4.2.10 Model validation
Model validation is carried out to ensure that the results are not specific only to
the sample data and can be generalized to the population. Different methods can be used
for validating the model. Those methods are (1) split sample; (2) testing the model on a
new sample drawn from a general population and comparing the two models; and (3)
bootstrapping (Stevens 2002 and Hair et al 1998, Barlas 1989).
In this study, it was not possible to validate the research findings using the first
method. When tried to split the sample randomly into two halves, an estimation
sub-sample for calibration and the holdout sub-sample for validation, the results indicated
an unstable path coefficient estimate. Hence, splitting the sample into calibration and
holdout sub-sample for validation was not feasible.
Usage of method 2 (e.g., testing the model on a new sample) for validating the
model was also not feasible because of cost, time pressure, and non-availability of
respondents. Hence bootstrapping (method 3) was used for model validation.
Efron (1981) introduced the method of bootstrapping for model estimation. The
bootstrap approach is used in research when replication with split sample is not possible.
In case of bootstrapping, the original data are repeatedly sampled with replacement for
model estimation (Hair et al 1998).
According to Schumacker and Lomax (2004) the bootstrap method treats a
random sample of data as a substitute for the population and resamples the original
sample a specified number of times to generate a larger number of new samples, each a
random subset of the original sample, to generate sample bootstrap estimates and
standard errors. These bootstrap estimates and standard error are then averaged to obtain