Each city contributes records for three different points in time. Dummy variables for 1980 and 1990 were created to indicate decade differences with 1970 as the reference category. The first model contains only the decade controls. In this model, 29 percent of the overall variation in family structure by city is explained. Cities in both 1980 and 1990 were more likely to have female-headed families than cities in 1970.The final step in this analysis is to examine what proportion of the differences in female-headed families is due to decade change and what part is due to compositional changes. The compositional changes minus decade changes are included in Model 2, while Model 3 contains both types of change. When compositional changes are added to decade changes a much larger percent of the city variation in family structure is explained. Compositional changes explain about 60 percent of the variation in family structure between cities. These compositional variables had the same basic relationships with female-headed families as they did in the previous decade analysis, especially in decade models for 1980 and 1990. Similar to Model 1, in the combined model the dummy variables for 1980 and 1990 remained significant indicating that regardless of the composition of the city, decade differences in family structure still remain. Cities in 1980 and 1990 were more likely to have female-headed families than cities in 1970. Because these decade variables remained significant, decade differences in composition of cities cannot be used to fully explain city differences in female-headed families. In other words, decade differences in female-headed families were independent of other contributors to the growth of female-headed families. Thus, there were compositional differences in cities that contributed to family structure changes. This indicates further support for the second hypothesis that employment dislocation was an important factor in the g...
