Nov 19, 2016

These findings are less than exciting:

Comparing across the panels of Figure 4 shows the decay in the effect of Treatment C over time. Although the effect remains statistically significant, the coefficient decreases steadily. In Panel A, the point estimate of -.27 indicates that the daily rate of the use of the word “_” decreased by .27 more among subjects in Treatment condition C than among subjects in the control condition. This average treatment effect for condition C decreased in magnitude to -.17 in Panel B and -.11 in Panel C. I collected data for 2 months, but these results are not shown because none of the treatments are significant.

...

Encouragingly, these effects persisted over time, for the first month under study, although not for two months. Also, the effect was significant at p < .05 in the two week time period, but it was only significant at p < .10 in the one week and one month time periods.

Wow! The subjects cut back on their slurs for the next month (and what is the practical difference in .27 fewer daily slurs compared to the control group?)....and then they started doing it again.

http://kmunger.github.io/pdfs/Twitter_harassment_final.pdf

Can't help but wonder how you get a Washington Post article out of that. It's a good graduate-level semester paper. What connections does this guy have?