Scaling regression inputs by dividing by two standard deviations

By Andrew Gelman. Statistics in Medicine, 2008

2,405 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/f28d9338c8bd198b90b53042c9d9ac411c5e7586

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

Relationships between structural complexity, coral traits, and reef fish assemblages (2017, methodology): “…Method, 259 εij ~ N (0, σ2 ×|Complexity|2δ) 260 Prior to analysis, all input variables were centered to a mean of zero and 261 standardized to a standard deviation of ± 1, in order to allow for direct comparisons 262 of variable effect sizes (in R, function ‘rescale’ in package arm; Gelman 2008).”
Recovery of amphibian, reptile, bird and mammal diversity during secondary forest succession in the tropics (2019, methodology): “To compare across the predictor variables, each numeric variable was scaled by centering around the mean and dividing by two standard deviations (Gelman 2008).”
Splitting a Predictor at the Upper Quarter or Third and the Lower Quarter or Third (2009, methodology): “5 (Gelman 2008), so we can interpret the main effects and the interaction directly as comparisons between high and low values of the predictors.”
The effect of institutional proximity in non-local university–industry collaborations: An analysis based on Chinese patent data (2013, methodology): “Moreover, to ensure computation stability and to deal with the potential problem of perfect separation, we utilised Bayesian version of logistic regression (Gelman, 2008) to estimate the following models. After obtaining the estimates from the logistic regression in R (R Development Core Team, 2012)”
The importance of project finance for renewable energy projects (2018, methodology): “Finally, the ratios are scaled to ensure regression coefficients are comparable to those of the binary variables (dividing by two standard deviations following the standard approach proposed by Gelman (2008)).”
Simple means to improve the interpretability of regression coefficients (2010, methodology): “Therefore, Gelman (2008) recommended to standardize continuous input variables by dividing two standard deviations (instead of one) by default.”
A brief introduction to mixed effects modelling and multi-model inference in ecology (2018, methodology): “Gelman (2008) provides strong 593 arguments for standardising continuous variables by 2 SDs when binary predictors are 594 in the model.”
Conclusions beyond support: overconfident estimates in mixed models (2008, methodology): “After z-transformation, slopes are estimated as standardized slopes that can vary between 21 and 1 and are equivalent to correlation coefficients (alternatively, Gelman [2008] suggested a standardization by dividing by 2 SDs).”
Accurate First Impressions Leave a Lasting Impression (2013, neutral): “Looking at the effect size estimates (bs), which can be interpreted as equivalent to Cohen’s d (Gelman, 2008), it is clear that positive bias had the strongest associations with Time 1 indicators of relationship development, more than doubling the size of the distinctive assumed similarity and disti”
Does satisfactory neighbourhood environment lead to a satisfying life? An investigation of the association between neighbourhood environment and life satisfaction in Beijing (2017, neutral): “To facilitate the interpretation and comparability of regression coefficients, all of the continuous variables are first mean centred, and then scaled by dividing by twice their standard deviations (Gelman, 2008).”
Review and evaluation of penalised regression methods for risk prediction in low‐dimensional data with few events (2015, background): “[29] suggested centering all predictors and scaling continuous predictors to have standard deviation of 0.”