I'm not buying some cop white crime conspiracy. Sometimes it's something simple as when you ask the question or who you are asking.
Suppose you wanted to do research on what scares 6 year old children. You could send someone to ask 1000 6yr old what scares them. Then you send someone to ask 1000 adults what scared them when they were 6yrs old. You would have two entirely different sets of data. The 6yr old one would be all over the map, dogs, other kids, elmo, monsters, just random shit. The adult data would be more consistent but filled with generalities. Which is more accurate? Do you combine both data sets? What if you were to ask parents of 6yr olds what scares their children? Again, you would have completely different answers then the first two data sets.
This is the same as many crime stats. Do you ask the victim? The criminal? The arresting officer? Go by what the perpetrator was charged with? Or go by the crimes they were ultimately convicted of? Again, you would have different sets of data. All in some ways would be right. If the collection of data is flawed or not uniform the data reported is not accurate.