Moderation (statistics)

Last updated January 30, 2025

In statistics and regression analysis, moderation (also known as effect modification) occurs when the relationship between two variables depends on a third variable. The third variable is referred to as the moderator variable (or effect modifier) or simply the moderator (or modifier).^[1]^[2] The effect of a moderating variable is characterized statistically as an interaction;^[1] that is, a categorical (e.g., sex, ethnicity, class) or continuous (e.g., age, level of reward) variable that is associated with the direction and/or magnitude of the relation between dependent and independent variables. Specifically within a correlational analysis framework, a moderator is a third variable that affects the zero-order correlation between two other variables, or the value of the slope of the dependent variable on the independent variable. In analysis of variance (ANOVA) terms, a basic moderator effect can be represented as an interaction between a focal independent variable and a factor that specifies the appropriate conditions for its operation.^[3]

Example

Moderation analysis in the behavioral sciences involves the use of linear multiple regression analysis or causal modelling.^[1] To quantify the effect of a moderating variable in multiple regression analyses, regressing random variable Y on X, an additional term is added to the model. This term is the interaction between X and the proposed moderating variable.^[1]

Thus, for a response Y and two variables: x₁ and moderating variable x₂,:

Y=b_{0}+b_{1}x_{1}+b_{2}x_{2}+b_{3}(x_{1}\times x_{2})+\varepsilon \,

In this case, the role of x₂ as a moderating variable is accomplished by evaluating b₃, the parameter estimate for the interaction term.^[1] See linear regression for discussion of statistical evaluation of parameter estimates in regression analyses.

Multicollinearity in moderated regression

In moderated regression analysis, a new interaction predictor ( $x_{1}x_{2}$ ) is calculated. However, the new interaction term may be correlated with the two main effects terms used to calculate it. This is the problem of multicollinearity in moderated regression. Multicollinearity tends to cause coefficients to be estimated with higher standard errors and hence greater uncertainty.

Mean-centering (subtracting raw scores from the mean) may reduce multicollinearity, resulting in more interpretable regression coefficients.^[4]^[5] However, it does not affect the overall model fit.

Post-hoc probing of interactions

Like simple main effect analysis in ANOVA, in post-hoc probing of interactions in regression, we are examining the simple slope of one independent variable at the specific values of the other independent variable. Below is an example of probing two-way interactions. In what follows, the regression equation with two variables A and B and an interaction term A*B,

Y=b_{0}+b_{1}A+b_{2}B+b_{3}A*B+\varepsilon

will be considered.^[6]

Two categorical independent variables

If both of the independent variables are categorical variables, we can analyze the results of the regression for one independent variable at a specific level of the other independent variable. For example, suppose that both A and B are single dummy coded (0,1) variables, and that A represents ethnicity (0 = European Americans, 1 = East Asians) and B represents the condition in the study (0 = control, 1 = experimental). Then the interaction effect shows whether the effect of condition on the dependent variable Y is different for European Americans and East Asians and whether the effect of ethnic status is different for the two conditions. The coefficient of A shows the ethnicity effect on Y for the control condition, while the coefficient of B shows the effect of imposing the experimental condition for European American participants.

To probe if there is any significant difference between European Americans and East Asians in the experimental condition, we can simply run the analysis with the condition variable reverse-coded (0 = experimental, 1 = control), so that the coefficient for ethnicity represents the ethnicity effect on Y in the experimental condition. In a similar vein, if we want to see whether the treatment has an effect for East Asian participants, we can reverse code the ethnicity variable (0 = East Asians, 1 = European Americans).

One categorical and one continuous independent variable

If the first independent variable is a categorical variable (e.g. gender) and the second is a continuous variable (e.g. scores on the Satisfaction With Life Scale (SWLS)), then b₁ represents the difference in the dependent variable between males and females when life satisfaction is zero. However, a zero score on the Satisfaction With Life Scale is meaningless as the range of the score is from 7 to 35. This is where centering comes in. If we subtract the mean of the SWLS score for the sample from each participant's score, the mean of the resulting centered SWLS score is zero. When the analysis is run again, b₁ now represents the difference between males and females at the mean level of the SWLS score of the sample.

Cohen et al. (2003) recommended using the following to probe the simple effect of gender on the dependent variable (Y) at three levels of the continuous independent variable: high (one standard deviation above the mean), moderate (at the mean), and low (one standard deviation below the mean).^[7] If the scores of the continuous variable are not standardized, one can just calculate these three values by adding or subtracting one standard deviation of the original scores; if the scores of the continuous variable are standardized, one can calculate the three values as follows: high = the standardized score minus 1, moderate (mean = 0), low = the standardized score plus 1. Then one can explore the effects of gender on the dependent variable (Y) at high, moderate, and low levels of the SWLS score. As with two categorical independent variables, b₂ represents the effect of the SWLS score on the dependent variable for females. By reverse coding the gender variable, one can get the effect of the SWLS score on the dependent variable for males.

Coding in moderated regression

When treating categorical variables such as ethnic groups and experimental treatments as independent variables in moderated regression, one needs to code the variables so that each code variable represents a specific setting of the categorical variable. There are three basic ways of coding: dummy-variable coding, contrast coding and effects coding. Below is an introduction to these coding systems.^[8]^[9]

Effects coding

Dummy coding compares a reference group (one specific condition such as a control group in the experiment) with each of the other experimental groups. In this case, the intercept $b_{0}={\bar {Y}}_{\text{reference}}$ is the mean of the reference group. Each unstandardized regression coefficient $b_{A}={\bar {Y}}_{A}-{\bar {Y}}_{\text{reference}}$ is the difference in the dependent variable mean between a treatment group and the reference group. This coding system is similar to ANOVA analysis.

Contrast coding investigates a series of orthogonal contrasts (group comparisons). The intercept $b_{0}={\overline {{\bar {Y}}_{1\dots n}}}$ is the unweighted mean of the individual group means. The unstandardized regression coefficient $b_{A,B}={\bar {Y}}_{B}-{\bar {Y}}_{A}$ ^{[ verification needed ]} represents the difference between two groups. This coding system is appropriate when researchers have an a priori hypothesis concerning the specific differences among the group means.

Effects coding is used when there is no reference group or orthogonal contrasts. The intercept $b_{0}$ ^{[ cleanup needed ]} is the grand mean (the mean of all the conditions). The regression coefficient $b_{A}={\bar {Y}}_{A}-{\overline {{\bar {Y}}_{1\dots n}}}$ is the difference between one group mean and the mean of all the group means. This coding system is appropriate when the groups represent natural categories.

Two continuous independent variables

If both of the independent variables are continuous, it is helpful for interpretation to either center or standardize the independent variables, X and Z. (Centering involves subtracting the overall sample mean score from the original score; standardizing does the same followed by dividing by the overall sample standard deviation.) By centering or standardizing the independent variables, the coefficient of X or Z can be interpreted as the effect of that variable on Y at the mean level of the other independent variable.^[10]

To probe the interaction effect, it is often helpful to plot the effect of X on Y at low and high values of Z (some people prefer to also plot the effect at moderate values of Z, but this is not necessary). Often values of Z that are one standard deviation above and below the mean are chosen for this, but any sensible values can be used (and in some cases there are more meaningful values to choose). The plot is usually drawn by evaluating the values of Y for high and low values of both X and Z, and creating two lines to represent the effect of X on Y at the two values of Z. Sometimes this is supplemented by simple slope analysis, which determines whether the effect of X on Y is statistically significant at particular values of Z. A common technique for simple slope analysis is the Johnson-Neyman approach.^[11] Various internet-based tools exist to help researchers plot and interpret such two-way interactions.^[12]

Higher-level interactions

The principles for two-way interactions apply when we want to explore three-way or higher-level interactions. For instance, if we have a three-way interaction between A, B, and C, the regression equation will be as follows:

Y=b_{0}+b_{1}A+b_{2}B+b_{3}C+b_{4}A\cdot B+b_{5}A\cdot C+b_{6}B\cdot C+b_{7}A\cdot B\cdot C+\varepsilon .

Spurious higher-order effects

It is worth noting that the reliability of the higher-order terms depends on the reliability of the lower-order terms. For example, if the reliability for variable A is 0.70, the reliability for variable B is 0.80, and their correlation is r = 0.2, then the reliability for the interaction variable A * B is $((0.7\times 0.8)+0.2^{2})/(1+0.2^{2})=0.577$ .^[13] In this case, low reliability of the interaction term leads to low power; therefore, we may not be able to find the interaction effects between A and B that actually exist. The solution for this problem is to use highly reliable measures for each independent variable.

Another caveat for interpreting the interaction effects is that when variable A and variable B are highly correlated, then the A * B term will be highly correlated with the omitted variable A²; consequently what appears to be a significant moderation effect might actually be a significant nonlinear effect of A alone. If this is the case, it is worth testing a nonlinear regression model by adding nonlinear terms in individual variables into the moderated regression analysis to see if the interactions remain significant. If the interaction effect A*B is still significant, we will be more confident in saying that there is indeed a moderation effect; however, if the interaction effect is no longer significant after adding the nonlinear term, we will be less certain about the existence of a moderation effect and the nonlinear model will be preferred because it is more parsimonious.

Moderated regression analyses also tend to include additional variables, which are conceptualized as covariates of no interest. However, the presence of these covariates can induce spurious effects when either (1) the covariate (C) is correlated with one of the primary variables of interest (e.g. variable A or B), or (2) when the covariate itself is a moderator of the correlation between either A or B with Y.^[14]^[15]^[16] The solution is to include additional interaction terms in the model, for the interaction between each confounder and the primary variables as follows:

Y=b_{0}+b_{1}A+b_{2}B+b_{3}C+b_{4}A\cdot B+b_{5}A\cdot C+b_{6}B\cdot C+\varepsilon

Related Research Articles

In statistics, the Pearson correlation coefficient (PCC) is a correlation coefficient that measures linear correlation between two sets of data. It is the ratio between the covariance of two variables and the product of their standard deviations; thus, it is essentially a normalized measurement of the covariance, such that the result always has a value between −1 and 1. As with covariance itself, the measure can only reflect a linear correlation of variables, and ignores many other types of relationships or correlations. As a simple example, one would expect the age and height of a sample of children from a primary school to have a Pearson correlation coefficient significantly greater than 0, but less than 1.

In statistics, the logistic model is a statistical model that models the log-odds of an event as a linear combination of one or more independent variables. In regression analysis, logistic regression estimates the parameters of a logistic model. In binary logistic regression there is a single binary dependent variable, coded by an indicator variable, where the two values are labeled "0" and "1", while the independent variables can each be a binary variable or a continuous variable. The corresponding probability of the value labeled "1" can vary between 0 and 1, hence the labeling; the function that converts log-odds to probability is the logistic function, hence the name. The unit of measurement for the log-odds scale is called a logit, from logistic unit, hence the alternative names. See § Background and § Definition for formal mathematics, and § Example for a worked example.

Analysis of covariance (ANCOVA) is a general linear model that blends ANOVA and regression. ANCOVA evaluates whether the means of a dependent variable (DV) are equal across levels of one or more categorical independent variables (IV) and across one or more continuous variables. For example, the categorical variable(s) might describe treatment and the continuous variable(s) might be covariates (CV)'s, typically nuisance variables; or vice versa. Mathematically, ANCOVA decomposes the variance in the DV into variance explained by the CV(s), variance explained by the categorical IV, and residual variance. Intuitively, ANCOVA can be thought of as 'adjusting' the DV by the group means of the CV(s).

In statistics, an interaction may arise when considering the relationship among three or more variables, and describes a situation in which the effect of one causal variable on an outcome depends on the state of a second causal variable. Although commonly thought of in terms of causal relationships, the concept of an interaction can also describe non-causal associations. Interactions are often considered in the context of regression analyses or factorial experiments.

In statistics, multicollinearity or collinearity is a situation where the predictors in a regression model are linearly dependent.

In statistics, the coefficient of determination, denoted R² or r² and pronounced "R squared", is the proportion of the variation in the dependent variable that is predictable from the independent variable(s).

In statistics, ordinary least squares (OLS) is a type of linear least squares method for choosing the unknown parameters in a linear regression model by the principle of least squares: minimizing the sum of the squares of the differences between the observed dependent variable in the input dataset and the output of the (linear) function of the independent variable. Some sources consider OLS to be linear regression.

Functional data analysis (FDA) is a branch of statistics that analyses data providing information about curves, surfaces or anything else varying over a continuum. In its most general form, under an FDA framework, each sample element of functional data is considered to be a random function. The physical continuum over which these functions are defined is often time, but may also be spatial location, wavelength, probability, etc. Intrinsically, functional data are infinite dimensional. The high intrinsic dimensionality of these data brings challenges for theory as well as computation, where these challenges vary with how the functional data were sampled. However, the high or infinite dimensional structure of the data is a rich source of information and there are many interesting challenges for research and data analysis.

In statistics, multinomial logistic regression is a classification method that generalizes logistic regression to multiclass problems, i.e. with more than two possible discrete outcomes. That is, it is a model that is used to predict the probabilities of the different possible outcomes of a categorically distributed dependent variable, given a set of independent variables.

Multilevel models are statistical models of parameters that vary at more than one level. An example could be a model of student performance that contains measures for individual students as well as measures for classrooms within which the students are grouped. These models can be seen as generalizations of linear models, although they can also extend to non-linear models. These models became much more popular after sufficient computing power and software became available.

Proportional hazards models are a class of survival models in statistics. Survival models relate the time that passes, before some event occurs, to one or more covariates that may be associated with that quantity of time. In a proportional hazards model, the unique effect of a unit increase in a covariate is multiplicative with respect to the hazard rate. The hazard rate at time $is the probability per short time d t that an event will occur between and given that up to time no event has occurred yet. For example, taking a drug may halve one's hazard rate for a stroke occurring, or, changing the material from which a manufactured component is constructed, may double its hazard rate for failure. Other types of survival models such as accelerated failure time models do not exhibit proportional hazards. The accelerated failure time model describes a situation where the biological or mechanical life history of an event is accelerated.$

In statistics, a mediation model seeks to identify and explain the mechanism or process that underlies an observed relationship between an independent variable and a dependent variable via the inclusion of a third hypothetical variable, known as a mediator variable. Rather than a direct causal relationship between the independent variable and the dependent variable, a mediation model proposes that the independent variable influences the mediator variable, which in turn influences the dependent variable. Thus, the mediator variable serves to clarify the nature of the causal relationship between the independent and dependent variables.

In probability theory and statistics, partial correlation measures the degree of association between two random variables, with the effect of a set of controlling random variables removed. When determining the numerical relationship between two variables of interest, using their correlation coefficient will give misleading results if there is another confounding variable that is numerically related to both variables of interest. This misleading information can be avoided by controlling for the confounding variable, which is done by computing the partial correlation coefficient. This is precisely the motivation for including other right-side variables in a multiple regression; but while multiple regression gives unbiased results for the effect size, it does not give a numerical value of a measure of the strength of the relationship between the two variables of interest.

In statistics, the variance inflation factor (VIF) is the ratio (quotient) of the variance of a parameter estimate when fitting a full model that includes other parameters to the variance of the parameter estimate if the model is fit with only the parameter on its own. The VIF provides an index that measures how much the variance of an estimated regression coefficient is increased because of collinearity.

In statistics, principal component regression (PCR) is a regression analysis technique that is based on principal component analysis (PCA). PCR is a form of reduced rank regression. More specifically, PCR is used for estimating the unknown regression coefficients in a standard linear regression model.

In statistics and machine learning, lasso is a regression analysis method that performs both variable selection and regularization in order to enhance the prediction accuracy and interpretability of the resulting statistical model. The lasso method assumes that the coefficients of the linear model are sparse, meaning that few of them are non-zero. It was originally introduced in geophysics, and later by Robert Tibshirani, who coined the term.

In statistics and in machine learning, a linear predictor function is a linear function of a set of coefficients and explanatory variables, whose value is used to predict the outcome of a dependent variable. This sort of function usually comes in linear regression, where the coefficients are called regression coefficients. However, they also occur in various types of linear classifiers, as well as in various other models, such as principal component analysis and factor analysis. In many of these models, the coefficients are referred to as "weights".

In statistics, linear regression is a model that estimates the linear relationship between a scalar response and one or more explanatory variables. A model with exactly one explanatory variable is a simple linear regression; a model with two or more explanatory variables is a multiple linear regression. This term is distinct from multivariate linear regression, which predicts multiple correlated dependent variables rather than a single dependent variable.

In statistics, the class of vector generalized linear models (VGLMs) was proposed to enlarge the scope of models catered for by generalized linear models (GLMs). In particular, VGLMs allow for response variables outside the classical exponential family and for more than one parameter. Each parameter can be transformed by a link function. The VGLM framework is also large enough to naturally accommodate multiple responses; these are several independent responses each coming from a particular statistical distribution with possibly different parameter values.

Functional regression is a version of regression analysis when responses or covariates include functional data. Functional regression models can be classified into four types depending on whether the responses or covariates are functional or scalar: (i) scalar responses with functional covariates, (ii) functional responses with scalar covariates, (iii) functional responses with functional covariates, and (iv) scalar or functional responses with functional and scalar covariates. In addition, functional regression models can be linear, partially linear, or nonlinear. In particular, functional polynomial models, functional single and multiple index models and functional additive models are three special cases of functional nonlinear models.

References

1 2 3 4 5 Cohen, Jacob; Cohen, Patricia; Leona S. Aiken; West, Stephen H. (2003). Applied multiple regression/correlation analysis for the behavioral sciences. Hillsdale, N.J: L. Erlbaum Associates. ISBN 0-8058-2223-2.
↑ Schandelmaier, Stefan; Briel, Matthias; Varadhan, Ravi; Schmid, Christopher H.; Devasenapathy, Niveditha; Hayward, Rodney A.; Gagnier, Joel; Borenstein, Michael; van der Heijden, Geert J.M.G.; Dahabreh, Issa J.; Sun, Xin; Sauerbrei, Willi; Walsh, Michael; Ioannidis, John P.A.; Thabane, Lehana (2020-08-10). "Development of the Instrument to assess the Credibility of Effect Modification Analyses (ICEMAN) in randomized controlled trials and meta-analyses". Canadian Medical Association Journal. 192 (32): E901 –E906. doi:10.1503/cmaj.200077. ISSN 0820-3946. PMC 7829020 . PMID 32778601.
↑ Baron, R. M., & Kenny, D. A. (1986). "The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations", Journal of Personality and Social Psychology, 5 (6), 1173–1182 (page 1174)
↑ Iacobucci, Dawn; Schneider, Matthew J.; Popovich, Deidre L.; Bakamitsos, Georgios A. (2016). "Mean centering helps alleviate "micro" but not "macro" multicollinearity". Behavior Research Methods. 48 (4): 1308–1317. doi: 10.3758/s13428-015-0624-x . ISSN 1554-3528. PMID 26148824.
↑ Olvera Astivia, Oscar L.; Kroc, Edward (2019). "Centering in Multiple Regression Does Not Always Reduce Multicollinearity: How to Tell When Your Estimates Will Not Benefit From Centering". Educational and Psychological Measurement. 79 (5): 813–826. doi:10.1177/0013164418817801. ISSN 0013-1644. PMC 6713984 . PMID 31488914.
↑ Taylor, Alan. "Testing and Interpreting Interactions in Regression-In a Nutshell" (PDF).
↑ Cohen Jacob; Cohen Patricia; West Stephen G.; Aiken Leona S. Applied multiple regression/correlation analysis for the behavioral sciences (3. ed.). Mahwah, NJ [u.a.]: Erlbaum. pp. 255–301. ISBN 0-8058-2223-2.
↑ Aiken L.S., West., S.G. (1996). Multiple regression testing and interpretation (1. paperback print. ed.). Newbury Park, Calif. [u.a.]: Sage Publications, Inc. ISBN 0-7619-0712-2.{{cite book}}: CS1 maint: multiple names: authors list (link)
↑ Cohen Jacob; Cohen Patricia; West Stephen G.; Aiken Leona S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3. ed.). Mahwah, NJ [u.a.]: Erlbaum. pp. 302–353. ISBN 0-8058-2223-2.
↑ Dawson, J. F. (2013). Moderation in management research: What, why, when and how. Journal of Business and Psychology. doi : 10.1007/s10869-013-9308-7.
↑ Johnson, Palmer O.; Fay, Leo C. (1950-12-01). "The Johnson-Neyman technique, its theory and application". Psychometrika. 15 (4): 349–367. doi:10.1007/BF02288864. ISSN 1860-0980. PMID 14797902. S2CID 43748836.
↑ "Interpreting interaction effects".
↑ Busemeyer, Jerome R.; Jones, Lawrence E. (1983). "Analysis of multiplicative combination rules when the causal variables are measured with error". Psychological Bulletin. 93 (3): 549–562. doi:10.1037/0033-2909.93.3.549. ISSN 1939-1455.
↑ Keller, Matthew C. (2014). "Gene × Environment Interaction Studies Have Not Properly Controlled for Potential Confounders: The Problem and the (Simple) Solution". Biological Psychiatry. 75 (1): 18–24. doi:10.1016/j.biopsych.2013.09.006. PMC 3859520 . PMID 24135711.
↑ Yzerbyt, Vincent Y.; Muller, Dominique; Judd, Charles M. (2004). "Adjusting researchers' approach to adjustment: On the use of covariates when testing interactions". Journal of Experimental Social Psychology. 40 (3): 424–431. doi:10.1016/j.jesp.2003.10.001.
↑ Hull, Jay G.; Tedlie, Judith C.; Lehn, Daniel A. (1992). "Moderator Variables in Personality Research: The Problem of Controlling for Plausible Alternatives". Personality and Social Psychology Bulletin. 18 (2): 115–117. doi:10.1177/0146167292182001. ISSN 0146-1672. S2CID 145366173.

Hayes, A. F., & Matthes, J. (2009). "Computational procedures for probing interactions in OLS and logistic regression: SPSS and SAS implementations." Behavior Research Methods, Vol. 41, pp. 924–936.

This page is based on this Wikipedia article
Text is available under the CC BY-SA 4.0 license; additional terms may apply.
Images, videos and audio are available under their respective licenses.

[cohen-1] 1 2 3 4 5 Cohen, Jacob; Cohen, Patricia; Leona S. Aiken; West, Stephen H. (2003). Applied multiple regression/correlation analysis for the behavioral sciences. Hillsdale, N.J: L. Erlbaum Associates. ISBN 0-8058-2223-2.

[2] Schandelmaier, Stefan; Briel, Matthias; Varadhan, Ravi; Schmid, Christopher H.; Devasenapathy, Niveditha; Hayward, Rodney A.; Gagnier, Joel; Borenstein, Michael; van der Heijden, Geert J.M.G.; Dahabreh, Issa J.; Sun, Xin; Sauerbrei, Willi; Walsh, Michael; Ioannidis, John P.A.; Thabane, Lehana (2020-08-10). "Development of the Instrument to assess the Credibility of Effect Modification Analyses (ICEMAN) in randomized controlled trials and meta-analyses". Canadian Medical Association Journal. 192 (32): E901 –E906. doi:10.1503/cmaj.200077. ISSN 0820-3946. PMC 7829020 . PMID 32778601.

[3] Baron, R. M., & Kenny, D. A. (1986). "The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations", Journal of Personality and Social Psychology, 5 (6), 1173–1182 (page 1174)

[4] Iacobucci, Dawn; Schneider, Matthew J.; Popovich, Deidre L.; Bakamitsos, Georgios A. (2016). "Mean centering helps alleviate "micro" but not "macro" multicollinearity". Behavior Research Methods. 48 (4): 1308–1317. doi: 10.3758/s13428-015-0624-x . ISSN 1554-3528. PMID 26148824.

[5] Olvera Astivia, Oscar L.; Kroc, Edward (2019). "Centering in Multiple Regression Does Not Always Reduce Multicollinearity: How to Tell When Your Estimates Will Not Benefit From Centering". Educational and Psychological Measurement. 79 (5): 813–826. doi:10.1177/0013164418817801. ISSN 0013-1644. PMC 6713984 . PMID 31488914.

[6] Taylor, Alan. "Testing and Interpreting Interactions in Regression-In a Nutshell" (PDF).

[7] Cohen Jacob; Cohen Patricia; West Stephen G.; Aiken Leona S. Applied multiple regression/correlation analysis for the behavioral sciences (3. ed.). Mahwah, NJ [u.a.]: Erlbaum. pp. 255–301. ISBN 0-8058-2223-2.

[8] Aiken L.S., West., S.G. (1996). Multiple regression testing and interpretation (1. paperback print. ed.). Newbury Park, Calif. [u.a.]: Sage Publications, Inc. ISBN 0-7619-0712-2.{{cite book}}: CS1 maint: multiple names: authors list (link)

[9] Cohen Jacob; Cohen Patricia; West Stephen G.; Aiken Leona S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3. ed.). Mahwah, NJ [u.a.]: Erlbaum. pp. 302–353. ISBN 0-8058-2223-2.

[10] Dawson, J. F. (2013). Moderation in management research: What, why, when and how. Journal of Business and Psychology. doi : 10.1007/s10869-013-9308-7.

[11] Johnson, Palmer O.; Fay, Leo C. (1950-12-01). "The Johnson-Neyman technique, its theory and application". Psychometrika. 15 (4): 349–367. doi:10.1007/BF02288864. ISSN 1860-0980. PMID 14797902. S2CID 43748836.

[12] "Interpreting interaction effects".

[13] Busemeyer, Jerome R.; Jones, Lawrence E. (1983). "Analysis of multiplicative combination rules when the causal variables are measured with error". Psychological Bulletin. 93 (3): 549–562. doi:10.1037/0033-2909.93.3.549. ISSN 1939-1455.

[14] Keller, Matthew C. (2014). "Gene × Environment Interaction Studies Have Not Properly Controlled for Potential Confounders: The Problem and the (Simple) Solution". Biological Psychiatry. 75 (1): 18–24. doi:10.1016/j.biopsych.2013.09.006. PMC 3859520 . PMID 24135711.

[15] Yzerbyt, Vincent Y.; Muller, Dominique; Judd, Charles M. (2004). "Adjusting researchers' approach to adjustment: On the use of covariates when testing interactions". Journal of Experimental Social Psychology. 40 (3): 424–431. doi:10.1016/j.jesp.2003.10.001.

[16] Hull, Jay G.; Tedlie, Judith C.; Lehn, Daniel A. (1992). "Moderator Variables in Personality Research: The Problem of Controlling for Plausible Alternatives". Personality and Social Psychology Bulletin. 18 (2): 115–117. doi:10.1177/0146167292182001. ISSN 0146-1672. S2CID 145366173.

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]