Table 4.

Severity of publication bias, S, in Li et al. (2018) required to attenuate ${\widehat{\mu }}^{\prime }$ or ${\widehat{\mu }}^{\mathrm{ub}\prime }$ (the upper limit of the CI for ${\widehat{\mu }}^{\prime }$) to the null or to q ^′=0.90†

Model	$S({\widehat{\mu }}^{\prime },\mathit{1})$	$S({\widehat{\mu }}^{\mathrm{ub}\prime },\mathit{1})$	$S({\widehat{\mu }}^{\prime },\mathit{0.90})$	$S({\widehat{\mu }}^{\mathrm{ub}\prime },\mathit{0.90})$
Common effect	14	5	3	2
Robust (independent)	8	2	2	1
Robust (clustered)	8	2	2	1

†‘1’ indicates that the statistic is already greater than or equal to q ^′. Values are conservatively rounded down to the nearest integer.