Interpreting the results from the logistic regression is done by looking at the odds ratio rather than at
the coefficient, as it provides an intuitive interpretation: for example a constant odds ratio of 0.3
implies a 30% predicted chance and a variable odds-ratio of 1.1 implies an increase of 10% per 1
increase in the variable, meaning that a variable value of 3 results in a predicted chance of 39%. When
looking at the second model (including cart value as a significant covariate), it shows that the
predicted chance of completing an order increases with 0.4% per Euro of cart value on the constant,
which is relatively large given the average order value of 38.41 Euros (implying 9.2% base point average increase in predicted chance of purchase). Time of day and weekday versus weekend does not
significantly influence the predicted chance of completing a purchase.
Taking into account all models with all covariates, there were no significant influences of the cart
page designs on conversion. The approximate 3% to 4% increase over the 60% to 70% baseline
predicted chance of completing a purchase is highly insignificant with p-values around 0.72. The
stronger negative influence of the cross-selling enabled cart page design of approximately 10% over
the 60% to 70% baseline predicted chance of completing purchase is insignificant at the p-value of
0.27. More data is needed to investigate the effects of the cart page designs and more longitudinal
data is needed to get more insights into the effects of the covariates.