Post questions for Paul and Eva about the survey here
If you have questions about what sorts of items are on the survey, post them here.
posted by Paul Gronke at 11/06/2006 10:53:00 PM
Monday, November 06, 2006
Previous Posts
9 Comments:
Hi Paul & Eva,
I've been unable to find a variable (military service and/or family member military service) in the 2000 study to complement my work on the 2004 study. It might be that I'm blind or incompetent, but neither the Demographics sections (nor the rest of the dataset) seem to contain such a variable. Plese let me know if you can find one.
Thanks,
Will Brown
Ok, I agree. I don't see military service in the 2000 study. Wierd.
__Will McNitt__
I have been trying to run the immigration, Hispanic descent, and turnout data from the cumulative datafile. I ran a series of regressions, the first of which looked at the effect of being Hispanic on turnout. I am concerned because the T-stat was 10.5 or so. Furthermore, even after controlling for party it was still 8.758. Those values seem far too significant, but perhaps they are correct. Any suggestions? My dependent (turnout) variable is v702(r: 1=1; 0=2). My independent variable for Hispanic descent is v107(r: 1=1-4; 0=7). The independent for immigration stance is v879a(r: 1=1; 0 = 3,5). The independent for party ID is v303(r: 1=1; 2=3; 3=2).
Will,
can you send me the output or post it here? send as a PDF please. That statistic sounds just fine.
Did you run the WHOLE cum file? Did you include a variable for "time"?
I have two regression models, each with four dummy vars and turnout as my dependent. If the R-Squared values are similar, can I conclude that the two models have a similar strength effect on turnout? Would it be better to just run all eight in one regression and try to identify which set of four explains more of the correlation?
Also, I am trying to create a new variable with inputs are not all binomial and some of them need to be flipped, but I can't get recoded inputs (r: 0=5)or dummy (d: 1) to give me a distribution for the single variable. How do you recode when making new variables?
I don't think what I have now really means anything:
trustelection=(count(v5202, v5204, v5243, v5247(1)))/4
devin,
Ok. Some answers. First, you have a model with just dummy variables in the regression, right? So yes, if the two models have similar R2s, you can say that the models have similar explanatory power. But the power of regression is such that there is really no reason not to include all 8 in one regression.
On the recoding, the final recode you have essentially counts the number of "1's" in those variables, then divides by 5. Make sure the "1" means the same thing for all, though. Is it "up" or "Trust outcome" for all variables?
On the previous recode, you ahve to use ALL options in the recode:
VAR(r:0=5;1=1)
the other one:
VAR(d:1) ought to work.
I was just going over the sample paper, and on the 5th page there are some stats i don't understand. You multiply (6*4.9) the 4.9 is the B for one of your variables, but I don't understand the 6.
I was also wondering about controls. I know we should include them in our regression, but what are they for? You included age in the sample, and in your analysis you concluded that there was no effect. Was this the expected result, therefore confirming your hypothesis and the accuracy of your regression?
Thanks
McGarry
I was just going over the sample paper, and on the 5th page there are some stats i don't understand. You multiply (6*4.9) the 4.9 is the B for one of your variables, but I don't understand the 6.
I was also wondering about controls. I know we should include them in our regression, but what are they for? You included age in the sample, and in your analysis you concluded that there was no effect. Was this the expected result, therefore confirming your hypothesis and the accuracy of your regression?
Thanks
McGarry
Natalie,
Party ID runs from 0-6. So if you move it six points, from the strongest democrat to the strongest republican, you move six points (0-1, 1-2, 2-3, 3-4, 4-5, 5-6).
Each of these are a one unit movement, so 6 x 4.9 = total movement.
Controls are needed in a regression because you cannot be sure that some of your independent variables (such as media use) are not actually measuring something like education.
So it's generally good to control for obvious things that are related to your DV, such as income and education.
It's not necessary in order to do this paper, however.
Post a Comment
<< Home