as missing data only the "No Answer" code (9). 6 variable label killem 'Favor execution (V961197) made ordinal'. ^ Labels the new variable, KILLEM (not really necessary; cosmetic) 7 value labels killem -1 'Oppose execution' 0 'Undecided' 1 'Favor execution'. ^ Labels the values in KILLEM (not really necessary) 8 crosstabs /tables = v961197 by killem /missing=include / cells=count. ^ Checks the coding validity for KILLEM. The subcommand "missing=include" forces CROSSTABS to show all data codes.

V961197 96PO: Does R favor/oppose the death pen by KILLEM KILLEM Count | |Oppose e Undecide Favor ex |xecution d ecution Row | -1.00| .00| 1.00| 9.00| Total V961197 --------+--------+--------+--------+--------+ 0 | | 180 | | | 180 0. Inap, no post | | | | | 10.5 +--------+--------+--------+--------+ 1 | | | 1168 | | 1168 1. Favor | | | | | 68.1 +--------+--------+--------+--------+ 5 | 316 | | | | 316 5. Oppose | | | | | 18.4 +--------+--------+--------+--------+ 8 | | 40 | | | 40 8. DK | | | | | 2.3 +--------+--------+--------+--------+ 9 | | | | 10 | 10 9. NA | | | | | .6 +--------+--------+--------+--------+ Column 316 220 1168 10 1714 Total 18.4 12.8 68.1 .6 100.0

Now you have a new variable, KILLEM, with this distribution:

KILLEM Favor execution (V961197) made ordinal Valid Cum Value Label Value Frequency Percent Percent Percent Oppose execution -1.00 316 18.4 18.5 18.5 Undecided .00 220 12.8 12.9 31.5 Favor execution 1.00 1168 68.1 68.5 100.0 9.00 10 .6 Missing ------- ------- ------- Total 1714 100.0 100.0

Although this is not an ideal measure of attitudes toward the death penalty, it will do for our class. It will enable you to use multiple regression.