# How to show covariates are balanced after IPTW

#### felisleo987

##### New Member
I am attempting to perform an inverse probability of treatement weighting on observational data. I can perform the IPTW, however the problem I have is after obtaining the weights/balancing groups, how would i go about showing a baseline table of the new "pseudo population" to show covariates are balanced. For example lets say in the original baseline table i have 30% of diabetic patients in one group and 15% in the other group (p<0,05), but after the IPTW there is no difference between the groups (p>0.05), how would i show this via a baseline table for the pseudo-population( the percentages as in the original sample)?

I can calculate the weighted means after the IPTW, would those weighted means for binomial variables equal to the percentages or am i way of?

Thank you in advance if anyone is able to help me out.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
So are you planning to use strata or scores based on the IPTWs? I am imagining scores, but it wasn't apparent given your text. Also how may covariates do you have in the propensity model? Are you familiar with the positivity assumption and did you do any trimming at this point?

#### felisleo987

##### New Member
Sorry about not being clear in the first part. I am planing on using scores based on IPTWs. I have 12 covariates. I am familiar with the positivity assumption, and i have not yet performed any trimming.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Well, I believe the basic approach is to compare the IPTWs histograms for the two treatment groups or create a scatterplot of the values stratified by treatment group and overlay a boxplot. So you visually examine the scores based on treatment group. I don't have a source to support this, but I believe when you have so many covariates the scores are more of a blend across the confounders (some negative or positive confounders) thus examining at the covariate level - differences may still be visible since you are using a score based on multiple covariates in a model. If you had a particular variable of interest of extreme importance you could check it out, but I believe balance is examined at the score level. You could also hold an important variable out of the PS model and put it in the outcome model (I believe), if it is of great importance.

Let me know if you find resources that don't support this.