Increasingly complex models are being fit to data these days. This is especially the case for Bayesian modelling making use of Markov chain Monte Carlo methods. Tailored model diagnostics are usually lacking behind. This is also the case for Bayesian mediation models. In this paper, we developed a method for the detection of influential observations for a popular mediation model and its extensions in a Bayesian context. Detection of influential observations is based on the case-deletion principle. Importance sampling with weights which take advantage of the dependence structure in hierarchical models is utilized in order to identify the part of the model which is influenced most. We make use of the variance of log importance sampling weights as the measure of influence. It is demonstrated that this approach is useful when interest lies in the impact of individual observations on a subset of model parameters. The method is illustrated on a three-level data set from the field of nursing research, which was previously used to fit a mediation model of patient satisfaction with care. We focused on influential cases on both the second and the third level of the data.

Department of Probability and Mathematical Statistics Faculty of Mathematics and Physics Charles University Prague Czech Republic

Leuven Biostatistics and Statistical Bioinformatics Centre University of Leuven Leuven Belgium

Leuven Institute for Healthcare Policy University of Leuven Leuven Belgium

Zobrazit více v PubMed

Bayarri M.J. and Morales J., Bayesian measures of surprise for outlier detection, J. Statist. Plann. Inference. 111 (2003), pp. 3–22. doi: 10.1016/S0378-3758(02)00282-3 DOI

Bradlow E.T. and Zaslavsky A.M., Case influence analysis in Bayesian inference, J. Comput. Graph. Stat. 6 (1997), pp. 314–331.

Cook R.D., Assessment of local influence, J. R. Stat. Soc. Ser. B (Methodol.) 48 (1986), pp. 133–169.

Cook R.D. and Weisberg S., Residuals and Influence in Regression, New York, Chapman and Hall, 1982.

Hodges J.S., Some algebra and geometry for hierarchical models, applied to diagnostics, J. R. Stat. Soc. Ser. B (Methodol.) 60 (1998), pp. 497–536. doi: 10.1111/1467-9868.00137 DOI

Jackson D., White I.R., and Carpenter J., Identifying influential observations in Bayesian models by using Markov chain Monte Carlo, Stat. Med. 31 (2012), pp. 1238–1248. doi: 10.1002/sim.4356 PubMed DOI PMC

Kass R.E., Tierney L., and Kadane J.B., Approximate methods for assessing influence and sensitivity in Bayesian analysis, Biometrika 76 (1989), pp. 663–674. doi: 10.1093/biomet/76.4.663 DOI

McCulloch R.E., Local model influence, J. Amer. Statist. Assoc. 84 (1989), pp. 473–478. doi: 10.1080/01621459.1989.10478793 DOI

Millar R.B. and Stewart W.S., Assessment of locally influential observations in Bayesian models, Bayesian Anal. 2 (2007), pp. 365–383. doi: 10.1214/07-BA216 DOI

Peruggia M., On the variability of case-deletion importance sampling weights in the bayesian linear model, J. Amer. Statist. Assoc. 92 (1997), pp. 199–207. doi: 10.1080/01621459.1997.10473617 DOI

Preacher K.J., Rucker D.D., and Hayes A.F., Addressing moderated mediation hypotheses: Theory, methods, and prescriptions, Multivariate Behav. Res. 42 (2007), pp. 185–227. doi: 10.1080/00273170701341316 PubMed DOI

Robert C.P. and Casella G., Monte Carlo Statistical Methods, 2nd ed, Springer-Verlag, New York, 2004.

Rusá Š., Komárek A., Lesaffre E., and Bruyneel L., Multilevel moderated mediation model with ordinal outcome, Stat. Med. 37 (2018), pp. 1650–1670. doi: 10.1002/sim.7605 PubMed DOI

Sermeus W., Aiken L.H., Van den Heede K., Rafferty A.M., Griffiths P., Moreno-Casbas M.T., Busse R., Lindqvist R., Scott A.P., Bruyneel L., Brzostek T., Kinnunen J., Schubert M., Schoonhoven L., and Zikos D., Nurse forecasting in Europe (RN4CAST): Rationale, design and methodology, BMC. Nurs. 10 (2011), pp. 1–9. doi: 10.1186/1472-6955-10-6 PubMed DOI PMC

Thomas Z.M., MacEachern S.N., and Peruggia M., Reconciling curvature and importance sampling based procedures for summarizing case influence in Bayesian models, J. Amer. Statist. Assoc. 113 (2018), pp. 1669–1683. doi: 10.1080/01621459.2017.1360777 DOI

VanderWeele T.J., A three-way decomposition of a total effect into direct, indirect, and interactive effects, Epidemiology 24 (2013), pp. 224–232. doi: 10.1097/EDE.0b013e318281a64e PubMed DOI PMC

Weiss R., An approach to Bayesian sensitivity analysis, J. R. Stat. Soc. Ser. B (Methodol.) 58 (1996), pp. 739–750.

Weiss R.E. and Cho M., Bayesian marginal influence assessment, J. Statist. Plann. Inference. 71 (1998), pp. 163–177. doi: 10.1016/S0378-3758(98)00015-9 DOI

Zaslavsky A.M. and Bradlow E.T., Posterior predictive outlier detection using sample reweighting, J. Comput. Graph. Stat. 19 (2010), pp. 790–807. doi: 10.1198/jcgs.2010.08141 DOI