Hint: The “margins” and “marginsplot” commands in Stata are quite adept at this. 24Within the table, I should also see clear headings and variable names. This is particularly true of regression results or other depictions of quantitative analysis. Often, students will use variable names in their statistical software that are not immediately self-descriptive. For instance, a student working on a model of how spending a ects election outcomes may obtain a dataset with a spending variable named something like “camfin.” It is fine to use this variable name in your analysis, but before moving it into the paper, you must convert this variable name back to plain English. Remember, while you may have spent hours with these data, and undoubtedly know them inside and out, the reader does not have the same 12
luxury. Thus, you would rename “camfin” to something like “Spending by campaign, in 2010 dollars.” With that name, as a reader I know exactly what the result reflects.Similarly, figures should be clearly labeled not with variable names, but with the concepts they measure. Figures should be appropriately sized, clearly titled, and designed in such a fashion that they clearly depict the desired pattern.8.3 Tests and SignificanceWhile it is conventional in many political science outlets to show significance tests at a number of levels (– = .05, – = .01, etc.), doing so in my view does not reflect good statistical practice. A result with a p-value of .01 is not “more significant,” as statistical significance is a dichotomous condition, the level of which is pre-determined by the analyst. Almost always, it will be acceptable to seek 95% confidence in the result (– = .05). Your paper should reflect this decision, with “statistical significance” determined by your pre-established threshold. Tables containing statistical results should also have some sort of fineprint in which the student explains what sort of tests and significance thresholds were used.Based on his suggestions, I suggest that before arriving at DV, I measure the respondent support by question and condition through datacracker.com. You can see the figures in the second email sent that last night. These are just recommendations-- do with it as you please.Leadership: Based on my hypothesis, I expected that women will be rated lower in leadership.13
Voting: Respondents’ self-reported voting decisions according to the candidates’ gender To translate the findings above into a more comprehensive understanding of the real-life voting decisions these respondents may make, Table (?) provides a broad assessment of their voting decisions by condition. 15
One unexpected finding emerged. The participants, who were randomly assigned to Condition 2, were more likely to respond with “strongly agree” to the question: “The candidate’s gender doesn’t matter as long as the candidate is qualified.” (see briefly updated front-end for support on this).