Every cause is checked against every result
Each day's answers line up by date. We compare every cause question in your study with every result question, one pair at a time. A study with 7 causes and 2 results gets 14 separate comparisons. To see what they may add up to together, there's Ask AI.
Back to top ↑The right test for each kind of question
Questions are either yes/no or a number on a scale. Each combination gets the test that fits it:
Do higher values of one tend to come with higher, or lower, values of the other?
Spearman's rank correlation
Was the number different on “yes” days than on “no” days?
Mann–Whitney U test
How much more likely was the result on days you did the cause?
Fisher's exact test
Which test runs is decided by how the two questions are set up, before any data exists, never by picking whichever looks best afterward. These are standard methods suited to the small samples common in personal tracking.
Back to top ↑Effects that show up later
Some things don't show up the same day. By default, each pair is checked from the same day through 3 days later. You can change the maximum lag in a study's settings, from 0 to 14 days.
We headline the delay with the strongest statistical evidence, based on the smallest adjusted q-value; effect size breaks a tie. We also show the delays tested so you can see how the estimated effect changes across them.
Back to top ↑Guarding against lucky coincidences
Checking many combinations has a catch: some can look meaningful by chance. A study with 7 causes and 2 results, checked at 4 delays, runs 56 tests. If none of the patterns were real, about 2.8 could still pass an uncorrected 0.05 significance cutoff.
We correct for all tested pairs and delays together using the Benjamini–Hochberg procedure. A link has to stand out against everything else checked, not just clear a cutoff on its own.
Adding questions makes each individual link harder to confirm, which is why a focused study is often more useful than tracking everything at once.
Back to top ↑How to read a result
Every link gets two separate labels: how big is the effect? and how much evidence supports it? A large estimated effect can still be uncertain, and a well-supported effect can be small.
Spent time outside in the morning → How rested I felt
Typical restfulness was higher on days you spent time outside, with the clearest statistical evidence the same day.
Illustrative example only; actual results depend on your data.
Effect — how big
Based on the absolute effect size for the test used.
Trust — how much evidence
Combines statistical significance and overlapping days.
For the statistically curious: strong effects are at least 0.5, moderate effects at least 0.3, and a result is significant when its adjusted q-value is below 0.05. “Possible” means an effect of at least 0.3 with q below 0.20. A pair needs at least 8 overlapping days before we compute its result; under 20 overlapping days is treated as an early signal.
Results use the most recent 30 days by default; each study can instead use a different window or all available history.
Back to top ↑Skipped days and “Not applicable”
A day you didn't log, or a question you marked Not applicable, is left out of that one comparison. It's never counted as zero or “no”, and other comparisons in your study can still use that day's answers.
Back to top ↑