However, the simple correlation of REFRIG and HXP is .90 and REFRIG may not have a separate role in the model.
Figure 1 shows the relationship between price elasticity, the percentage of family income spent on housing (HXP), rural/urban location (URB) and the proportion of customers that are minimum bill customers (MBILL).
For alternative values of HXP, rows 1 and 2 estimate pure marginal price elasticities and rows 11 and 12 estimate pure intra-marginal elasticities.
REFRIG was statistically insignificant, was highly correlated with HXP and may not have a role separate from HXP and other variables in the model.
Finally, an equation explaining HXP, the share of income going to housing, was estimated with earlier data and with electricity price entered.(14) All variables were important with the exception of household income.
Re-estimated price elasticities, reported in Figure 3, indicate only marginal differences from those in Figure 2 where HXP is assumed fixed.
First, while price indices individually and as a group were important in the equation explaining HXP, they were not important when entered separately from HXP in the demand model.
In fact the sample correlation between refrigerator saturation (REFRIG) and the percentage of income devoted to housing (HXP) is .90.
For example and assuming no intra-marginal effects (MBILL = 0%), the short-run price elasticity lies in the range from -.182 to -.2 when HXP is as large as 12.6%.
Here, price indices are important both individually and as a group in explaining HXP, but when entered with HXP in the demand model have limited significance.
REFRIG is highly correlated with HXP (% of income allocated to housing costs) and, a priori, might not have an independent role in the model.