When analyzing a problem, it is always useful to ask, "why are these persons, places, times, or events troublesome, when other similar entities are far less troublesome?" Answering such questions requires you to compare problem cases to non-problem cases.
This kind of comparison is called a "case-control study." A case-control study involves comparing troublesome persons, places, times, or events with untroublesome persons, places, times, or events. The troublesome cases are called the "cases." The cases to which they are compared are called the "controls."
An example comes from data supplied by Susan Wernike, a crime analyst for the Shawnee, Kansas Police Department. For every bar in Shawnee, she calculated the calls per 100 persons of rated capacity. This adjusted for bars of various sizes (see Step 27). The figure shows the bars ranked from highest to lowest rate. A basic case-control study could be applied here. The bars with the highest rates would be compared to those with the lowest rates to see if there are systematic differences in the ways they operate, the behaviors in the bars, and the types of customers they attract.
Case-control studies are very helpful when the troublesome cases are rare relative to the untroublesome ones. This is frequently the situation in problem solving.
To conduct a valid case-control study you should do the following:
- Define your cases precisely.
- Select a representative sample of these cases.
- Define a group of controls that could have been troublesome but did not become troublesome even though they were exposed to similar conditions (e.g., in the same neighborhood or city, serve the same types of clients, etc.).
- Select a representative sample of these controls.
- Compare the characteristics of the cases to the characteristics of the controls.
Substantial differences indicate characteristics that might be contributing to the problem. Similarities indicate characteristics that are probably not contributing to the problem. Let's explore these steps with an example.
The simple example of Shawnee bars illustrates the first four steps for a very small number of locations. We can also examine a more complex example to point out some of the details of case-control studies.
Define the cases precisely. In the early 1990s, John Eck was interested in why a few places were persistent drug dealing locations and most other nearby places were not. With the help of the San Diego Police Department, California, over 300 cases of persistent drug places were identified in one San Diego neighborhood. These were identified based on citizen calls, drug enforcement actions, field interrogation records, arrest data, and patrol officer observations. To be classified as persistent, each site either had to have more than one drug arrest, call, or field interrogation on separate days; or a warrant for a raid; or be identified by a patrol officer. Because multiple indicators were used to locate these sites, it was reasonable to believe that most persistent sites were located.
Select a representative sample of cases that were troublesome. For each block in the neighborhood with a single dealing location, the dealing location was selected for study. If two dealing locations were on the block, both were selected. If there were more than two, two were randomly selected. This provided a representative sample of 189 locations. Selecting all the cases was another valid option, but that would have raised the costs of the study. Simple random sampling could also have been used, but would not have guaranteed coverage of all affected blocks.
Define a group of controls that could have been troublesome. Controls were places in the neighborhood that showed no evidence of drug dealing. Nearby places were useful because drug sellers looking for a dealing site would know them. Therefore, the nearby places were exposed to dealers but had not been selected by dealers. Controls should be entities that could have been cases, but for mysterious reasons did not become so. The objective of the study is to solve the mystery.
Select a representative sample of controls. On each block the same number of non-dealing places was selected as dealing places. These sites were randomly selected (as would be the case if they were in a lottery) from a list of places on each block that had no indicators of drug activity. Selecting all non-dealing locations would have been impractical, since there were thousands of nondealing locations. By making sure cases and controls were from the same block, the selection process assured that the controls were exposed to drug dealers. Random selection assured us that the controls were representative of all non-dealing locations.
Compare the cases to the controls. Observers were sent to the cases and the controls to record information about the sites. This included information on: the type of structure (business, apartment building, single family home, vacant lot, etc.); the type of street it was on (number of lanes, one-way/two way, etc.); the distance from the nearest interstate highway; the types of surrounding buildings; the proximity of lighting; the number of apartment units; the presence of fences and other security; adjacency to alleys and paths; and many other factors. The objective was to see if the dealing locations differed substantially from the non-dealing locations with regard to any of these characteristics. Two patterns were found, one for crack-dealing sights and the other for methamphetamine sites. Compared to controls, crack locations were more likely to be in small apartment buildings and have a lockable gate in a fence. Compared to controls, methamphetamine sites were more likely to be in single-family homes and adjacent to paths. The seeming preference of drug dealers for rental units in small buildings (either buildings with few apartments or single family homes) suggested that they were looking for places with low place management. In a later experiment, Eck found that intervening with landlords did reduce drug-related crime.
Case-control studies are different from most other studies and require special techniques to analyze data. Step 33 describes one technique that is particularly useful.
Case-control studies are very useful in problem analysis. The approach is flexible enough to be applied to a small number of places (as in the Shawnee example) or a very large number (as in the San Diego example). Though the examples focused on places, the same process can be applied to people, times, and events.
COMPARING HIGH AND LOW CALL RATE BARS - Shawnee, KS
Do Not Perform an Uncontrolled Case Study
A common mistake is to collect data only on the persons, places, times, or events that are troublesome. This can provide misleading results because you learn only about characteristics common among the troublesome cases, but not if they are different from untroublesome cases. An example of this is a study conducted by the FBI in the early 1990s to understand killings of police officers. The researchers collected information on officers killed in the line of duty, but did not collect information on officers who were exposed to similar conditions but not killed. Consequently, we do not know which, if any, of the characteristics of the dead officers contributed to their killing. Living officers exposed to the same conditions may share many of these same characteristics.
- Loftin, Colin and David McDowall. (1988). "The Analysis of Case-Control Studies in Criminology." Journal of Quantitative Criminology 4:85-98.