If the figures really do not change, where is the analysis about why it was found that there was a reduction in the rate when it was first put in ?
This is a very difficult analysis to perform. Basically you are looking to potential confounding factors, but one you identify them proving which (if any) are responsible for the paradox in effectively impossible.
Most likely the reason is that external environment has changed. Vehicles are much safer and have better lights than was the norm in the 1980s, so it would be very unlikely that switching lights off would show the reversal of any original results.
It is also likely that the reductions seen when lighting was introduced, were not entirely due to the new lighting but included a contribution from external confounding factors like regression to the mean, and more importantly the increasingly rapid safety improvements in vehicles and highways that occurred at the time.
A more remote possibility is that the loss of lighting has in fact increased casualties, but that this has been offset by some other factor reducing casualties at the same time.
With road safety typically involving interventions on multiple fronts at the same time, it is very difficult to separate out any individual intervention and accurately assess it's impact - which is why you see so many valid challenges to the claims made to support some of the, shall we say, less popular interventions.