Congruence Research: Methodological Advancement Can Speed Its Impact on the Public s Health L. ALISON PHILLIPS RUTGERS UNIVERSITY APRIL 28, 2011
Example Congruence Topics Congruence between X and Y patient satisfaction, adherence, or health outcomes X and Y: Patients preference for and providers actual sharing of medical information Patients preference for and perceived amount of participation in decision making Patients beliefs and their spouses beliefs regarding the patients control over his/her illness Patients beliefs and providers beliefs regarding whether an illness is primarily psychological or physical in nature Christensen et al., 2010; Cvengros et al. 2007, 2009; Jhang et al. 2005; Sterba et al., 2008; Greer & Halgin, 2006
Algebraic (raw) difference score model Z = a+b(x-y)+e Used in regression to determine importance of congruence Interpretation: Patient adherence increases as X approaches Y and continues to increase as X exceeds Y. X = Provider s estimate of the patient s self-assessed health (SAH) Y = Provider s rating of the patient s health b=.35, R 2 =.10, p<.001 1.5 1 0.5 0-0.5-1 -1.5 Patient Adherence -3-2 -1 0 1 2 3 X - Y
Absolute Value Difference Score Z = a+b X-Y +e Used in regression to determine effect of degree of incongruence Interpretation: Patient adherence decreases linearly and symmetrically as the difference between X and Y increases b=.41, p<.001 Adjusted R 2 =.10, p<.001 0.4 0.2 0-0.2-0.4-0.6-0.8-1 -1.2 Patient Adherence -3-2 -1 0 1 2 3 X - Y
Combination: Difference score and congruence groups Algebraic difference score (X-Y) split into three groups based on values or tertiles Used in ANOVA to determine effect of direction of incongruence Interpretation: Adherence is significantly lower only when the provider s rating is higher than the provider s estimate of the patient s rating F (2, 124) = 13.40, p<.001 0.4 0.2 0-0.2-0.4-0.6-0.8 Patient Adherence Patient > Provider Congruence Provider > Patient
Fundamental Source of Problems Congruence indices represent 3-dimensions in 2-D space, losing statistical and theoretical meaning 1,2 Patient Adherence Patient Adherence 1.5 1 1.5 0.5 1 0.5 0-0.5-1 -1.5-3 -2-1 0 1 2 3 X Y 3 0-3 -3 0 3-0 -0.5-1 1.5 1 Johns, 1981, 2 Edwards, 2002
Specific Problems: Constraints Increase Type II Error risk Loss of unique statistical information for components Loss of statistical power/attenuation of effect sizes and teststatistics Does this mean they provide conservative tests? Increase Type I Error risk E.g., X or Y can be strong predictor of Z and result in significant relationship between X-Y and Z Constrain theoretical hypotheses Limited to concluding congruence is/is not important rather than how and to what degree it is important for the outcome Johns, 1981; Cronbach, 1953
Polynomial Regression A 3-D analysis that maintains unique contributions of separate predictor components Confirmatory method: Allows for empirical assessment of the implied constraints of difference score models Empirical data: the algebraic difference score was accurate Exploratory method: Finds best-fitting unconstrained polynomial model (requires replication) Allows for tests of more theoretically complex hypotheses
50 40 30 20 Z 10 0 3 1.8-10 0.6 Y -0.6-1.8-3 -3-1.8-0.6 0.6 X 1.8 3-20 A perfect congruence relationship Outcome maximized when X=Y, uniformly for all values of X and Y Outcome most negatively affected when X is maximally incongruent with Y Outcome equally affected when X>Y as when Y>X
50 40 30 20 Z 10 0 3 1.8-10 0.6 Y -0.6-1.8-3 -3-1.8-0.6 0.6 X 1.8 3-20 Outcome maximized when X=Y, uniformly for all values of X and Y Outcome most negatively affected when X is maximally incongruent with Y *Added complexity: the outcome is maximized when Y is just slightly greater than X
50 40 30 20 Z 10 0 3 1.8-10 0.6 Y -0.6-1.8-3 -3-1.8-0.6 0.6 X 1.8 3-20 Outcome maximized when X=Y, uniformly for all values of X and Y Outcome most negatively affected when X is maximally incongruent with Y *Added complexity: the outcome is more negatively affected when X>Y than when Y>X.
5 4.5 Z 4 3.5 3 2.5 2 1.5 2 1.2 0.4-0.4 Y -1.2-2 -2-1.2-0.4 0.4 X 1.2 2 1 0.5 0 Outcome maximized when X=Y Outcome most negatively affected when X is maximally incongruent with Y Outcome equally affected when X>Y as when Y>X *Added complexity: absolute values of X/Y matter
Take Aways Theory Method/Analysis We learn first how the methods and analyses we choose depend on the theory and hypotheses we wish to test The reverse is also true: the methods we choose, or have available to us, can determine the theoretical hypotheses and conclusions we make
References Christensen, A. J., Howren, M. B., Hillis, S. L., Kaboli, P., Carter, B. L., Cvengros, J. A., et al. (2010). Patient and physician beliefs about control over health: Association of symmetrical beliefs with medication regimen adherence. JGIM, 25, 397-402. Cronbach, L. J. (1958). Proposals leading to analytic treatment of social perception scores. In Tagiuri R, Petrullo L, eds. Person Perception and Interpersonal Behavior. Stanford, CA: Stanford University Press: 353-379. Cvengros, J. A., Christensen, A. J., Cunningham, C., Hills, S. L., Kaboli, P. J. (2009). Patient preferences for and reports of provider behavior: Impact of symmetry on patient outcomes. Health Psychology, 28, 660-667. Cvengros, J. A., Christensen, A. J., Hillis, S. L., Rosenthal, G. E. (2007). Patient and physician attitudes in the health care context: Attitudinal symmetry predicts patient satisfaction and adherence. Annals of Behavioral Medicine, 33, 262-268. Greer, J., & Halgin, R. (2006). Predictors of physician-patient agreement on symptom etiology in primary care. Psychosomatic Medicine, 68, 277-282. Jahng, K. H., Martin, L. R., Golin, C. E., & DiMatteo, M. R. (2005). Preferences for medical collaboration: Patient-physician congruence and patient outcomes. Patient Education and Counseling, 57, 308-314. Sterba, K. R., DeVellis, R. F., Lewis, M. A., DeVellis, B. M., Jordan, J. M., Baucom, D. H. (2008). Effect of couple illness perception congruence on psychological adjustment in women with rheumatoid arthritis. Health Psychology, 27, 221-229.
Further Reading Johns, G. (1981). Difference score measures of organizational behavior variables: A critique. Organizational Behavior and Human Performance, 27, 443-463. Edwards, J. R. (1994). The study of congruence in organizational behavior research: Critique and proposed alternative. Organizational Behavior &Human Decision Processes, 58, 51-100. Edwards, J. R. (2001). Ten difference score myths. Organizational Research Methods,4,265-287. Edwards, J. R. (2002). Alternatives to difference scores: Polynomial regression analysis and response surface methodology. In Drasgow F, Schmitt NW, eds. Advances in Measurement and Data Analysis. San Francisco, CA: Jossey-Bass: 350-400. Edwards, J. R., & Harrison, R. V. (1993). Job demands and worker health: Threedimensional reexamination of the relationship between person-environment fit and strain. Journal of Applied Psychology, 78, 628-648. MacCallum, R. C., Zhang, S., Preacher, K. J., Rucker, D. D. (2002). On the practice of dichotomization of quantitative variables. Psychological Methods, 7, 19-40. Shanock, L. R., Baran, B. E., Gentry, W. A., Pattinson, S. C., Heggestad, E. D. (2010). Polynomial regression with response surface analysis: A powerful approach for examining moderation and overcoming limitations of difference scores. Journal of Business and Psychology, 25, 543-554.
Extra Slides
Table of Constraints Model Constrained Equation Unconstrained Equation Constraints Algebraic Difference Score Z=a+b(X-Y)+e Z=a+b 1 X+b 2 Y+e b 1 =-b 2 =a+bx-by+e Absolute Value Difference Score Z=a+b X-Y +e =a+b(1-2w)(x-y)+e =a+bx-by-2bwx +2bWY+e Z=a+b 1 X+b 2 Y+b 3 W +b 4 WX+b 5 WY+e b1=-b2 b4=-b5 b3=0 b4=-2b1 Note. e is error; in the absolute value difference score model, W is a constant =0 when X>Y, =1 when Y>X, and randomly assigned 0 or 1 when X = Y (Edwards, 2002). Adapted from Organizational Behavior and Human Decision Processes, volume 58, Jeffrey R. Edwards, The study of congruence in organizational behavior research: Critique and a proposed alternative, page 54, 1994,with permission from Elsevier.
Polynomial Regression: Confirmatory Approach Use if a priori hypothesize 3D difference score model Scale-center the two predictor components 3 Options to deal with outliers/influential cases: Remove using conservative criteria on leverage, Cook s D, residuals and report results with/without cases included Split data into initial and replication samples Use bootstrapping to determine coefficients for terms Create all required terms (higher order, e.g.) Run the unconstrained equation in regression or GLM
Confirmatory Approach cont In order to find support for hypothesized 3D difference score model: Unconstrained equation must be significant Individual terms must be as implied by model (in significance and direction) Constraints must not significantly decrease variance explained Higher-order terms (as a set) must not explain significant incremental variance to the unconstrained terms
Polynomial Regression: Exploratory Approach Use if: Difference score constraints are rejected No a priori hypotheses regarding the particular congruence model Hypothesize more complex relationships than can be tested with confirmatory approach Create terms from centered predictor components Check outliers/influential cases Test sets of terms of same order until set is n.s. (linear, quadratic, cubic, quartic) Interpret final model, using RSM 1 Bollen & Jackman, 1990, 2 Edwards, 2002
Response Surface Methodology 3 traits of a quadratic surface that can be used to test congruence hypotheses: Stationary point Principle Axes First Principle Axis Second Principle Axis Slopes and curvatures of the surface along lines of interest Y=X, in the XY-plane Y=-X, in the XY-plane