Raking example

This example describes the raking procedure calculated by a poisson regression model. Raking adjusts the marginal totals of a table to match known totals. Sample survey analysts use this procedure to compute sample weights to adjust the marginal totals of a table to census totals. Here, we reproduce the schooling and attitude towards abortion example from the book from Agresti, Categorical data analysis. The table gives the relation of the amount of schooling to attitudes towards abortion. The table is adjusted in such a way that the rows and columns of the table sum to 100. The response is a column of pseudo data with values 100/3. An independence model with schooling and attitude is fitted to these pseudo data, where the logarithm of the observed data is included as an offset. The logarithm can be computed from the data menu, where you click transform. The estimated table is given in the output as the predicted values from the model. This estimated table has the same odds ratio structure as the observed table, although the marginal totals are different. This standardization procedure is used to compare different tables with varying marginal totals.

Summary: Steps in the analysis

• Calculate the logarithm of the observed frequencies,do this by selecting transform from the data menu, and then select ln from the list.
• Select Poisson regression from the statistics menu
• Select the pseudo data as the response
• Select attitude and schooling as the explanatory variables
• Click options and select the logarithm of the observed frequencies as an offset
• Click done, your done

The output is given here, note that the plots, parameter estimates and deviance table have no interpretation in this example because this model is fitted to pseudo data.