Writing
The link between sleep and mood, and how to test it on yourself
That sleep affects mood is well established. What that means for you specifically — how much, at what threshold — is not, and is testable.
What is well established
The relationship between sleep and next-day mood is one of the sturdier findings in this area, and it is worth naming the evidence rather than gesturing at it. Pilcher and Huffcutt's 1996 meta-analysis in Sleep pooled the experimental sleep-deprivation literature and found something people usually get backwards: mood was affected more than cognitive performance, and more than motor performance. The thing most degraded by losing sleep is not how well you think. It is how you feel.
The longer-range evidence points the same way. Baglioni and colleagues, in a 2011 meta-analysis in the Journal of Affective Disorders, pooled longitudinal studies and found that people with insomnia were around twice as likely to go on to develop depression as people without it. That is an association in observational data rather than a demonstrated cause, and it should be read as one — but it is a large, replicated association, and it is measured over years rather than a single bad night.
Beyond those, it shows up in daily-diary studies where people report both over weeks, and in the clinical literature, where sleep disturbance is entangled with mood disorders closely enough that it appears as both a symptom and a risk factor.
It also appears to run in both directions. Poor sleep degrades next-day mood; low mood degrades that night's sleep. This matters for interpretation, because a correlation you find in your own data will not tell you which way the arrow points — and in this particular case there is good reason to think it points both ways.
What is not established, for you
Population averages describe populations. They are a poor guide to an individual, and the size of the individual differences here is large.
Your threshold is a specific example. "Adults need seven to nine hours" is a population range, not a personal prescription — some people are genuinely fine at six and a half, some are impaired at seven and a half, and the average tells you which is more common, not which you are. The same applies to whether duration or consistency matters more for you, and to how long the effect lasts.
None of that is knowable from an article. It is knowable from a few weeks of your own paired records.
How to actually test it on yourself
Record both sides daily, at a fixed time. Mood is best captured at a consistent point — evening works for most people — because a score taken at 8am and one taken at 11pm are not measuring the same thing, and mixing them adds noise you will later mistake for signal.
Pair the right days. Last night's sleep against today's mood is the comparison you want if you are asking whether sleep drives mood. Getting the offset wrong is the most common way a personal analysis of this ends up meaningless.
Collect at least 20 paired days before drawing anything. Below that a correlation coefficient swings wildly on single observations, and you will find a strong-looking relationship in ten days of data that evaporates in thirty.
Decide in advance what you are testing. If you compute every possible comparison and report whichever looks strongest, you will find something — that is guaranteed by the number of comparisons, not by your data. Deciding the question first is what makes the answer mean anything.
Expect nothing, often. A null result after a month is a real and common outcome. It is more informative than a spurious finding, because acting on a spurious finding costs you weeks.
A worked example
Suppose you have recorded 46 paired days: each night's sleep duration, and the following evening's mood on a 1-to-5 scale. You pair last night's hours with today's mood, run the correlation, and get r = 0.31.
First, what that is. A coefficient of 0.31 means roughly 10% of the day-to-day variation in your mood scores moves together with your sleep — r squared, 0.31 × 0.31, is about 0.096. Ninety percent of the variation is something else: what happened at work, whether you saw anyone, the weather, and a large helping of noise in a self-reported five-point scale. A real effect and a small one are not opposites.
Second, whether it is worth believing. With 46 observations, r = 0.31 sits around p = 0.036 on a two-sided test — under the conventional 0.05, so on its own it would be called significant. But if you tested eight habits against mood that month, the threshold is no longer 0.05. Dividing it across the eight comparisons gives 0.00625, and 0.036 does not clear that. The same number is either a finding or not depending entirely on how many questions you asked, which is why deciding the questions in advance is not a technicality.
Third, what to do with it. Not "sleep eight hours". The coefficient describes a linear trend across your observed range; it says nothing about a threshold. If you want to know whether there is a cliff at seven hours, that is a different comparison — split the days into above and below seven and compare the two groups' mood — and it needs enough days on both sides of the line to be worth running.
Reading the number you get
A correlation coefficient runs from -1 to 1. For self-tracked daily data, anything above about 0.25 in absolute value is worth attention and anything above 0.4 is strong for this kind of noisy, single-subject measurement. Do not expect 0.8; that is not what human behavioural data looks like.
And the coefficient alone is not enough. You need the sample size beside it and a p-value corrected for how many comparisons you ran. A coefficient of 0.5 over 12 days and one of 0.3 over 90 days are very different claims, and only one of them survives contact with more data.
This is the work Crescendo does automatically — pairing the days, applying the offset, holding the hypotheses fixed in advance, correcting for the number of tests, and showing the coefficient and sample size on every card whether or not it cleared the threshold. But none of it requires the app. A spreadsheet and a month of honest recording will answer the question too.