You are on page 1of 7

ORIGINAL ARTICLE

The Mediation Proportion
A Structural Equation Approach for Estimating the Proportion of Exposure Effect on Outcome Explained by an Intermediate Variable
Susanne Ditlevsen,* Ulla Christensen,† John Lynch,‡ Mogens Trab Damsgaard,§ and Niels Keiding*

Abstract: It is often of interest to assess how much of the effect of an exposure on a response is mediated through an intermediate variable. However, systematic approaches are lacking, other than assessment of a surrogate marker for the endpoint of a clinical trial. We review a measure of “proportion explained” in the context of observational epidemiologic studies. The measure has been much debated; we show how several of the drawbacks are alleviated when exposures, mediators, and responses are continuous and are embedded in a structural equation framework. These conditions also allow for consideration of several intermediate variables. Binary or categorical variables can be included directly through threshold models. We call this measure the mediation proportion, that is, the part of an exposure effect on outcome explained by a third, intermediate variable. Two examples illustrate the approach. The first example is a randomized clinical trial of the effects of interferon-␣ on visual acuity in patients with age-related macular degeneration. In this example, the exposure, mediator and response are all binary. The second example is a common problem in social epidemiology—to find the proportion of a social class effect on a health outcome that is mediated by psychologic variables. Both the mediator and the response are composed of several ordered categorical variables, with confounders present. Finally, we extend the example to more than one mediator. (Epidemiology 2005;16: 114 –120)

Supplemental material for this article is available with the online version of the Journal at www.epidem.com. Submitted 21 November 2002; final version accepted 16 September 2004. From the *Department of Biostatistics, †Department of Social Medicine, and §Department of Social Medicine, Institute of Public Health, University of Copenhagen, Denmark; and ‡Department of Epidemiology, University of Michigan, Ann Arbor. Supported by the Danish Research Councils (Ref. no. 9801268). Correspondence: Susanne Ditlevsen, Department of Biostatistics, Panum Institute, Blegdamsvej 3, DK-220, Copenhagen N, Denmark. E-mail: S.Ditlevsen@pubhealth.ku.dk Copyright © 2004 by Lippincott Williams & Wilkins ISSN: 1044-3983/05/1601-0114 DOI: 10.1097/01.ede.0000147107.76079.07

question often arising in epidemiology is how much of the effect of an exposure on outcome is explained by a third, intermediate variable. For example, a common problem in social epidemiology involves estimating how much of the effect of social class on a given health outcome is mediated by psychologic or lifestyle factors. The problem also arises in clinical trials when using surrogate markers, also called intermediate endpoints, to assess effects of interventions or possible risk factors. An example of this scenario is the question of how much of the effect of a cholesterol-lowering drug on coronary heart disease can be assessed by the serum cholesterol level at a point between administration of the drug and the onset of the disease.1 A further example is the extent to which human papillomavirus infection mediates the relationship between number of sexual partners and cervical dysplasia.2– 4 As always, the causal direction is most obvious in longitudinal studies (cause before effect); in cross-sectional studies, the causal pathway has to be postulated. The method we propose applies to both situations. Current terminology defines an intermediate variable as one lying at least in part on the causal pathway between exposure and response. This is in contrast to a confounder, for which the effects of exposure should usually be corrected. As emphasized by current texts in epidemiology,5 it is misleading to correct for an intermediate variable; however, as a result, intermediate variables often are ignored. It can nonetheless be useful to ask how much of the effect of the exposure variable A on the response C has been mediated by an intermediate variable B. The usual approach to this question is to compare a multiple regression model containing B with one without, where B could represent multiple mediators. This is achieved by comparing differences in either effect measures or overall model fit. This approach was used by Lynch et al6 in a study of acute myocardial infarction using Cox proportional hazard models. To assess the impact of risk factor adjustment on the age-adjusted relative hazard (RH), the proportion of excess relative risk accounted for by risk adjustment was calculated
Epidemiology • Volume 16, Number 1, January 2005

A

114

8. or at least that the conditional distribution of (mediator. mediators. P͑C ϭ 1͉A ϭ 0. for example.1 We then embed the joint normal distribution as basic latent variables in a structural equation model. Validation of surrogate markers in clinical trials has enjoyed a lively methodological debate in the last decade. in this case. B ϭ 1͒ P͑B ϭ 1͉A ϭ 1͒ Wang and Taylor9 also noted that their discussion considerably simplifies when B and C are jointly normal. and response. where model B included the risk factors. allowing for measurement error in the explanatory as well as the dependent variables. Calculation of confidence intervals for the mediation proportion also becomes routine. The contribution of these explanatory factors was measured by the percentage reduction in the odds ratios (ORs) of childhood socioeconomic groups compared with the first model. The assumption of normal distribution of the latent variable is primarily a technical device with no intention of restricting the practical application and interpretation. indeed. A similar method was used by van de Mheen and colleagues7 in a longitudinal study investigating the influence of childhood socioeconomic conditions and selection processes on adult inequalities in health. The structural equation models are regression equations with less-restrictive assumptions. but there is a considerable technical gain in initially assuming that all 3 variables (exposures. The exposition in this article acknowledges the conventional preference in epidemiology for discrete measures of exposure. defined in its simplest version for binary A. The numerator expresses that part of the treatment effect that is mediated through B. In formulas. their proposed measure coincides with that of Freedman et al. Explanatory factors were added to a logistic regression model with childhood or adult socioeconomic level and confounders only. A key reference is by Freedman et al1 These authors described the dependence of a binary effect measure C on a discrete surrogate marker B and a binary exposure variable A by an additive (no-interaction) logistic model logit P͑C ϭ 1͉A ϭ a. B. mediation. The central research question often involves quantities (“latent variables”) that are not directly observable. The models consist of factor analyses that permit direct and indirect effects between factors. which expresses what the probability of success when exposure A ϭ 0 would be if the values of the surrogate B are distributed as those in the exposure group A ϭ 1. and C as P͑C ϭ 1͉A ϭ 0͒ Ϫ S P͑C ϭ 1͉A ϭ 0͒ Ϫ P͑C ϭ 1͉A ϭ 1͒ ϭ 0) of success when exposure A ϭ 0 and (2) the probability S. or cynical hostility. namely the difference between (1) the probability P(C ϭ 1͉A © 2004 Lippincott Williams & Wilkins 115 . Wang and Taylor9 proposed a measure of proportion explained. The assumptions of normally distributed variables underlying the observed ordered categorical variables in the structural equation models through threshold models avoid problems of more or less arbitrary construction of scales and selection of cut-points.1 The surrogate marker discussion originated in clinical trials in which the exposure measure is binary. we propose to measure the mediating effect by the indirect effect as a fraction of the total effect. In the case of normally distributed variables. B ϭ b͒ ϭ ␮ ϩ ␣a ϩ ␣b where the regression coefficient ␣a is the usual log odds ratio of the effect of A on C corrected for B. and defined the proportion of the exposure effect explained by the intermediate end point as (1 Ϫ ␣a/␣Јa). response) given exposure is bivariate normal. This allows for the (possibly ordered) categorical variables. Apparently. that are so commonly encountered in epidemiology. There are various advantages of using structural equation models. confounding. usually by insight or postulate Fϭ where the denominator is the treatment effect on the probability scale disregarding the surrogate marker B. to be estimated by replacing ␣a and ␣Јa by their usual maximum likelihood estimates. We will call this measure the mediation proportion. In that context. B ϭ 0͒ P͑B ϭ 0͉A ϭ 1͒ ϩ P͑C ϭ 1͉A ϭ 0. Number 1. we include some introductory exposition of this methodology. They compared that with a logistic model logit P͑C ϭ 1͉A ϭ a͒ ϭ ␮Ј ϩ ␣Јa for the uncorrected effect. Freedman et al1 acknowledged the nonuniqueness of this measure due to the arbitrary choice of the logistic link function and the arbitrary choice of the log OR as effect measure. This article addresses situations in which the causal pathway already is specified.9 Buyse and Molenberghs10 pointed out that when B and C are jointly normally distributed. January 2005 The Mediation Proportion using the formula (RHmodel A Ϫ RHmodel B)/(RHmodel A Ϫ 1). We shall arrive at methods covering these common situations. response) are jointly normally distributed. possibly but not necessarily measured with error. visual acuity. they did not find it problematic that the 2 above-mentioned models cannot be mathematically consistent (a mixture of logistic regressions is not itself a logistic regression) except in special situations. the analysis of the proportion explained in Freedman et al1 corresponds to elementary analysis of a bivariate normal distribution.Epidemiology • Volume 16. general health. our measure coincides with those of Wang and Taylor9 and Freedman et al. The formula used was (ORmodel A Ϫ ORmodel B)/(ORmodel A Ϫ 1). however. Because structural equations are not widely used in epidemiologic practice.

Because the averaging operations across omitted variables are always linear. cynical hostility. a unit increase in A would cause an increase of ␥1 ؉ ␥2␥3 in the expectation of C where ␥2␥3 is the part that goes through B. The random variables are either latent. and we illustrate by the analysis of the clinical trial from ophthalmology and the cross-sectional survey of social class and health. © 2004 Lippincott Williams & Wilkins MEDIATION PROPORTION Assume exposure A. This example. hence. It is well known that social class affects health. and response C. Number 1. 116 . The measurement model consists of the structural equations that describe the relations between latent and observed variables. That is. We want to decompose ␤ into a sum of a “direct” effect of A on C and an “indirect” effect of A on C via B.Ditlevsen et al Epidemiology • Volume 16. and possibly aided by the natural temporal order of events. The system of structural equations is divided into 2 parts: the measurement model and the model for the latent variables. ␤. Diagram illustrating the regressions in the model of exposure A. We need a framework that allows for confounders as well as for measurement errors on the exposure and intermediate variables and permits generalizations to several intermediate variables and exposures. In particular. Now the effect of A on B is expressed by the regression B ϭ ␥2A and the effects of A and B on C by the regression C ϭ ␥1A ϩ ␥3B (Fig. we do not consider time-dependent confounders that change from confounder to mediator depending on the temporal order of events. intermediate variable B. We define the mediation proportion as the dimensionless proportion of the effect of A on C mediated through B. available with the electronic version of this article. A more detailed description can be found in several references.9 The methodology is further illustrated using the study in social epidemiology that initiated our interest in this concept. intermediate variable B. No normality assumption is necessary for the exposure A as it enters the analysis only as a regressor. and symptom load are such latent variables and are assumed normally distributed.16 –19 The model consists of a system of structural equations containing random variables and structural parameters.12 and by Cox and Wermuth. was earlier discussed by Buyse and Molenberghs10 and by Wang and Taylor.14 The methodology is first illustrated on a randomized clinical trial15 in ophthalmology studying the effects of interferon-␣ on visual acuity in patients with age-related macular degeneration. not all variables of interest are continuous or normally distributed. and response C. Latent variables relate to concepts and are thus hypothetical variables that are not directly observed. Visual acuity.” We calculate the mediation proportion and finally enlarge the model also to include the intermediate variable self-efficacy. Obviously the interpretation is easiest when all entering regression coefficients ␥1␥2␥3 and. Then ␤ ϭ ␥1 ϩ ␥2␥3 (see Appendix A. self-efficacy. The effect of exposure A on response C would usually be reported by calculating the regression of C on A: C ϭ ␤A. January 2005 based on substantive considerations. as defined in the scale of the underlying linear normal statistical models. illustrating how the method applies to several pathways within the same model.11 Finally. STRUCTURAL EQUATION MODELS Structural equation models are generalizations of linear regression models. we instead measure other variables (termed “indicators”) that we assume are correlated with the latent variable. In epidemiologic practice. They allow latent concepts (such as health. observed or “disturbance terms. The question asked here is how much of this effect is mediated through the psychologic variable “cynical hostility. for the derivation) so that we have FIGURE 1. The indicators are related to the latent variables through factor analytic models. In the remainder of this work. in which the conditional joint distribution of B and C is bivariate normal. by items in a questionnaire or clinical measurements. In the case of cynical hostility these are typically items from a questionnaire. we show how structural equation models may contribute to handling these generalizations. are positive. We briefly outline the idea behind the structural equation models. obtained the desired decomposition of the total effect ␤ of A on C into the direct effect ␥1 and the indirect effect ␥2␥3. visual acuity. where the 3 entering variables are all binary. 1). Because we cannot measure a latent variable. nor do we consider variables that are partially caused by the exposure and correlated with the outcome. cynical hostility. or power) that usually are only indirectly observed. for example. (1) The mediation proportion is the percentage change of the regression coefficients when we include an intermediate variable in the model. intelligence. or Indirect effect Total effect ϭ ␥2 ␥3 ␥1 ϩ ␥2 ␥3 . we make the basic assumption of no effect modification. A technical advantage of reasoning in the framework of the underlying models is that these allow the study of the situation with and without intermediate variables within a single statistical model.” with the latter modeling the variation among subjects. we are less prone to the many problems and paradoxes so carefully discussed for the binary-binary-binary case by Robins and Greenland.13 and followed up in the modern theory of graphical models.

The data are presented in Table 1. Err. and poorer visual acuity at 6 months increases the risk of losing more visual acuity at 1 year. Thus. available with the electronic version of this article). The idea is to superimpose the distribution of answers in each category onto a normal distribution.878 (95% confidence interval ͓CI͔ ϭ 0. otherwise.050 0. but other types of indicators such as ordered categorical variables (the typical response from a questionnaire) can be handled within the framework of structural equation models by threshold models (see Appendix B.9.201– 1. Number 1. Results The model was fitted using M-plus20 (see Appendix C. Regression Estimates in Structural Equation Model for Ophthamology Data Param Estimate 0. respectively. which uses the conventional notation of structural equation models (squares symbolize measured variables and ovals symbolize latent variables). January 2005 The Mediation Proportion In the simplest description the indicators are assumed to be normally distributed.878 { if patient randomized to placebo. The response is whether the patient had lost at least 3 lines of vision at 1 year. is 0. we first reanalyze an example used by both Buyse and Molenberghs10 and Wang and Taylor.727 to 0. if patient had lost less than three lines of vision at one year. the visual acuity at 6 months is a good surrogate marker for the visual acuity at 1 year.827 ϭ 0.796) (0.435 ⅐ 0.248 to 0. otherwise. Visual acuity is assumed to vary continuously. Contrary to what was hypothesized from pilot studies. available with the electronic version of this article): Indirect effect Total effect ϭ 0. and T variables. Structural equation model for relations between visual acuity at 1 year and at 6 months. we connect the latent variables to the measurement model through the indicators to estimate the structural parameters (routinely done in standard software such as M-plus.435 ⅐ 0.15 The treatment group received interferon-␣. 0.152 0. ␥3 B ϩ ␥2 A ϩ ␨1 ␥1 A ϩ ␨2 (2) where ␥i are regression coefficients and ␨j error terms. see Appendix A.827 0. MECOSA or SAS).184 0. All estimates are positive: a person exposed to interferon-␣ has higher risk of losing visual acuity at both 6 months and at 1 year. if patient randomized to interferon-␣. The exposure variable is binary and directly measured. A patient’s visual acuity was assessed through the ability to read lines on a vision chart.051 (95% CI) (Ϫ0. and the exposure is whether the patient received placebo or interferon-␣.348) (0. © 2004 Lippincott Williams & Wilkins TABLE 2.Epidemiology • Volume 16. and exposure to interferon-␣. The model is illustrated in Figure 2. The mediation proportion of the effect of interferon-␣ on visual acuity at 1 year. delta-method. and we describe it as normally distributed. B C ϭ ϭ TABLE 1. APPLICATION TO OPHTHALMOLOGY DATA To illustrate the concepts. { { if patient had lost less than two lines of vision at six months. interferon-␣ showed no benefit and even a detrimental effect. measured indirectly through loss of ability to read lines on a vision chart.9 This is a randomized clinical trial in ophthalmology on the effects of interferon-␣ in patients with age-related macular degeneration.074 to 0. and the control group received placebo. The intermediate variable is loss of at least 2 lines of vision at 6 months. Note that the model allows for random variation in the intermediate variable B. S.15 Adopting the notation used in earlier articles.10 the variables are defined as 0 Zϭ 1 0 Sϭ 1 0 Tϭ 1 FIGURE 2. available with the electronic version of this article). Data From a Randomized Clinical Trial in Ophthalmology* Z 0 0 1 1 S 0 1 0 1 T‫؍‬0 56 8 31 9 T‫؍‬1 9 30 9 38 *See text for meaning of Z. as well as controlling for confounders. Estimates are listed in Table 2.435 0. mediated through visual acuity at 6 months. The model for the latent variables consists of a system of linear regression models.827 Std.927) ␥1 ␥2 ␥3 117 .555. Because nothing is observed in the model for the latent variables.050 ϩ 0.

mediator (cynical hostility). The model was fitted using M-plus. The Mediation Proportion of One Intermediate Variable Consider model (2).y13 and social class are included in the model using the threshold models described in Appendix B (available with the electronic version of this article) because they are all ordered categorical.121 ϩ 0.Ditlevsen et al Epidemiology • Volume 16. social relations.207 0.207 THE EFFECT OF SOCIAL CLASS ON SYMPTOM LOAD MEDIATED THROUGH CYNICAL HOSTILITY The study that triggered our interest in the problem is based on a random sample of 40.. has had more symptoms the last 4 weeks.x8. and cynical hostility. Structural equation model for relations between symptom load. it seems more natural to base the model on a continuous variable. affective. and B is cynical hostility.. sickness benefits. Wang and Taylor assumes it can be approximated by a binary variable.445 (Ϫ0. if the underlying data-generating mechanism of a binary or a categorical variable is believed to be continuous. If visual acuity is thought to deteriorate smoothly over the course of time. Cynical hostility is measured by the 8 items x1 Ϫ x8 and symptom load by the 13 items y1 Ϫ y13. The discrepancy between the mediation proportion and Wang and Taylor’s measure in this example stems from the underlying assumptions. We focus on determining the role of cynical hostility (defined as a persistent negative attitude toward others involving cognitive.210 ⅐ 0.264 (0. response (self-reported symptom load). A full substantive analysis would usually include more variables in more complex modeling.. respectively.30 to 4. and psychologic factors. This sample is part of the Danish Longitudinal Study on Work. y1. January 2005 For comparison. The model is illustrated in Figure 3.. with an additional social class VI representing people on transfer income.17 to 3. We restrict attention to exposure (social class). Freedman’s measure for proportion of treatment explained was P ϭ 0.264 (3) Assuming the postulated causal pathways. Age is included as a confounder in all entering regression models. Estimates are listed in Table 3. In general. and has lower socioeconomic position. The classification is similar to the British Registrar General’s Classification I-V. we find the description by a continuous variable more appropriate. disability pension. including unemployment benefits.and 50-year-old individuals from the Danish general population. Social class is considered as directly observed.35). Our question was simple: How much of the effect of social class on symptom load is mediated through cynical hostility? ϭ ϭ 0. and social class means that the person is feeling more hostile. which included variables on demographics. available with the electronic version of this paper).. Symptom load is measured by 13 questions on physiological and mental symptoms within the last 4 weeks. social class.20 (see Appendix C. health behavior.21–24 Social class based on occupation was coded in accordance with the standards of the Danish National Institute of Social Research. Suppose A is social class.210 ⅐ 0. and confounders (age and sex). FIGURE 3. Unemployment and Health. A higher level of cynical hostility. and other social benefits. The analyses were performed stratified by sex and adjusted for age.. occupational health. symptom load.218 – 0.. The data are based on a postal questionnaire. We assume that visual acuity can be approximated by a normal variable. and is drawn from the Institute of Local Government Studies in Denmark’s longitudinal register at Statistics Denmark. © 2004 Lippincott Williams & Wilkins 118 . social capital. as the analyses are merely for illustration.12). socioeconomic factors.315) Indirect effect Total effect 0. and behavioral components) as a step in the pathway between social class and self-reported symptom load.690 (0.10 and Wang and Taylor’s measure was F ϭ 0. Cynical hostility was measured by 8 items derived from the Cook-Medley Hostility Scale.9 Our analysis indicates a stronger role of the pathway via the mediator than previous analyses. we would conclude that around a fourth of the effect of social class on symptom load can be explained for men by the fact that lower social class increases risk of being hostile and that being hostile increases risk of poor health.9. somatic and mental health. Number 1. C is symptom load. Symptom load and cynical hostility are modeled as latent variables. The mediation proportion of the effect of social class on symptom load mediated through cynical hostility is 0.. Only results for men are shown. The variables x1. The arrows connecting the error terms of the indicator variables are explained in Appendix B (available in the electronic version of this article).

0046 (95% CI) (0. Structural equation model for relations between symptom load.220 0. we include a second intermediate variable. Social Class. where the relative size of the regression coefficients along the various pathways measures the effect of an intermediate variable on the association of interest.297) (0.013 0. The model is illustrated in Figure 4. and self-efficacy. The total effect of social class on symptom load can be expressed as Total effect ϭ Direct effect ϩ Indirect effects ϭ ␥6 ϩ ␥1␥2␥4ϩ␥1␥5ϩ␥3␥4 The indirect effect of social class through self-efficacy.011 0. we can quantify how much of the effect is mediated through that third variable by dividing the indirect effect by the total effect.021 (95% CI) (0.099–0. Number 1.25 It is included as a latent variable. Estimates of the Mediated Effects for Men FIGURE 4. 4). we would conclude that approximately a fourth of the effect of social class on symptom load is mediated for men through lack of self-efficacy.010–0. In the larger model the effect of social class on symptom load is the sum of one direct and 3 indirect effects.028 0.232) (0. Then. Given the model.189–0. we quantify a given pathway from one variable to another by multiplying the regression coefficients in each individual path contained in that pathway. The mediation proportions of the effect of social class on symptom load mediated through indirect pathways are listed in Table 4.207 0. © 2004 Lippincott Williams & Wilkins Pathway Self-efficacy Cynical hostility Both Estimate 0.210 Only Results for Men are Shown. as illustrated in Figure 4. but not through cynical hostility. Regression Estimates in Structural Equation Model for Symptom Load. cynical hostility.182–0. The total effect of one variable on another is the sum of all pathways leading from the former to the latter variable. Std. Err.028) 119 . January 2005 The Mediation Proportion TABLE 3.275) (0. together forming all the pathways from social class to the response health. The approach is through structural equation models.165–0.Epidemiology • Volume 16. social class. The arrows connecting the error terms of the indicator variables are explained in Appendix B (available in the electronic version of this article).121 0. Within this model. and approximately 2% mediated first through lack of self-efficacy and then through hostility. is the product of the 2 regression coefficients on that pathway(␥1␥5. DISCUSSION The mediation proportion provides an estimate of how much of the effect of a given exposure is mediated through an intermediate variable. 0.143) (0. measured by items x9 Ϫ x18 using the threshold models described in Appendix B (available with the electronic version of this paper) because they are all ordered categorical. such that the part of the effect of social class on symptom load mediated through self-efficacy can be measured in the following way: Indirect effect of social class through self-efficacy Total effect ϭ ␥1␥5 ␥6 ϩ ␥2␥4 ϩ ␥1␥5 ϩ ␥3␥4 . The indirect effect of a third variable that lies in 1 or more pathways between the 2 main variables is calculated by taking the sum of the pathways going through this third variable. The methods are standard in path analysis but do not seem to have TABLE 4. 0. A higher level of (lack of) self-efficacy means that the person is feeling less self-confident. and Cynical Hostility Param Estimate 0.028 0.019 Std.243 0. Fig.251) ␥1 ␥2 ␥3 The Mediation Proportion of Two Intermediate Variables To further illustrate the possibilities in the structural equation approach to the measuring of the contribution of intermediate variables.169–0. with 20 –25% mediated directly through cynical hostility. Err. Selfefficacy (regarded as a self-confident view of one’s capability to deal with certain life stressors) was measured by the 10 items in the version of the Generalized Self-Efficacy Scale.

Biometrika. Medley DM.16: 1515–1527. Budtz-Jørgensen E. Biometrics. This example was cross-sectional. The application of latent variables and structural equation models is a practical way to acknowledge the inherent problems of measuring variables that are only partially related to the concepts of real concern. REFERENCES 1. Freedman LS. 1992. et al. Stronks K. latent variables indirectly measured by a set of several related variables. The Sociology of Health Inequalities. and the postulated causal pathways could not be motivated by temporal order. Weinberg CR. Molenberghs G. et al. 2003.58:803– 812. 25. 115:865– 872. Julkunen J. the mediation proportion also covers other situations (eg. with the option of including 2 or more intermediate variables.. anger. Pers Individ Dif.14:105–120. 19. 1993. Preliminary analysis of data should be performed to check whether this is an acceptable assumption.3:143–155. 23. 7. 1999. Blane D.Ditlevsen et al Epidemiology • Volume 16. et al. Budtz-Jørgensen E. January 2005 been routinely applied to epidemiologic problems of the kind addressed here. there was a temporal ordering that made the causal pathways unambiguous. van de Mheen HD. Construct validity and sex differences in cook-medley hostility. Estimation of health effects of prenatal mercury exposure using structural equation models. Jerusalem M. Cox DR. indeed. 5. et al. Schwarzer R.10:37– 48. Results of a prospective randomized placebo-controlled clinical trial. Buyse M. Schatzkin A. Available at: http://www. Mplus. Cook-Medley hostility. and acute myocardial infarction? Am J Epidemiol. Jeremy Taylor. Greenglass ER.54:1014 –1029. Pearl J. Kaplan GA. Test reliability is critically important to molecular epidemiology: an example from studies of human papillomavirus infection and cervical neoplasia. Modern Epidemiology. 1992:195–213. our measure coincides with the measure of Wang and Taylor. Toward a clearer definition of confounding. ACKNOWLEDGMENTS We are grateful to David Cox. the general link function incorporated in their measure. By pretending that we are directly measuring these variables through other variables such as items in a questionnaire. Validation of intermediate end points in cancer research. Lynch JW. In our first example on detrimental effects of interferon-␣ on visual acuity. Friedrichsdorf. et al. A note on the quadratic exponential binary distribution. 3. 15.82: 1746 –1752. 1994. 1998:193–216. 22.68:1059 –1066. 16. J Natl Cancer Inst. Wittenberg J. handling of categorical variables. Greenland S. Biometrics. 1997. Graubard BI. The differences come about in the way the measures are generalized and. 1998. 6. Freedman LS. 12. Schiffman MH. Oxford. Proposed hostility and pharisaic-virtue scales for the MMPI. ed. eds. Rothman KJ. UK: Blackwell. Do cardiovascular risk factors explain the relation between socioeconomic status. Arch Ophthalmol.11:167–178. the mediation proportion offers an easily interpretable measure that does not depend on arbitrary constructions of scales or choices of cut-points. 2004.54(7 Suppl): 1944s–1947s. Schiffman MH. and John W. Robins JM. Washington. 21. Structural Equations with Latent Variables. 1996. Wermuth N. thereby covering a broader range of distributions). net/home/. 11. Greenland S. Germany: ADDITIVE GmbH. 1954. Arminger G. De Gruttola V. However. Schatzkin A.ehjournal. Pharmacological Therapy for Macular Degeneration Study Group. 1996. In the simplest setting of jointly normally distributed variables. Schatzkin A. This should be considered only a partial solution to the general problem of mediating effects. 1991. Stat Med. J Appl Psychol. Los Angeles: Muthe ´n & Muthe ´ n. Lin DY. Moreover. thus reducing the problem of bias. Epidemiology. 1998. 9. and the Type A behavior in Finland. Environ Health ͓serial online͔. Criteria for the validation of surrogate endpoints in randomized experiments.38:414 – 418.1:2. Freedman LS.137:1– 8. A measure of the proportion of treatment effect explained by a surrogate marker. 18. 20. 2. 4. Kauhanen J. User’s Guide. 1997.81:403– 408. 8. Hostility and increased risk of mortality and acute myocardial infarction: the mediating role of behavioral risk factors. Psychol Rep. Surrogate end points in cancer research: a critique. Estimating the proportion of treatment effect explained by a surrogate marker.144:934 –942. Davey G. Muthe ´ n LK. Fleming TR. Am J Epidemiol. New York: John Wiley. Cancer Epidemiol Biomarkers Prev. Mackenbach JP. 1997. The proposed framework allows for error terms on intermediate variables and indicator variables.146:142–152. We also used the mediation proportion to assess how much of the effect of social class on symptom load is mediated through cynical hostility. In: Bartley M. 2002. 1992. Schatzkin A. 1989. Keiding N. Statistical validation of intermediate endpoints for chronic disease. exploration of various indirect pathways within the same model). Keiding N. Muthe ´ n BO. Identifiability and exchangeability for direct and indirect effects. Taylor JMG. cardiovascular mortality. Our statistical analysis. Everson SA. 17. Grandjean P. Dorgan J. DC: Hemisphere. 1994. Schepers A. The structural equation approach is a way to model latent concepts not directly observable. given a substantively motivated model. could not compensate for the lack of time relationship in the observed data. Robins JM. Wang and Taylor include situations not covered by the mediation proportion (eg. Cook WW. However. Self-efficacy as a resource factor in stress appraisal processes.10:209 –218. Kaplan GA. Greenglass ER. Epidemiology. 10. risk of all-cause mortality. as is always the case. Bollen KA. the threshold models yield useful ways of handling ordered categorical variables. Accessed January 15. Julkunen J. 2002. Am J Epidemiol. Cancer Res. Environmetrics. Smith. Grandjean P. McDonald for advice and suggestions for further references. 1990. Wang Y.Interferon alfa-2a is ineffective for patients with choroidal neovascularization secondary to age-related macular degeneration. 1989. 24. Statistical methods for the evaluation of health effects of prenatal mercury exposure. Cohen RD. Greenland S. The discrepancy between the mediation proportion and Wang and Taylor’s measure in this example stems from our modeling of a continuous distribution of visual acuity. Philadelphia: Lippincott Raven. Self-Efficacy: Thought Control of Action. Stat Med. 13. 120 © 2004 Lippincott Williams & Wilkins . we run a risk of getting unpredictably biased results. et al. Number 1. 14. A limitation of the proposed method is the assumption of no effect modification. A lifecourse perspective on socio-economic inequalities in health. 1998. Our analysis indicated a substantially stronger role of the mediator than previous analyses. 1996. In: Schwarzer J. 5:947–953. MECOSA 3 User Guide. Causal diagrams for epidemiological research.