Several studies have been conducted on the relationship between smoking and depressive symptoms using observational study data. This manuscript examines if smoking has a causal effect on depressive symptoms using the 8th Korea national health and nutrition examination survey data. We conduct propensity score matching to reduce the self-selection bias of the observational study through balancing covariates to estimate the average treatment effect for the treatment group. In the propensity score matching process, this paper focuses on two major cautions. First, if the outcome variable is binary, the marginal effect may not coincide with the conditional effect. Thus, we use G-computation to estimate the marginal effect, robustly. Second, in the complex survey sampling data, we consider complex survey design when the target population is entire, not the group of respondents. The result shows that smoking can lead to an increase in depressive symptoms based on the causal inference approach, and it is the same between the entire population and the group of survey respondents.