You want to estimate the impact of a feature launch when you could not randomize users. The feature was rolled out first to a treated group and not to a comparable untreated group. In the 4 weeks before launch, the treated group had 18,400 active users and averaged 2.10 weekly creations per user, while the untreated group had 21,600 active users and averaged 1.95. In the 4 weeks after launch, the treated group averaged 2.42 and the untreated group averaged 2.08. The standard error of the estimated difference-in-differences effect from a regression is 0.041, and you are testing at the 5% level.
How would you estimate the causal effect with a quasi-experimental design here, test whether the effect is statistically significant, and explain what assumptions you would need for the estimate to be credible?