When it is the parent who answers about the child

Before seven or eight, a child does not complete a reliable questionnaire about their own conduct or sleep. Asking the parent is the only route feasible at scale, and it has the advantage of covering weeks of observation rather than one session.

The instruments most used in the field are built for this source and validated with it, which makes comparison between studies possible but locks the field into a single mode of collection.

A meta-analysis conducted by Achenbach and colleagues in the late 1980s established a now classic result: agreement between two adults describing the same child is moderate, and weaker still between an adult and the child themselves.

More recent work, carried out with other instruments and in other countries, has found comparable orders of magnitude. Disagreement between informants thus ranks among the most consistent results of measurement in child psychology.

That disagreement is not a flaw to be corrected. Parents and teachers observe the child in different situations, and part of the gap measures a real variation in conduct by context.

The adult's own state weighs on their description. Several studies show that a parent with depressive symptoms reports more difficulties in their child, including when other sources report none.

To which are added social desirability, which pushes towards understating certain conduct, and a halo effect: a favourable general judgement colours answers to precise and unrelated questions. These biases do not all run in the same direction, which complicates correcting them: a parent may understate a behaviour judged unacceptable and overstate a difficulty that worries them, in the same questionnaire.

The recommended practice is to collect several sources and treat them separately rather than averaging them, which would erase the information contained in the disagreement.

So-called objective measures, actigraphy, audio recording of speech addressed to the child, observation coded by a third party, are developing. They are costly, applicable to small samples, and they often measure something other than what the questionnaire measures.

An article reporting an association between a parenting practice and a trait of the child, both described by the same parent, partly measures the respondent's own consistency. This is known as shared method variance.

It does not cancel the result, but it limits its reach, and it explains why an association that is strong in self-reported data becomes weak as soon as an independent source is introduced.

The distinction often comes down to a single phrase. A study concluding that certain children are more anxious is really saying that their parents describe them as more anxious, and that nuance is almost always lost when the results are picked up.