Hybrid working from home improves retention without damaging performance

One key motivation for Trip.com in running the experiment was to evaluate how hybrid WFH affected employee attrition and job satisfaction. The net effect was to reduce attrition over the experiment by 2.4%, which against the control-group base of 7.2% was a one-third (33%) reduction in attrition (meancontrol = 7.20, meantreat = 4.80, t(1610) = 2.02, P = 0.043). Consistent with this reduction in quit rates, employees in the treatment group also registered more positive responses to job-satisfaction surveys (meancontrol = 7.84, meantreat = 8.19, t(1343) = 4.17, P < 0.001). Employees were anonymously surveyed on 21 January 2022, and employees in the treatment group showed significantly higher scores on a scale from 0 (lowest) to 10 (highest) in ‘work–life balance’, ‘work satisfaction’, ‘life satisfaction’ and ‘recommendation to friends’, and significantly lower scores in ‘intention to quit’ (Extended Data Table 2).

One possible explanation for the lower quit rates in the treatment group is that quit rates in the control group increased because the individuals in this group were annoyed about being randomized out of the experiment. However, quit rates in the same Airfare and IT divisions were 9.8% in the six months before the experiment—higher than the rate for the control group during the experimental period. Quit rates over the experimental period in the two other Trip.com divisions for which we have data (Business Trips and Marketing) were 10.5% and 9.8%—again higher than that for the control group during the experimental period. This suggests that, if anything, the control-group quit rates were reduced rather than increased by the experiment, possibly because some of them guessed (correctly) that the policy would be rolled out to all employees once the experiment ended.

Figure 2 shows the change in attrition rates by three splits of the data. First, we examined the effect on attrition for the 1,217 non-managers and 395 managers separately. We saw a significant drop in attrition of 3.3 percentage points for the non-managers, which against a control-group base of 8.6% is a 40% reduction (meancontrol = 8.59, meantreat = 5.33, t(1215) = 2.23, P = 0.026). By contrast, there was an insignificant increase in attrition for managers (meancontrol = 2.96, meantreat = 3.13, t(393) = −0.098, P = 0.922). We also found that non-managers were more enthusiastic before the experiment, with a volunteering rate of 35% (versus 22% for managers), matching the media sentiment that although non-managerial employees are enthusiastic about WFH, many managers are not (t(1610) = 4.86, P < 0.001).

Second, we examined the effect on attrition by total commute length, splitting the sample into people with shorter and longer total commutes on the basis of the median commute duration (two-way commutes of 1.5 h or less versus those exceeding 1.5 h, with 648 and 611 employees, respectively). We found that there was a larger reduction in quit rates (52%) for those with a long commute (meancontrol = 6.00, meantreat = 2.89, t(609) = 1.87, P = 0.062). The reduction in quit rates was similarly large for employees with a long commute if we instead defined a long commute as a two-way commute time exceeding 2 h (meancontrol = 7.33, meantreat = 1.89, t(307) = 2.31, P = 0.021). Employees who volunteered to take part in the experiment had longer one-way commute durations (Extended Data Table 3; meannon-volunteer = 0.80, meanvolunteer = 0.89, t(1257) = −3.68, P < 0.001). This is not surprising given that the most frequently cited benefit of WFH is no commute.

Third, we examined the effect on attrition by gender, examining the 570 female and 1,042 male employees separately. We found that there was a 54% reduction in quit rates for female employees (meancontrol = 9.2, meantreat = 4.2, t(568) = 2.40, P = 0.017). For male employees, there was an insignificant 16% reduction in quit rates (meancontrol = 6.15, meantreat = 5.15, t(1040) = 0.70, P = 0.487). This greater reduction in quit rates among female individuals echoes the findings of previous studies, which suggest that women place greater value on remote work than men do. Notably, although the treatment effect of WFH was significantly larger for female employees, volunteers were less likely to be female (meannon-volunteer = 0.37, meanvolunteer = 0.32, t(1610) = −2.02, P = 0.043); this might suggest that women have greater concerns about negative career signalling by volunteering to WFH.