View article

The missing data problem has long been an issue for data analysis of all kinds, and structural equation modeling (SEM) was, in the early days, not exempt from such problems. The main problem, as was true of virtually all statistical procedures of the time, was that statistical algorithms were devised for complete data, that is, perfectly rectangular data sets with no missing values. In those early days, new statistical procedures first focused on well-behaved data (eg, normal distributions, no missing values), and only later began to address problem data (eg, non-normal data, missing values). An important year in the history of missing data analysis was 1987. Although important missing data work was done prior to that (eg, Dempster, Laird, & Rubin, 1977), what happened in 1987 was a kind of a missing data revolution. This year was the beginning of the end of the reign of complete-cases analysis in scientific research …