No, that's not correct. That's one possible factor but it isn't the only one. For example the roads with a limit over 40 stat you posted is probably irrelevant as the number of 30mph roads now dropped to 20 where there was a 50+ alternative that no one used is probably small.What you are looking for, is a significant reduction that is an outlier from the trend and then a check of "other" roads to ensure that there isn't a corresponding increase - suggesting avoidance rather than improvement. Then you are looking for the same exposure to risk. i.e. not a change in vehicle use etc.
Its also equally possible that roads were just more dangerous (hence the greater 40mph+ accidents) but the 20mph zones were even better.
But you can't cut this data too much, there aren't enough deaths to exclude noise if you start drilling into tinier and tinier subsets.
