On the prevalence of uninformative parameters in statistical models applying model selection in applied ecology

Abstract

1On the prevalence of uninformative parameters in statistical models applying model 2 selection in applied ecology 3 4 5 Author: Shawn J. Leroux (ORCID ID: 0000-0001-9580-0294) Abstract 14 Research in applied ecology provides scientific evidence to guide conservation policy and 15 management. Applied ecology is becoming increasingly quantitative and model selection via 16 information criteria has become a common statistical modeling approach. Unfortunately, 17 parameters that contain little to no useful information are commonly presented and interpreted as 18 important in applied ecology. I review the concept of an uninformative parameter in model 19 selection using information criteria and perform a literature review to measure the prevalence of 20 uninformative parameters in model selection studies applying Akaike's Information Criterion 21 (AIC) in 2014 in four of the top journals in applied ecology (Biological Conservation, 22 Conservation Biology, Ecological Applications, Journal of Applied Ecology). Twenty-one 23 percent of studies I reviewed applied AIC metrics. Many (31.5 %) of the studies applying AIC 24 metrics in the four applied ecology journals I reviewed had or were very likely to have 25 uninformative parameters in a model set. In addition, more than 40 % of studies reviewed had 26 insufficient information to assess the presence or absence of uninformative parameters in a 27 model set. Given the prevalence of studies likely to have uninformative parameters or with 28 insufficient information to assess parameter status (71.5 %), I surmise that much of the policy 29 recommendations based on applied ecology research may not be supported by the data analysis. I 30 provide warning signals and a decision tree to help reduce the prevalence of uninformative 31 parameters in studies applying model selection with information criteria. The four warning 32 signals and decision tree should assist authors, reviewers, and editors to screen for uninformative 33 parameters in studies applying model selection with information criteria. In the end, careful 34 thinking at every step of the scientific process and greater reporting standards are required to 35 detect uninformative parameters in studies adopting an information criteria approach. 3 36 38 39 Introduction 40 Conservation biology emerged as a crisis discipline in the 1970s in response to evidence of 41 widespread declines in biodiversity [1]. Along with the evolution of new technologies (e.g.42 Remote Sensing, Geographic Information Systems) and increasing availability of environmental 43 (e.g. Land-use) and biodiversity (e.g. species occurrence records) data, the discipline has 44 developed into a rigorous quantitative science [2][3][4]. These advances in methods and data allow 45 applied ecologists to tackle complex problems at larger temporal and spatial scales than before.46 The application of quantitative analyses and the interpretation of these analyses in applied 47 ecology is particularly important as research in this field often informs policy and...

Trophic rewilding establishes a landscape of fear: Tasmanian devil introduction increases risk‐sensitive foraging in a key prey species

“…In both analyses, we ran all simpler combinations of explanatory variables, and selected the best models using an information theoretic approach based on small-sample corrected Akaike information criterion (AICc) (Burnham et al 2011). Some models showed strong signs of containing a 'pretending variable' (sensu Anderson 2007), otherwise known as an uninformative parameter (Leroux 2019). These variables can be identified when the addition of a variable to a simpler nested model does not improve model fit (i.e.…”

“…These variables can be identified when the addition of a variable to a simpler nested model does not improve model fit (i.e. the log-likelihood) and increases the AIC value by approximately the penalty of two (Anderson 2007, Leroux 2019. In such cases, we excluded models containing a pretending variable, as recommended by Anderson (2007) and Leroux (2019).…”

“…the log-likelihood) and increases the AIC value by approximately the penalty of two (Anderson 2007, Leroux 2019. In such cases, we excluded models containing a pretending variable, as recommended by Anderson (2007) and Leroux (2019). For completeness, we present the model selection table containing the pretending variables in Supplementary material Appendix 1 Table A1, A2.…”

Responses of grassland snakes to tallgrass prairie restoration

“…We used the AICcmodavg R package to rank models with QAIC c (quasi‐Akaike's information criterion corrected for small sample sizes; Mazerolle ). We excluded models with uninformative parameters (= pretending variables; identified by ΔQAIC c ≤ 2, log likelihood similar to a simpler model, and parameter confidence limits that overlap zero as in Lerouz ) and generated model averaged parameter estimates from models with ΔQAIC c ≤ 2 (Burnham & Anderson ; Powell & Gale ).…”

Prey availability and ambient temperature influence carrion persistence in the boreal forest

“…We selected the top models using Akaike's Information Criterion, corrected for small sample sizes (AICc; Burnham & Anderson, 2002) and examined relative variable importance from the entire model set using the MuMin package in r (Barton, 2019). We then excluded candidate models that contained uninformative parameters (Leroux, 2019), which were identified when the addition of a parameter to a simpler nested model had no improvement on model fit (i.e. log-likelihood), and increased AICc by approximately the penalty of two (as in Cunningham, Johnson, Hollings, Kreger, & Jones, 2019).…”