Latent Regression Analysis Based on the Rating Scale Model

This paper sets out a latent structure model built on top of the rating scale model, together with a procedure for estimating its parameters. The guiding idea is to place a regression directly on the mean values of the latent population densities, so that shifts in those means can be traced back to measurable covariates.

To see the approach in action, Erling B. Andersen applies it to a group of housing standard questions drawn from the Danish Welfare Study. Other variables collected in that same study are brought in as explanatory variables, which lets the housing responses be read against respondents' wider circumstances.

A test is proposed as part of the method. Its purpose is to determine which of the explanatory variables account for a meaningful share of the variation in the housing index that the housing standard questions produce, separating the covariates that matter from those that add little.