Total, Direct, and Indirect Effects in Logit and Probit Models
The DOI matches its Crossref record; 47 of the 48 references it cites were found in a registry or an index.
Abstract
This article presents a method for estimating and interpreting total, direct, and indirect effects in logit or probit models. The method extends the decomposition properties of linear models to these models; it closes the much-discussed gap between results based on the “difference in coefficients” method and the “product of coefficients” method in mediation analysis involving nonlinear probability models models; it reports effects measured on both the logit or probit scale and the probability scale; and it identifies causal mediation effects under the sequential ignorability assumption. We also show that while our method is computationally simpler than other methods, it always performs as well as, or better than, these methods. Further derivations suggest a hitherto unrecognized issue in identifying heterogeneous mediation effects in nonlinear probability models. We conclude the article with an application of our method to data from the National Educational Longitudinal Study of 1988.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The KHB test was developed for nonlinear dependent variables(Breen et al., 2013), but in STATA it can be applied to linear ones as well, in which case it is the same as the Sobel test.”
From Rape or Homicide: Which Is Worse? (2024) · cites this paper as Breen et al., 2013
In the citing paper
“To avoid the identification anomalies and non-linear scale distortions of logistic mediation, the second stage utilizes a linear probability specification, which permits an exact additive decomposition of conditional indirect effects (Breen et al., 2013) and formal evaluation of the Index of Moderated Mediation (a 3 × b 1 ; H3).”
From Shopping by algorithm: How agentic AI deploys human heuristics as a surrogate consumer (2026) · cites this paper as Breen et al., 2013
Citations
Cites
(1990) · Choice Reviews Online · doi:10.5860/choice.27-3338 · Cited by 35,305
(2001) · RePEc: Research Papers in Economics · Cited by 27,577
Estimating causal effects of treatments in randomized and nonrandomized studies
Rubin, D. B. (1974) · Journal of Educational Psychology · doi:10.1037/h0037350 · Cited by 9,436 · cited 1×
“While most treatments of causal mediation analysis use the potential outcomes framework of Rubin (1974, 1978), we refrain from doing this here and rather use the equivalent expression in terms of conditional independence between covariates and unobservables.”
MethodKupper, L. L., Hosmer, D. W., & Lemeshow, S. (1990) · Journal of the American Statistical Association · doi:10.2307/2290035 · Cited by 7,599
Regression Models for Categorical and Limited Dependent Variables.
MTW, & Long, J. S. (1997) · Journal of the American Statistical Association · doi:10.2307/2965458 · Cited by 6,898
A general approach to causal mediation analysis.
Imai, K., Keele, L., & Tingley, D. (2010) · Psychological Methods · doi:10.1037/a0020761 · Cited by 4,143 · cited 1×
“Using Monte Carlo simulations, Karlson et al. (2012) compared the method we propose with other methods for comparing coefficients across samesample nested logit or probit models.”
MethodA statistical model for the analysis of ordinal level dependent variables
McKelvey, R. D., & Zavoina, W. (1975) · The Journal of Mathematical Sociology · doi:10.1080/0022250x.1975.9989847 · Cited by 2,599
Bayesian inference for causal effects: The role of randomization
Rubin, D. B. (1978) · The Annals of Statistics · doi:10.1214/aos/1176344064 · Cited by 2,503
Statistical Methods for Comparing Regression Coefficients Between Models
Clogg, C. C., Petkova, E., & Haritou, A. (1995) · American Journal of Sociology · doi:10.1086/230638 · Cited by 2,072
The Analysis of Cross-Classified Categorical Data
Drane, J. W. (1979) · Technometrics · doi:10.1080/00401706.1979.10489789 · Cited by 1,941
Estimating Mediated Effects in Prevention Studies
Mackinnon, D. P., & Dwyer, J. H. (1993) · Evaluation Review · doi:10.1177/0193841x9301700202 · Cited by 1,937
Identifiability and exchangeability for direct and indirect effects
Robins, J. M., & Greenland, S. (1992) · Epidemiology · doi:10.1097/00001648-199203000-00013 · Cited by 1,793 · cited 1×
“In recent years, methodologists have criticized mediation analysis for lacking a causal interpretation (e.g., Jo 2008; Pearl 2001; Robins and Greenland 1992; Sobel 2008).”
BackgroundEXPLAINING EDUCATIONAL DIFFERENTIALS: TOWARDS A FORMAL RATIONAL ACTION THEORY
Breen, R., & Goldthorpe, J. H. (1997) · Rationality and Society · doi:10.1177/104346397009003002 · Cited by 1,644
Self-citationIdentification, Inference and Sensitivity Analysis for Causal Mediation Effects
Imai, K., Keele, L., & Yamamoto, T. (2010) · Statistical Science · doi:10.1214/10-sts321 · Cited by 1,628 · cited 1×
“Using Monte Carlo simulations, Karlson et al. (2012) compared the method we propose with other methods for comparing coefficients across samesample nested logit or probit models.”
MethodThe Decomposition of Effects in Path Analysis
Alwin, D. F., & Hauser, R. M. (1975) · American Sociological Review · doi:10.2307/2094445 · Cited by 1,423
Path Analysis: Sociological Examples
Duncan, O. D. (1966) · American Journal of Sociology · doi:10.1086/224256 · Cited by 1,261
Karlson, K. B., Holm, A., & Breen, R. (2012) · Sociological Methodology · doi:10.1177/0081175012444861 · Cited by 1,204 · cited 16×
“Indeed, all three alternative methods discussed in Karlson et al. (2012) are based on estimating different models, thereby reflecting changes in the fit of the error to the assumed distribution.”
BackgroundMethodResultInfluentialSelf-citationComparing Logit and Probit Coefficients Across Groups
ALLISON, P. D. (1999) · Sociological Methods & Research · doi:10.1177/0049124199028002003 · Cited by 1,047
Comparing Coefficients of Nested Nonlinear Probability Models
Kohler, U., Karlson, K. B., & Holm, A. (2011) · The Stata Journal: Promoting communications on statistics and Stata · doi:10.1177/1536867x1101100306 · Cited by 1,041 · cited 1×
“Perhaps most usefully, the method can be applied very easily using the Stata routine khb (Kohler et al. 2011).”
MethodSelf-citationPearl, J. (2022) · Probabilistic and Causal Inference · doi:10.1145/3501714.3501736 · Cited by 1,033
Cameron, S. V., & Heckman, J. J. (1998) · Journal of Political Economy · doi:10.1086/250010 · Cited by 1,028
On Education, opportunity, and social inequality
Ramsöy, N. R. (1975) · Social Science Information · doi:10.1177/053901847501400606 · Cited by 925
Identification, Inference, and Sensitivity Analysis for Causal Mediation Effects
Imai, K., & Keele, L. J. (2008) · Cited by 887
Statistical Methods for Categorical Data Analysis
Powers, D. A. (1999) · Cited by 880
Regression Models with Ordinal Variables
Winship, C., & Mare, R. D. (1984) · American Sociological Review · doi:10.2307/2095465 · Cited by 796 · cited 2×
“They found that, for estimating mediation effects, their method is always as good as or better than the linear probability model, APEs based on the logit or probit (Cramer 2007; Wooldridge 2002), and the method of Y standardization (Long 1997; Winship and Mare 1984).”
BackgroundMethodChmura Kraemer, H., Kiernan, M., Essex, M., & Kupfer, D. J. (2008) · Health Psychology · doi:10.1037/0278-6133.27.2(suppl.).s101 · Cited by 600 · cited 1×
“However, a special case arises when the predictor variable and the mediator variable are interacted (Kraemer et al. 2008).”
BackgroundTotal, Direct, and Indirect Effects in Structural Equation Models
Bollen, K. A. (1987) · Sociological Methodology · doi:10.2307/271028 · Cited by 449
The Analysis of Cross-Classified Categorical Data.
Holt, D., & Fienberg, S. E. (1978) · Journal of the Royal Statistical Society. Series A (General) · doi:10.2307/2344495 · Cited by 420
Specification Error in Probit Models
Yatchew, A., & Griliches, Z. (1985) · The Review of Economics and Statistics · doi:10.2307/1928444 · Cited by 388 · cited 1×
“Thus, including a z orthogonal to x—ensuring that the y–x relationship is not mediated or confounded by z— would change the coefficient of x simply as a consequence of rescaling, as noted by Winship and Mare (1984), Yatchew and Griliches (1985), and Wooldridge (2002).”
BackgroundOn class differentials in educational attainment
Erikson, R., Goldthorpe, J. H., Jackson, M., Yaish, M., & Cox, D. R. (2005) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0502433102 · Cited by 349 · cited 1×
“…effects on educational decisions have received considerable attention, because they link to a theoretical model developed in a classic work on inequality of educational opportunity by Raymond Boudon (1974) and its generalization by Breen and Goldthorpe (1997; see Erikson et al. 2005; Morgan 2012).”
BackgroundDecomposing primary and secondary effects: A new decomposition method
Karlson, K. B., & Holm, A. (2011) · Research in Social Stratification and Mobility · doi:10.1016/j.rssm.2010.12.005 · Cited by 327 · cited 8×
“Indeed, all three alternative methods discussed in Karlson et al. (2012) are based on estimating different models, thereby reflecting changes in the fit of the error to the assumed distribution.”
BackgroundMethodResultInfluentialSelf-citationLogit Models from Economics and Other Fields
Cramer, J. S. (2003) · Cambridge University Press eBooks · doi:10.1017/cbo9780511615412 · Cited by 285
Amemiya, T. (1975) · NBER Chapters · Cited by 284
Modellvergleich und Ergebnisinterpretation in Logit- und Probit-Regressionen
Best, H., & Wolf, C. (2012) · KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie · doi:10.1007/s11577-012-0167-4 · Cited by 210
Structural Equations and Path Analysis for Discrete Data
Winship, C., & Mare, R. D. (1983) · American Journal of Sociology · doi:10.1086/227834 · Cited by 208 · cited 2×
“While these decomposition principles apply to linear models, total effects in logit and other nonlinear binary probability models do not decompose into direct and indirect effects as in linear models (Fienberg 1977; Karlson, Holm, and Breen 2012; MacKinnon and Dwyer 1993; Winship and Mare 1983).”
BackgroundMethodIdentification of Causal Parameters in Randomized Studies With Mediating Variables
Sobel, M. E. (2008) · Journal of Educational and Behavioral Statistics · doi:10.3102/1076998607307239 · Cited by 202
Ambition and Social Class: A Respecification
Keller, S., & Zavalloni, M. (1964) · Social Forces · doi:10.1093/sf/43.1.58 · Cited by 198
The Measurement and Decomposition of Causal Effects in Nonlinear and Nonadditive Models
Stolzenberg, R. M. (1980) · Sociological Methodology · doi:10.2307/270872 · Cited by 189 · cited 2×
“Using the properties of linear models and the basic results of path analysis (see Alwin and Hauser 1975; Duncan 1966; Stolzenberg 1980), we find the well-known results that d ¼ byx byx z ¼ yzx byz x: ð5Þ This result shows that the ‘‘difference in coefficients’’ method is equivalent to the ‘‘product…”
MethodResultCausal inference in randomized experiments with mediational processes.
Jo, B. (2008) · Psychological Methods · doi:10.1037/a0014207 · Cited by 151 · cited 2×
“For inference on the indirect effect, see the significance test in Karlson et al. (2012). Breen et al.”
BackgroundFemlogit—Implementation of the Multinomial Logit Model with Fixed Effects
Pforr, K. (2014) · The Stata Journal Promoting communications on statistics and Stata · doi:10.1177/1536867x1401400409 · Cited by 97
Robustness of Logit Analysis: Unobserved Heterogeneity and Mis‐specified Disturbances*
Cramer, J. S. (2007) · Oxford Bulletin of Economics and Statistics · doi:10.1111/j.1468-0084.2007.00445.x · Cited by 84 · cited 2×
“They found that, for estimating mediation effects, their method is always as good as or better than the linear probability model, APEs based on the logit or probit (Cramer 2007; Wooldridge 2002), and the method of Y standardization (Long 1997; Winship and Mare 1984).”
BackgroundMethodTubeuf, S., Jusot, F., & Bricard, D. (2011) · Health Economics · doi:10.1002/hec.2815 · Cited by 73
Morgan, S. L. (2012) · Sociological Methods & Research · doi:10.1177/0049124112440797 · Cited by 59 · cited 1×
“…effects on educational decisions have received considerable attention, because they link to a theoretical model developed in a classic work on inequality of educational opportunity by Raymond Boudon (1974) and its generalization by Breen and Goldthorpe (1997; see Erikson et al. 2005; Morgan 2012).”
BackgroundPais, J. (2014) · Demography · doi:10.1007/s13524-014-0330-9 · Cited by 56
Myrskylä, M., Elo, I. T., Kohler, I. V., & Martikainen, P. (2014) · Social Science & Medicine · doi:10.1016/j.socscimed.2014.06.008 · Cited by 28
Health Aid and Health Improvement in Sub Saharan AfricaPreprint
Mallaye, D., & Yogo, U. T. (2012) · SSRN Electronic Journal · doi:10.2139/ssrn.2047219 · Cited by 9
Tubeuf, S., Jusot, F., & Bricard, D. (2011) · RePEc: Research Papers in Economics · Cited by 8
Research works printed but not linked: 3
- Amemiya Takeshi, 1975, Annals of Economic and Social Measurement, 4, 363
- Blalock Hubert M., 1979, Social Statistics
- Pearl Judea, 2001, Proceedings of the Seventeenth Conference on Uncertainty and Artificial Intelligence, 411
Cited by
Interpreting and Understanding Logits, Probits, and Other Nonlinear Probability Models
Breen, R., Karlson, K. B., & Holm, A. (2018) · Annual Review of Sociology · doi:10.1146/annurev-soc-073117-041429 · Cited by 468
Self-citationPergelova, A., Manolova, T., Simeonova-Ganeva, R., & Yordanova, D. (2019) · Journal of Small Business Management · doi:10.1111/jsbm.12494 · Cited by 288
“We also decomposed the effect of the three digital capability measures on internationalization into total, direct, and indirect effects (Breen, Karlson, and Holm 2013). this additional analysis (table 3) shows that information management systems have the biggest direct and total effects of all…”
MethodJansen, T., Rademakers, J., Waverijn, G., Verheij, R., Osborne, R., & Heijmans, M. (2018) · BMC Health Services Research · doi:10.1186/s12913-018-3197-4 · Cited by 284
Atheists and Other Cultural Outsiders: Moral Boundaries and the Non-Religious in the United States
Edgell, P., Hartmann, D., Stewart, E., & Gerteis, J. (2016) · Social Forces · doi:10.1093/sf/sow063 · Cited by 251
Friis, K., Lasgaard, M., Rowlands, G., Osborne, R. H., & Maindal, H. T. (2016) · Journal of Health Communication · doi:10.1080/10810730.2016.1201175 · Cited by 228
“This command decomposes the total effect into direct effect (the effect of the independent variable (educational level) on the dependent variable (health behavior) when controlling for mediating variables (health literacy scales) and indirect effect (the effect of the independent variable on the dependent variable thr…”
Background
Metadata
- Authors
- Breen, R., Karlson, K. B., & Holm, A.
- Affiliations
- Yale University (US)
- Aarhus University (DK)
- University of Copenhagen (DK)
- Corresponding author
- Breen, Richard
- Venue
- Sociological Methods & Research
- Publisher
- SAGE Publications
- Published
- 2013-05
- Volume
- 42
- Issue
- 2
- Pages
- 164-191
- ISSN
- 0049-1241 (ISSN-L)
- 1552-8294
- OpenAlex
- W2110453893
- Cited by
- 818
- References
- 48
- Citations by year
- 2026: 64 · 2025: 75 · 2024: 78 · 2023: 81 · 2022: 99 · 2021: 87 · 2020: 73 · 2019: 63 · 2018: 66 · 2017: 48 · 2016: 37 · 2015: 27 · 2014: 14 · 2013: 6
- FWCI
- 24.64
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- journals.sagepub.com/page/policies/text-and-data-mining-license (tdm)
- Locations
- Sociological Methods & Research (publishedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Bayesian Inference
- Social Sciences › Sociology and Political Science › Intergenerational and Educational Inequality Studies
- Keywords
- mediation analysis, logit model, probit model, causal mediation analysis, sequential ignorability
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.