Attrition Bias

You start a diet study with 100 people. At the end, 50 people remain. They all lost weight. Conclusion: The diet works 100% of the time!

Wrong.

What happened to the other 50?

The Dropouts

They quit because the diet was too hard. Or they gained weight and felt ashamed. Or they got sick.

These are the failures. If you ignore them, you are cheating.

This is Attrition Bias. The loss of participants distorts the results.

The Drug Trial Trick

In drug trials, people often drop out because of side effects.

If 20% of the treatment group quits because of nausea, and you only analyze the survivors, the drug looks safe.

But in the real world, people will quit. The drug is not effective if people won't take it.

Intention-to-Treat

The solution is Intention-to-Treat (ITT) analysis.

Once you are randomized, you are analyzed. Even if you quit on day one. Even if you never took a pill.

If you drop out, you are counted as a failure. This is harsh. But it is honest.

The Diagnosis

Check the dropout rates. If the treatment group lost more people than the placebo group, be suspicious.

The missing data is often more important than the existing data.