sixty-nine weeks, an error component model as developed by Fuller and Battese (1974) is
considered most appropriate for this study. The general form of this model is:
v
Yqr = ∑ Xqrsβ. + μqr q = 1.2..... n ; r = 1.2..-. T
s =1
where N is the number of cross-sections. and T is the length of the time series for each cross-
section.
Six cross-sections and 69 observations per cross-section are included in the specified model for
this study. Fourteen equations are specified and estimated using the time series cross-section
regression (TSCSREG) procedure in SAS. The equations and included variables are specified as
follows:
Qikt = f(pikt. pjkt.. pmkt.SDUMkt.TEXPkt.Qikt-1)
Where Qikt is total ounces of sub-category i for store k in week t; i = 1. .... 14; k = 1. .... 6; t = 1.
.... 69; Pikt is a weighted-average price of sub-category i for store k in week t; Pjkts represents
weighted-average prices for competing sub-categories for store k in week t; Pmkt is identical to
Pikt for lower-income stores 4. 5. and 6. but 0 for all other stores (it is intended to capture price
elasticity differences for higher and lower income stores); SDUMkt are zero-one dummy
variables intended to capture store differences; TEXPkt represents total expenditures on fruit and
vegetables for store k in week t (intended as a proxy for consumer income); and Qikt-1 is total
ounces of sub-category i purchased in store k during the previous week. Descriptive statistics for
dependent and independent variables are provided in Table 1.
Prices are determined by expressing each fruit or vegetable sale as a ratio of all fruit and
vegetables sales within a given sub-category. Specifically. weighted prices for sub-category i in
each time period is: