Consideration data collected before making a choice can sometimes be different from the actual choice made. Different methods are used to handle this inconsistency. Researchers found that when attributes have a similar effect in both the consideration and choice phases, a mixed logit model works well. However, if attributes have different effects, two-stage models perform better. Stated consideration data doesn't significantly improve recovery of preference parameters compared to models that only use choice data. The study used mobile phone preferences to illustrate these findings.