I run logit model with a cross-sectional dataset of Indian individuals. I am using descriptive statistics of the same dataset to justify and interpret the estimations of logit model. However, I must report the descriptive statistics with sampling weights in the analysis. Using sampling weight creates big changes in the dataset and nullifies the findings of logit model. Rule says that I must use weight for reporting descriptive statistics. Another side, I cannot use weighted logit model as it gives spurious results by inflating the total observations from 1 million to 1.6 billion. I will be grateful if you suggest some solution.
As a general rule, you want to use sample weights if your analysis is supposed to be representative for the whole population. Whether the estimates please you should not be a factor. The fact that you have an Indian population of 1.6 billion (actual: 1.4 billion) suggests that either
- sample weights are specified incorrectly by the data supplies or
- you don't use them correctly
Either way, if descriptives are with sample weights, the regression should be with sample weights as well. The fact that you have 1.6 billion weighted observations is a natural consequence from this.
More on the use of sample weights: Solon et al (2015): What are we weighting for?