Assess match quality

[jo-fleck]

Possible approach: Compute the mean and median of all the matching variables for the k (=10) ACS households used to match to a given ASEC household. Then, for each ASEC household, compute the absolute difference in each of the matching variables using the mean and median of the k ACS households.

[jiaxitan]

I pushed the new version that computes, save, and plots match quality variables.

• I'm not sure what the best way to determine the match quality, so I wrote some code at the end of the main file to plot the distributions of the absolute differences. Maybe you can take a look and then we can discuss what to do next?
• From the plots, I think the KNN package tries to match levels of variables, regardless of how big the values of the variable is. For instance, observed from the distribution plots, variation in gross income is way larger than other coded variables, but KNN tries to match it as closely as other coded variables like race. In other words, I think it intrinsically put higher weights on variables with higher levels.