5 Econometric Estimation and Test Results
5.1 Demand Estimation Results
Using the data described in section 2.2, we have constructed observations of the households
choices of bottles of water over 13 periods of 4 weeks in 2006 using each purchase and an arbi-
trary rule consisting in assuming that the household product is the one purchased in the largest
quantity during each period. Doing such aggregation is however not essential for the results found.
Using other time periods or drawing at random a purchase event for each period did not change
significantly the results. The household purchase data finally allows to construct a sample of 2
836 households present over the whole 13 periods that is 36 868 observations. We have removed
households not present in the survey for some periods in 2006 in order to obtain a balancer panel
data set and also removed observations for which missing values exist in some variables. On this
sample, we estimated the demand model presented in section 4, as well as a standard multinomial
logit model. The estimates6 of the random-coefficients logit model and the simple multinomial logit
are in Table 2.
Coefficients |
Multinomial |
Random Coefficients |
(Std. error) |
(1) |
(2) |
Price (—a) |
-18.76 (0.003) |
-20.33 (0.004) |
Price (σα) |
6.42 (0.002) | |
Mineral water (J) |
1.28 (0.001) |
3.48 (0.001) |
Mineral water (σs ) |
3.83 (0.001) | |
Control η,jt (τ) |
17.06 (0.004) |
15.85 (0.005) |
Brand 1 |
3.00 (0.007) |
3.20 (0.001) |
Brand 2 |
5.08 (0.001) |
5.48 (0.001) |
Brand 3 |
1.86 (0.001) |
1.99 (0.001) |
Brand 4 |
0.97 (0.001) |
1.28 (0.001) |
Brand 5 |
2.25 (0.001) |
2.81 (0.001) |
Brand 6 |
0.88 (0.000) |
0.69 (0.001) |
Retailer 1 |
0.15 (0.000) |
0.36 (0.000) |
Retailer 2 |
0.69 (0.000) |
0.92 (0.000) |
Retailer 3 |
0.02 (0.001) |
0.25 (0.001) |
Retailer 4 |
0.45 (0.000) |
0.62 (0.000) |
Retailer 5 |
0.90 (0.000) |
0.11 (0.000) |
Retailer 6 |
-0.17 (0.001) |
0.03 (0.001) |
Table 2 : Estimation Results of Demand Models |
6Remark that we cannot provide the names of brand and retailer chains because of a confidentiality agreement
with TNS World Panel who provided us the data.
27