Facebook and mental wellbeing: a crossover randomised controlled study

Study design

This study used a crossover design with simple randomisation (1:1 ratio), involving a two-week period with minimal Facebook use, and a two-week control period. No changes were made to the study protocol after receiving ethics approval through the University of Oxford’s Medical Sciences Interdivisional Research Ethics Committee (Ref#: MSD-IDREC-C1-2014-035), which was granted before commencement of the study. Trial registration was completed retrospectively through the Australian New Zealand Clinical Trials Registry (Ref#: ACTRN12614000959673, www.anzctr.org.au/Trial/Registration/TrialReview.aspx?id=367015).

Participants

Participants were recruited between March 27 and May 25, 2014 and had to be full-time students at the University of Oxford, aged over 18, and regular (daily) Facebook users. No exclusion criteria were used. Students were recruited through adverts in student newsletters and Facebook pages, and departmental mailing lists. They were followed up for four weeks, with follow-ups starting on March 27, 2014 and ending June 22, 2014. There was no face-to-face contact between participants and researchers, or compensation for participants.

Eligibility data, informed consent, and questionnaire answers were obtained electronically. To confirm eligibility, students had to register with a valid University of Oxford email address.

Intervention

At the beginning of the intervention period, participants were asked to “restrict [their] Facebook use as much as possible for two weeks” and were told “During this time, please avoid using Facebook, including through mobile phones, tablets, and computers.” At the beginning of the control period, participants were told they “can use Facebook as much as [they] want” for two weeks. Participants were sent weekly reminders. Facebook use was measured through self-reporting only.

After each two-week period participants were sent links with questionnaires, and were asked to complete them regardless of whether they had followed the instructions given to them.

Outcomes

The primary outcome is within-individual differences in Warwick-Edinburgh Mental Well-Being Scale (WEMWBS) score between the two periods. WEMWBS is a 14-item measure of positive mental wellbeing which has been validated in UK student populations⁵. Secondary outcomes were stratification by two variables: gender, and relationship status.

Sample size

The sample size of 78 was calculated assuming a power of 0.8, an α of 0.05, a minimal detectable difference in means of 2 points in WEMWBS score, a 20% dropout rate, and a standard error of measurement (SEM) of 2.78 (mean of the five SEMs reported in Maheswaran et al.⁶).

Randomisation

Participants were randomised on a rolling basis after completing the informed consent form, using the rand() function in Excel 2013 (v.15.0). At randomisation, no data other than the participant’s email address had been collected. Participants were not blinded to group allocation.

Statistical methods

The statistical significance of the primary outcome was calculated using a paired t-test. Only participants who completed both sets of questionnaires were entered in this analysis.

In a sensitivity analysis, all observations were entered into a regression analysis using Stata’s clustered sandwich estimator (vce(cluster) function) to allow for within-participant correlation in data with repeated observations on individuals⁷, and to minimise missingness.

We present self-reported descriptive statistics on age, gender, relationship status, and whether participants reduced their use during the intervention period.

For secondary outcomes, we stratified analyses by gender (female vs. male) and relationship status (single vs. in a relationship).

All analyses were done in Stata, version 12. Observations were entered according to group allocation, following an intention-to-treat approach. Statistical assumptions for paired t-tests were tested using a histogram of the differences in scores (Assumption of Normality, Supplementary figure 1) and by plotting differences in scores against average scores (Supplementary figure 2).

Participants

At randomisation, 40 participants were allocated to Group AB and 38 to Group BA. Two-thirds of participants were female, 46% were single. There was no significant difference in dropout between Group AB and Group BA (Table 1).

During the intervention period 25/27 (93%) in Group AB and 28/30 (93%) in Group BA reported reduced Facebook usage (‘less than usual’ or ‘much less than usual’).

Outcomes

Participants obtained average WEMWBS scores of 46.0 (95% CI: 44.3 to 47.7) during the intervention period, and 43.7 (95% CI: 41.7 to 45.7) during the control period. The average difference was 2.3 points (95% CI: 0.4 to 4.2, p=0.016). This corresponds to a Cohen’s d of 0.33 (SD of difference = 7.0).

In a sensitivity analysis additionally including those with follow-up at 2 weeks only, the average difference in scores was 2.6 points (95% CI: 0.7 to 4.5, p=0.009).

We found no significant differences in effect by gender or relationship status (Table 2).

Limitations

Limitations include high dropout (possibly due to the lack of compensation) and a short follow-up period. Almost all students reported reduced Facebook use. Nevertheless, results could be affected by expectations, or perception of reduced use rather than actual changes in use.