The article discusses how using non-random samples can lead to biased estimates in behavioral studies. The researchers propose a two-stage estimator to correct for this bias, allowing analysts to use simple regression methods for more accurate results. The estimator's distribution is also determined for further analysis.