Descriptive Section

[jrspringman]

We need to show some descriptive insights we gleaned from the data. See the RAI paper for an example of how I'd approach this.

• [ ] Map 1: Map showing variation in the amount of cs activity (total non-999 articles/ total articles) over time. See the first map in the RAI paper as an example. There's code for those maps in the RAI paper's qmd, and you can find map code for civic space events here tracker/forecast_performance_final.R. I've already dealt with the annoying labeling stuff, so you won't have to.
• [ ] Map 2: What are the most reported-on civic space events for each country? How has this changed over time within countries?
• [ ] Figure of line plots with shock detection for some event. We could even do the months from the gpt validation section. But thinking of something really important that impacted a handful of countries would be perfect. I.e. my plot in the RAI paper showing what happens to Russian influence around the period that Sahel countries kick-out France.
• [ ] Figure showing readers what the data looks like on a month-to-month basis for some interesting recent-historical event. Maybe we could show the dive in civic activity around COVID? We could just show some line plots like the ones from the COVID report (posted on the mlp website) and then maybe describe a negative correlation between shocks to the SoE measure and and shocks to other categories. I'm not sure COVID is the right move, but it's worth considering. Here's a dropbox folder with the code from my pre-git era.

Another quick descriptive analysis would be to look at the PCA results. We generally see very similar factor loadings across countries (i.e. SoE, violence loads negative; corruption, activism load positively). Finding a quick way to visualize this could be of interest to folks and could be summarized in 1 paragraph. This could also help with a potential line of attack: that the normalization pits each event-type against each other, so a big increase in one necessarily drags down the others. If we show that the correlation across categories over time matches a sensible pattern, that might help?

[dmoratz]

Hey, which git is tracker/forecast_performance_final.R located in? I don't see it in either rai-activity-intro or ML4P-Civic-Space-Forecasting?