Table 3 Dealing with missing data for main analyses and sensitivity analyses

Main ^a

All participants will be included if they have the particular outcome being assessed measured at the follow-up.

Data are MAR

The number included in these main analyses will differ for each outcome e.g. based on comments above regarding likely levels of missing data for each specific outcome measure it is possible that fewer participants will contribute to accelerometer outcomes than questionnaire outcomes

An indicator variable (indicating whether baseline data are missing or not for each outcome) together with allocation of a ‘temporary’ value to those with baseline missing data, will be used to deal with missing baseline data[22]

S1

Similar to above but participants are only included for each measurement if they have both baseline and follow-up data observed for each outcome

As above

Numbers will differ for each outcome.

Allows assessment of whether those with missing baseline data differ in terms of the trial effect compared with those who do not have missing baseline data

S2

Similar to above but participants are only included if they have both baseline and follow-up data of all three primary outcomes

As above

For the three primary outcomes numbers will be the same numbers may differ for each secondary outcome.

Allows assessment of whether any apparent differences in effect for the three primary outcomes are due to differs between these outcomes in missing data mechanisms

S3

Similar to the main analyses but for any child with a missing follow-up measure the child is allocated a value that is 10% ‘healthier’ for a given outcome than all participants with observed data (irrespective of randomised group). This will be done by calculating the 10% value of the mean or median follow-up measure for each outcome and then adding or subtracting (depending on whether healthier levels are higher or lower for the particular outcome) this value to the outcome mean or median; this final value will then be imputed to the outcome value for every child with missing follow-up data.

Those with missing outcome data on average behave in a relatively healthy way.

Numbers will be the same for all outcomes.

Allows assessment of the possibility that missing data may be more likely to occur in families from higher SEP who may have missing data because of moving from state to private education. And to assess whether this form of missing data biases our assessment of the trial effect.

This will also test whether selection bias occurs as a result of limiting analyses only to those with the required wear-time for the accelerometer based outcomes (this outcome is likely to have more missing data than other outcomes). As these analyses include all recruited participants.

S4

Similar to the main analyses but for any child with a missing follow-up measure the child is allocated a value that is 10% ‘less healthy’ for a given outcome than all participants with observed data (irrespective of randomised group). This will be done by calculating the 10% value of the mean or median follow-up measure for each outcome and then adding or subtracting (depending on whether less healthy levels are higher or lower for the particular outcome) this value to the outcome mean or median; this final value will then be imputed to the outcome value for every child with missing follow-up data

Those with missing data on average behave in less healthy ways than those who do not have missing data through mechanisms that are not captured by observed data

Numbers will be the same for all outcomes.

Allows assessment of the possibility that missing data may be more likely to occur in families from lower SEP and who may have missing data because of being more dysfunctional and perhaps having to care for a relative at home or having higher rates of truancy. And to assess whether this form of missing data biases our assessment of the trial effect.

This will also test whether selection bias occurs as a result of limiting analyses only to those with the required wear-time for the accelerometer-based outcomes (this outcome is likely to have more missing data than other outcomes). As these analyses include all recruited participants