Mike Shores did his analysis of political bias in teachers’ essay grading and had some questions:
I ran the regression and tried to go over the results with my advisor at my school (Dr. Allyson Weseley), but we have a few questions and want to make sure we’re analyzing the results correctly. I have attached in pdf version the results of the regression.
My first question is for x2 (the condition of the essay the teacher saw)…I coded it as +1/-1 as you suggested, but I was wondering if this was arbitrary or it is a standard practice. Also, what happens if you want to “dummy code” something with 3 or more conditions…do you use 1,2,3 etc.?
Secondly, what do you think is the best way of displaying the results of a regression? Would it involve a scatterplot or just a table with the regression coefficients and p values?
Lastly, I interpreted the results as follows, and I was wondering if you agreed with these interpretations.
When looking at the results from SPSS, “mrlibc” is the +1/-1 for the essay version (where +1 is the liberal essay and -1 is the conservative essay). “Political” is the self rating of 1-6 (1 being “extremely liberal” and 6 “extremely conservative”), and “combo” is the interaction variable.
Here is what I believed these results mean:
1) The version of the essay had almost a significant effect on the grading of the essays where teachers gave almost significantly higher grades to the liberal essay.2) The higher the teacher rated himself/herself on the 6 point scale, the more likely they were to give the essay a lower grade.
However…
When I initially analyzed my results with an ANOVA, I found that conservative teachers gave higher grades to the conservative essay while liberal teachers gave slightly higher grades to the liberal essay. I can tell from the regression output that the interaction between the two variables is significant, but I cannot tell exactly what the interaction is; can you explain what in the output shows me that?
Can anyone help?
From the SPSS report, I cannot really see whether the interaction is "significant" or not. I suggest you switch to R: more people would be able to help you then.
I would be happy to put in R form, and I downloaded R…but I have no idea/understanding of the program. Could you give me directions on what you would need shown in R and how to do it?
Thank you.
Mike Shores
I would be interested to know what output R is going to produce for this regression that SPSS did not or could not produce. And is it something that Stata also could not produce? It is not clear to me why a less user friendly program (i.e., R), is a better choice here.
A quick internet search brought me to a R tutorial on regression. The summary fit at the end is very useful. The data can be loaded into R using the read.table command, and there is a description of that at OnLamp.
The issue is that I have neither SPSS nor Stata (expensive, commercial), so I cannot help with that software. I can, however, help with R.