In files that are not longitudinally edited, the user will notice illogical changes for some cases over time. In the 2004 and 2008 SIPP panels, data for each wave were edited independently. Changes in reporting or imputation for the same case in different waves were not reconciled as they would be in a longitudinally edited file. So, for example, users may notice that a small percentage of cases change age in illogical ways, with a significant jump in age between waves, or even by becoming younger over time. Similarly, some cases go from being married to being never married.

Source of the inconsistencies: The main reason for these inconsistencies is that each wave is edited independently of the others. Several other factors also affect these inconsistencies. In the SIPP 2004 and 2008 panels, there was a question asking for authorization to pass information forward.  This question was designed to provide an opportunity for respondents to opt out of having their data carried into the next wave, and it required the questions to be re-asked.  This increased the potential for getting inconsistent reported data.  

One additional source of longitudinal inconsistency is when people leave the survey and return later. Sometimes the interviewers re-add a person to the roster and it creates significant havoc. It can appear that people’s biological parents change, and users cannot be certain which is the same person from a prior wave.  This is very difficult to correct. Users may see this type of issue in all of the SIPP files.

How to address these inconsistencies in analysis: Some analysts choose the last respondent-reported value for sex, race and birth month/year and use those as longitudinal values.  It may also be advisable to calculate the difference  in age for the same person across the entire panel to determine if there was a month/wave where there would be a major concern--the person is a child in one wave and an adult in the other, or where there was a difference greater than some threshold such as 2 years.  Other types of inconsistencies, such as transitions into never married from an ever married state, or illogical changes in  educational attainment, could be addressed in a similar way. Essentially, the analyst must make an arbitrary decision about which point estimate is “correct” and then reconcile the rest of the values across the longitudinal panel.