Basic questions for the presentation.

[jmarkgraf]

Hi Malte,

I just created a first rough outline of the presentation. A number of basic questions are still to be answered:

1. I created a Beamer presentation, which is academic standard, but since we are supposed to engage wider public we could also create an html (ioslides) presentation (what Christopher usually uses in class). Let me know what you prefer. Related to this: We could also create the presentation in Latex format. That would potentially solve a number of formatting issues.
2. Did you get in touch with the two guys for the presentation feedback? I don't mind to give them spontaneous feedback, but it might be worth to have at least a quick look at their presentation (and actually the topic) in advance. I have no clue what they are working on.
3. I will work on the design (avoid separate slides for level-1 headers; centering the image - would be grateful for suggestions!). Furthermore, I would draft a script, so that you know what I'm going to say. I will make sure that I'm not talking longer than 5 minutes.
4. Maybe you could add a few bullet points, so that I get an idea of what you intend to say and share the regression outputs, so that I can add my thoughts about the interpretation of it.

[mberneaud]

1. I think an Ioslides presentation is fine. As a matter of personal preference, I actually prefer them over the html slides, actually, as they are easier to skim over and work without hassle everywhere.
2. I got in touch with Johannes and Mujahedul for the presentation feedback. They've agreed to be our discussants and told me they would send me the link to their presentation repo as soon as they started. In the meantime, they sent me link to their repo for the third pair assignment.
3. I'll have a look at the centering of the image later on tonight. Maybe I come across a solution you haven't tried yet.
4. I'll do that once I revised my R code and have the regression outputs ready. A central part of the second part of the presentation will be to figure out a way to graphically show our results, as Christopher said he hated regression tables. My first thought is to have a simulation as shown in Lecture 9, either done manually or with the help of the simGLM package.

[jmarkgraf]

Thanks for your work on the presentation! Somehow the code doesn't run yet though for me. Could you have a quick look at the working directories? I think they need to be defined for the Beamer presentation as well as for the R file. We have the PDF, but Christopher might want to replicate what we've done.

A few comments on your slides (as I see them in the PDF): That all looks good to me. I would maybe put a bit less content on each slide.
For the regression table: Why do you report t values instead of SE? And how do you interpret the >1 coefficient for Sparkassen membership? There seems something off as our response is binary {0;1}, right? Or do I misinterpret something? Remember that Christopher recommended to only report 1-2 digits after the comma. For the regression statistics: Maybe rather report adj. Rsquared instead of Akaike. We only calculate one model, so Akaike is not very informative.
For the plotted model: This is interesting but hard to interpret I would say. I suppose, as we do not control for whether an incumbent lost her position or retired, this might explain this unexpected outcome. An alternative way for the presentation would be to show a bar plot with the descriptive comparison of re-election mean of mayors with a board seat compared to mayors without a board seat. This you could use as a motivation to run the regression because you presume that the difference is affected by many other factors too and a regression enables you to understand the effect of each of those factors.

Unfortunately I have really little time to review everything as I'm busy with the end of year report and I have guests over the weekend, but I will have a look again this evening! Hope my comments are helpful though.

[mberneaud]

In order for the presentation to render, you have to set the working directory to the root directory of the repository, which is done automatically if you clone the repo directly from within RStudio, hence why I never do it. Because not everyone follows this workflow, I added a line with 'setwd(~/Git/PresentationAssignment)' and a comment asking the user to change the directory accordingly.

In order for the repository to run, you need to copy your data sets into the data/ directory, hence why I created the directory and put the placeholder file inside. The readme stated that one needed to manually include the data into the repo, but that can be easy to miss. To prevent that from happening again, I included another note about that in the Rmd file and had the readme state explicitly that the repository does not contain the data necessary to reproduce the RMarkdown files.

As for the interpretation of the coefficients: the numbers reported are not beta coefficients, but odds ratios as they allow for easier interpretation of binary variables. The interpretation of them is as follows in the case of mayors: Mayors who are board members of Sparkassen have 23% higher odds of getting re-elected. Generally, odds ratios show the ratio of the odds between the reference category and the category reported, so 1.23 means that the odds are 23% for the category (board member-mayors) under scrutiny in relation to the reference category (non-board-member.mayors). Likewise, an odds ratio below 1 can interpreted as the category underscrutiny having lower odds of getting re-elected in comparison to the reference category by a percentage that equals 1-odds ratio. In the case of female mayors, the interpretation is as follows: Female mayors have around 24% lower odds of getting re-elected.

As for the test statistics: I changed stargazer to report two decimal points only now and also excluded the Akaine statistic, but adj. R² can't be reported by stargazer, as it is not calculated by the glm function and thus is not stored in the object storing the regression estimates. I changed the reported t-values for p-values to get some stars on there. However, because I had stargazer exponentiate the coefficient estimates to convert them into the odds ratios, the standard errors come out wrong as they are recalculated using the new coefficient estimates. To circumvent this, I have disabled the automatic re-calculation of the test statistics and reported the p-values from the underlying beta coefficients along with the exponentiated beta coefficients (the odds ratios).