← Knowledge

Correlation is not diagnosis

Where statistical patterns start to make sense, where they deceive, and when the path leads to a doctor.

MethodologyBiasCaution

The clean sentence up front

Correlyn computes correlations between your food logs and your symptom logs. A correlation means: two events occur together more often than would be expected. This is not a statement that one caused the other.

If the tomato statistic shows 70% match in 2–24h, it is possible that tomato is your trigger. It is equally possible that when you eat tomato, three other things happen (late meal, red wine with it, post-work stress) which are the actual trigger.

Classic biases in self-tracking

1. Confirmation bias

You log more symptoms when you suspect that today matters. Tomato suspicion → you pay particular attention to headaches you would otherwise have overlooked. Result: the app “confirms” your suspicion artificially.

Antidote: also log negative days (no symptoms, nothing special — just “ok”).

2. Recall bias

You remember symptoms after tomato better than days with tomato and no symptoms. But the app only computes with what you actually enter.

Antidote: log promptly, ideally right after the meal / symptom. More than 6h later from memory = garbage data.

3. Selection bias

If you only log notable days, it looks as if everything correlates with everything.

Antidote: also track normal days — then the Wilson confidence has a denominator and can do its work.

4. Spurious correlations

With enough pairs compared, math will always find some correlation even if nothing is behind it (Schwarz et al. 2016 on multiple testing). If you track 20 foods × 8 symptoms × 4 windows that is 640 pairs — by pure chance some “notable” correlations appear.

Antidote: we show the Wilson lower bound and N. Look at both, not only at percent.

5. Confounders (third variable)

Antidote: the longer you log, the more this smooths out. But never completely.

When a pattern gets serious

A rule of thumb we find useful (no science, just experience):

If all four hold, an elimination / re-exposure trial is worthwhile. But: always discuss with someone qualified — family doctor, dietitian, allergist.

When to see a doctor

The honest tagline

Correlyn is a self-tracking tool. It turns your data into math, not math into diagnosis. The step from “pattern” to “trigger” happens outside the app — with your doctor, with an elimination test, or with healthy self-experiment.


This article is educational content, not a medical diagnosis or therapy recommendation.

Sources

Notice: This article is educational content, not a medical diagnosis or therapy recommendation. For concrete complaints, please seek medical evaluation.