How do I find suitable comparison people for each case? (matching cases and controls)

In brief

How do I find suitable comparison people for each case? The data assistant "Match cases and controls" (Edit data) looks for up to k controls (0) for each case (1) that are equal or similar in the chosen characteristics (matching). The columns "Gruppe" and "Abgleich" are created.

How it works in Statable

1. Choose the column with case (1) and control (0); optionally columns that must be exactly equal.
2. Choose number columns, each with a tolerance: the largest allowed distance. With 0 only an equal value counts.
3. Choose "Controls per case" (1 to 5).
4. "Calculate the matching", look at the result, then "Create new columns".

The rule

• The cases are worked through in the order of the rows.
• A control fits if it is not yet used, matches in all characteristics that need exact equality and is at most the tolerance away in each number.
• Among the fitting controls the one with the smallest distance wins, in a tie the one in the smallest row. The distance is the sum of the differences, each divided by the tolerance; with tolerance 0 it is 0.
• A control is used only once (without replacement).
• A row with a missing value in the case column or in a matching column does not take part. Statable counts such rows.

Reading the result

• The table shows: cases in total, fully matched (all k controls found), partly matched, without a control, the number of controls found and the rows that do not take part.
• For each number column the standardized mean difference before and after is shown: the difference between the means of cases and controls, divided by the common spread. Before means all cases and controls, after only the matched ones. A value closer to 0 afterwards means: the groups have become more similar in this characteristic.

Warnings

• The method is greedy and depends on the order of the rows: for each case in turn it takes the next free control and does not look for the best overall result. A different order can give different pairs.
• Cases without a control drop out of the comparison. This changes whom the comparison describes.
• The matching balances only the chosen characteristics. Everything else stays unequal, and the matching does not show a cause.
• The analysis afterwards must take the matching into account, for example with paired or conditional methods. Comparing only within the groups that remain is not an experiment.
• Do not calculate and choose from the same data until it fits: anyone who changes tolerance and characteristics until the result pleases can no longer trust the result.

Requirements and limits

• The case column has the values 1 (case) and 0 (control). At least one case and one control with all values are needed.
• A column may appear in only one place: as the case column, as an exactly equal column or as a number. A case weight that is set does not enter the matching.
• This step is entered in the log of the data changes.

Typical examples

• For each person with the illness (case) two persons without it (controls) of the same sex and at most 2 years apart in age.
• Form a comparison group for a support group by age and prior knowledge.