Traffic-calming efforts in Milwaukee have reduced speeding by an average of 27 percent.

At the 10 locations posted in the OP. Data is collected at the same time each year, and over a 48-hour period.

It's very easy to do, and since the number of vehicles recorded is often in the thousands, the trends would be very easy to see.

That sort of data actually is available these days, from brokers like Inrix that collect it from cellphone signals and traffic cameras and such.

Departments of Transportation trying to measure the impact of a specific project are still more likely to do it the semi-old-fashioned way, with traffic studies before and after, though.

You don't need to know every single instance of speeding. You just a good representative sample, and then you can plot your trendline from that. That is how statistics like that are measured.

You can't measure actual crimes committed. However, you can make a reasonable guess by factoring in your known unknowns. For example, assume we have a significant sample of reported crimes in a city. Let's pull the number 50/100k people out of our rear end for example purposes.

As you pointed out, we only have the data from the reports. The total number is known unknown. We could then look at a different data set, like a survey, that says something like 20% of people who are victims of the crime never report it, or 20% of people admit to doing it and not getting caught, whatever the case may be.

So, we can cross reference those two statistics to estimate that the "real" rate of that crime is closer to 60/100k people. Even though neither study can predict that number individually.