Measure	Questionnaire items	Analysis variable
Depressive symptoms	Participants completed the Mood and Feelings Questionnaire – short version (SMFQ) from which a summed score was created. The SMFQ comprises 13 items on affective symptoms in the last 2 weeks as follows: felt miserable or unhappy; didn't enjoy anything at all; so tired just sat around and did nothing; was very restless; felt I was no good anymore; cried a lot; found it hard to think properly or concentrate; hated myself; was a bad person; felt lonely; thought nobody really loved me; thought I could never be as good as other kids; did everything wrong.	Log transformed continuous variable used in modelling; generated dichotomous variable indicating clinically relevant symptoms (cut point ≥ 12)
Social media usea	Respondents were asked “On a normal week day during term time, how many hours do you spend on social networking or messaging sites or Apps on the internet such as Facebook, Twitter and WhatsApp?” (response categories: None, less than half an hour, half an hour to less than 1 h, 1 h to less than 2 h, 2 h to less than 3 h, 3 h to less than 5 h, 5 h to less than 7 h, 7 h or more).	Categories were collapsed to generate a variable as follows: none, < 1 h, 1 to < 3 h, 3 to < 5 h, ≥ 5 h.
Online harassmentb	“How often have other children sent you unwanted or nasty emails, texts or messages or posted something nasty about you on a website?”; “How often have you sent unwanted or nasty emails, texts or messages or posted something nasty about other children on a website?” response categories for both questions: most days; about once a week; about once a month; every few months; less often; never	Combined responses capturing any involvement as victim and/or perpetrator to generate a variable with 4 categories: no involvement; victim; perpetrator; perpetrator-victim. (Adapted from Fahy et al[12])
Sleep duration	“About what time do you usually go to sleep on a school night?” “About what time do you usually wake up in the morning on a school day?”.	A 4-category variable was generated: 7 h or less, 8, 9, 10 + h
Sleep latency	A sleep latency variable was constructed from answers to the question “During the last four weeks, how long did it usually take for you to fall asleep?”	A 3-category variable was created: 0–30, 30–60, > 60mins.
Sleep disruption	Disruptions to sleep were assessed using the question “During the last four weeks, how often did you awaken during your sleep time and have trouble falling back to sleep again?”	A 4-category variable was created: all/most of the time; often; a little of the time; and none of the time.
Self-esteem c	Self-esteem was assessed using the items on self-satisfaction from the Rosenberg scale: having good qualities; able to do things similar to others; person of value; and feel good about oneself.	A dichotomised variable (low vs normal/high) derived from the sum of the items, scores ≥ 7 (i.e. the top 20% of the distribution) indicate low self-esteem.
Happiness with appearance	Happiness with appearance was measured, as follows: “On a scale of 1 to 7 where ‘1’ means completely happy and ‘7’ means not at all happy, how do you feel about the way you look?”	A log transformed continuous variable was used in modelling. A dichotomised variable (1–6 vs 7) was used for display purposes in Table 1, Table 2.
Body weight satisfaction	Body weight satisfaction was assessed from 3 items: “Which of these do you think you are?” (underweight, about the right weight, slightly overweight, very overweight), “Have you ever exercised to lose weight or to avoid gaining weight?”, “Have you ever eaten less food, fewer calories, or foods low in fat to lose weight or to avoid gaining weight?”.	Responses other than ‘about the right weight’ or affirmative to exercising or eating to lose or maintain weight were combined to generate a body satisfaction variable (satisfied vs dissatisfied).

^aAlternative specifications that assumed a continuous normal distribution were rejected due to heteroscedasticity.

^bTreating the online harassment victim and -perpetrator variables as ordinal and testing for an interaction between the two variables resulted in an unwieldy number of parameters. Assuming the variables to be continuous was not tenable due to their distributional patterns.

^cThe Rosenberg Scale had a distinctly non-normal distribution for which no transformation was satisfactory. Regression models using the raw scale show significant heteroscedasticity.