This paper extends the concept of inverse probability weighted (IPW) M-estimation to a broader class of missing data schemes. The author allows selection probabilities to depend on unobserved variables, extending previous work that assumed selection variables were always observed. The paper demonstrates that estimating selection probabilities generally leads to a more efficient weighted estimator compared to using known probabilities. This result is particularly useful in cases where the selection mechanism is complex, such as variable probability sampling, censored response variables, and unobservability due to censoring of a second variable. The paper also discusses the properties of the IPW M-estimator when the selection probability model is possibly misspecified, providing insights into robust estimation of average treatment effects (ATEs). Additionally, the paper covers the case of exogenous selection, where the selection probability model is correctly specified, and provides conditions under which the weighted estimator is consistent. The paper concludes with a discussion on when to use weighted versus unweighted estimators, highlighting scenarios where weighting can be beneficial or detrimental.This paper extends the concept of inverse probability weighted (IPW) M-estimation to a broader class of missing data schemes. The author allows selection probabilities to depend on unobserved variables, extending previous work that assumed selection variables were always observed. The paper demonstrates that estimating selection probabilities generally leads to a more efficient weighted estimator compared to using known probabilities. This result is particularly useful in cases where the selection mechanism is complex, such as variable probability sampling, censored response variables, and unobservability due to censoring of a second variable. The paper also discusses the properties of the IPW M-estimator when the selection probability model is possibly misspecified, providing insights into robust estimation of average treatment effects (ATEs). Additionally, the paper covers the case of exogenous selection, where the selection probability model is correctly specified, and provides conditions under which the weighted estimator is consistent. The paper concludes with a discussion on when to use weighted versus unweighted estimators, highlighting scenarios where weighting can be beneficial or detrimental.