The Adoption
of Electronic Banking: Evidence from Liechtenstein
Dependent variable
(4)
Electronic Banking Adoption
(5)
(6)
Distance to closest 0.0019*** 0.0049%** 0.0079***
branch (min) (0.0006) (0.0015) (0.0024)
[0.0019] [0.0019]
Age -0.0044*** -0.0114*** -0.0183***
(0.0001) (0.0004) (0.0006)
[-0.0044] [-0.0044]
Sex 0.0735*** 0.1894*** 0.3051***
(0.0055) (0.0143) (0.0230)
[0.0734] [0.0735]
Constant 0.632 0.335 0.544
R? 0.040
Pseudo R? 0.029 0.029
Observations 31’511 31’511 31’511
Method OLS Probit Logit
Robust standard errors in parentheses
Marginal effects for the probit and logit models in brackets
* p<0.10, ** p<0.05, *** p<0.01
Table 5.2. Regression models distance to closest branch (min).
Notes: The dependent variable in this regression is Electronic Banking Adoption. Column (4) shows a
model in which the distance to the closest branch in minutes operates as the explanatory variable and we
simultaneously control for a client’s age and sex. Column (5) and (6) replicate the model from column
(4) with a non-linear probit and logit model. For each variable we report the raw coefficients from the
regression, together with robust standard errors in parentheses. For the probit and logit regression models
we add the marginal effects in brackets. To give an indication about the model's goodness of fit, we report
McFadden's R? for the probit and logit models, as well as the R? known from OLS. Definitions of the
variables are provided in Appendix C.
Source: Own table, based on the LLB dataset.
effect of à unit change in some explanatory variable on the dependent variable is simply the
associated coefficient on the relevant explanatory variable. However, for logit and probit models
obtaining measures of the marginal effect is more complicated, due to the fact that these models
are non-linear. We, therefore, follow the best practices for interpreting these results presented in
Hoetker (2007). When z; is a continuous variable, such as age in our model, its partial effect on
Pr (y = 1|x) is obtained from the partial derivative
OPr(y=1|x) O9G(zx)
Oz; c x;
where
dz
is the probability density function associated with G and where G is a function taking on values
strictly between zero and one: 0 « G(z) « 1, for all real numbers z. Notice that 0 < G(x) « 1
ensures that the estimated response probabilities are strictly between zero and one, which, thus,
addresses the main worries we had about using LPM. Because the density function is non-negative,
the partial effect of x; will always have the same sign as £;. Since our coefficients in table 5.2
University of St.Gallen 25