2. Propensity-weighted scores for causal assessment methods Random sampling from the overall user group, divided into two groups, experimental group: users who participated in the activity; control group: users who did not participate in the activity
Brought into the binary logistic regression model for iteration, calculated the propensity score P, and calculated the weight coefficient ws number list according to P W is used to balance the population distribution of the control group to ensure that the population distribution of the control group and the experimental group is basically the same. The detailed principle is as follows.
The propensity score is the probability value of subject i receiving treatment (T=1), conditioned on a set of covariates (X). The most commonly used calculation of this probability value is the logistic regression model, and models such as random forest and neural network can also be used. With similar scores, the distributions of treatment and control baseline data should be balanced.