The Impact of Cognitive versus Affective Aspects on Consumer Usage of Financial Service Delivery Channels

model would have to be estimated as a multinomial logit model, notwithstanding the fact that
the dependent variable is clearly ordinal.

Complete reliance in the conclusions of the Wald test can be risky (Hauck and
Donner, 1977; Jennings, 1986; Menard, 1995). For example, the Wald statistic can produce
erroneous conclusions, such as when very large logit coefficients lead to inflated standard
errors and to small Wald values (i.e., leading to Type II errors in which the effect of the
variable is deemed to be not significant when it is). Therefore, some authors suggest either to
complement the Wald test with the likelihood ratio test or even to use only the latter (Hosmer
and Lemeshow, 1989; Agresti, 1990; Agresti, 1996). This led to that, although some predictor
Wald coefficients were not individually significant, their joint test of significance with other
predictors of the model (either cognitive or affective depending on the individual variable in
question) and the likelihood ratio test were used to decide whether to keep or discard that
individual variable from the model.

According to Hosmer and Lemeshow (1989), “models should not be based entirely on
tests of statistical significance. (...) Successful modeling of a complex data set is part science,
part statistical methods, and part experience and common sense. (.) The objective is to have
the most parsimonious model that best describes the data”. Thus, after fitting the model and
estimating the coefficients, the contribution of each variable for the prediction of the
dependent variable should be assessed. This stage of variable testing was followed and so, the
models presented are the final models for each delivery channel.

Debit card channel

Table 2: Ordinal logistic regression of debit card’s usage

18

<<   <   16   17   18   19   20   21   22   >   >>