Decline effects are rare in ecology

Abstract

The scientific evidence base on any given topic changes over time as more studies are published. Currently, there is widespread concern about nonrandom, directional changes over time in the scientific evidence base associated with many topics. In particular, if studies finding large effects (e.g., large differences between treatment and control means) tend to get published quickly, while small effects tend to get published slowly, the net result will be a decrease over time in the estimated magnitude of the mean effect size, known as a “decline effect.” If decline effects are common, then the published scientific literature will provide a biased and misleading guide to management decisions, and to the allocation of future research effort. We compiled data from 466 meta‐analyses in ecology to look for evidence of decline effects. We found that decline effects are rare. Only ~5% of ecological meta‐analyses truly exhibit a directional change in mean effect size over time arising for some reason other than random chance, usually but not always in the direction of decline. Most apparent directional changes in mean effect size over time are attributable to regression to the mean, consistent with primary studies being published in random order with respect to the effect sizes they report. Our results are good news: decline effects are the exception to the rule in ecology. Identifying and rectifying rare cases of true decline effects remains an important task, but ecologists should not overgeneralize from anecdotal reports of decline effects.

Publication bias impacts on effect size, statistical power, and magnitude (Type M) and sign (Type S) errors in ecology and evolutionary biology

“…The observed power to determine a decline effect in the current set of 87 meta-analyses was considerably low (median = 13%). This low power was similar to that observed in another much larger survey of 464 ecological meta-analyses (median = 17%; [68,69]). Importantly, our second-order meta-analysis have found a statistically significant and homogeneous effect (Figure 6A), corroborating that decline effects are common in both subfields previously explored (status signalling: [70], plant and insect biodiversity: [17,71] and ocean acidification: [72]) and general fields of ecology and evolutionary biology [12,68].…”

How much does the typical ecological meta‐analysis overestimate the true mean effect size?

“…This assumption usually holds but not always. Approximately 3%–5% of ecological meta‐analyses exhibit true directional trends in mean effect size over time as more and more studies are published (Costello & Fox, 2022 ). Those directional trends arose because primary research studies were published in nonrandom order with respect to the effect sizes they reported, usually with larger‐magnitude effect sizes being published earlier (Costello & Fox, 2022 ).…”

“…However, even meta‐analytic estimates of mean effect size remain at least somewhat imprecise, particularly estimates from small meta‐analyses. The median ecological meta‐analysis comprises just 64 effect sizes from just 24 primary research studies (information extracted from data compiled by Costello & Fox, 2022 ). Just by chance, small meta‐analyses are more likely than large ones to produce imprecise, large‐magnitude estimates of mean effect size.…”

“…Here I address this question using meta‐meta‐analysis. A meta‐meta‐analysis, also called second‐order meta‐analysis, is a meta‐analysis of meta‐analyses (Barto & Rillig, 2012 ; Busch & Friede, 2018 ; Cafri et al, 2010 ; Costello & Fox, 2022 ; Da Costa et al, 2015 ; Eisend & Tarrahi, 2016 ; Eisend, 2015 ; Fanelli et al, 2017 ; Fanshawe et al, 2017 ; Jennions & Møller, 2002 ; Martin et al, 2022 ; Mathur & VanderWeele, 2021 ; Peterson, 2001 ; Rebar et al, 2015 ; Sáiz‐Vazquez et al, 2020 ; Steenbergen‐Hu et al, 2016 ; Tan et al, 2020 ; van Aert et al, 2019 ; Yang et al, 2022 ; Young, 2017 ). Meta‐meta‐analyses can be used for various purposes.…”

“…Meta‐meta‐analyses can be used for various purposes. Meta‐meta‐analyses can be used to estimate the prevalence and/or typical magnitude of publication biases (Barto & Rillig, 2012 ; Costello & Fox, 2022 ; Fanelli et al, 2017 ; Fanshawe et al, 2017 ; Jennions & Møller, 2002 ; Mathur & VanderWeele, 2021 ; Pietschnig et al, 2019 ; van Aert et al, 2019 ). Different meta‐analyses of the same effect can be combined into a meta‐meta‐analysis to obtain a more precise estimate of the mean effect size, estimate heterogeneity in mean effect size among meta‐analyses, and improve power to detect effects of moderator variables associated with variation in effect size (Castellanos & Verdú, 2012 ; Da Costa et al, 2015 ; Duke et al, 2017 ; Eisend & Tarrahi, 2016 ; Jüni et al, 2002 ; Martin et al, 2022 ; Mingebach et al, 2018 ; Peterson, 2001 ; Rebar et al, 2015 ; Sáiz‐Vazquez et al, 2020 ; Steenbergen‐Hu et al, 2016 ; Tan et al, 2020 ; Young, 2017 ).…”