Social Desirability and Social Approval Biases in Dietary Self-Report: Examples of Epidemiologic Effect Modification James R. Hebert, ScD Professor and Chair Department of Epidemiology and Biostatistics Norman J. Arnold School of Public Health University of South Carolina

Other Collaborators: Lynn Clemow, Ph.D. Ira S. Ockene, M.D. Cara B. Ebbeling, Ph.D. Yunsheng Ma, M.D., M.P.H. Thomas G. Hurley, M.Sc. Milagros C. Rosal, Ph.D. Charles E. Matthews, Ph.D. Judith K. Ockene, Ph.D.

 cognitive difficulties  emotional aspects issues of prying  misleading responses You’ve got to be kidding!

Emotional Aspects of Eating -  same part of brain processes sensations of taste, smell, and basic emotional and instinctual behavior.  social messages get overlaid on psychological and emotional ones.

Self-report of dietary intake could be biased by social desirability or social approval thus affecting risk estimates in epidemiologic studies. These constructs produce response set biases, which are evident when testing in domains characterized by easily recognizable correct or desirable responses. Given the social and psychological value ascribed to diet, assessment methodologies used most commonly in epidemiologic studies are particularly vulnerable to these biases. Background

Social Desirability is a Response Set reflecting the defensive tendency to respond in such a way as to avoid criticism in a situation perceived to be a test.

Social Approval is a Response Set reflecting the tendency to actively seek approval in a situation perceived to be a test.

Social desirability bias in dietary self-report may compromise the validity of dietary intake measures. International Journal of Epidemiology 1995; 24:

WORCESTER AREA TRIAL for COUNSELING in HYPERLIPIDEMIA WATCH

We Developed Seven-Day Diet Recall (7DDR) In WATCH The 7DDR looks very much like a FFQ, but it asks participants to recall specific meals and snacks over the past week. It includes 118 foods, 13 beverages, and a worksheet. Thus it combines elements of the FFQ and produces estimates of dietary intake more like trait measures than does 24HR.

WATCH External Validation Study 3 Weeks Pre 7DDR Post 7DDR (7 randomly selected 24-Hrs) Exactly two years later social desirability and approval assessed STUDY DESIGN

Hypothesis: Response Set Biases are expressed on structured questionnaires such as the Seven Day Dietary Recall (7DDR) or the FFQ. it is closed-ended and has obvious correct responses grid-like obvious response categories requires report of diet as a trait rather than a state   

Social desirability and social approval biases were tested by comparing nutrient scores derived from multiple twenty-four hour diet recalls (24HR) on randomly assigned days with those from two seven-day diet recalls (7DDR) (similar in some respects to commonly used food frequency questionnaires), one administered at the beginning of the test period (pre) and one at the end (post). Statistical analysis included correlation and multiple linear regression. Except for b describing relation between methods (where b should equal 1.0), H 0 :  =0. Methods:

Social DesirabilityApproval Assessment 33-item (true/false) Marlowe-Crowne Social Desirability Scale 20-item (5-point Likert) Martin-Larsen Approval Motivation Scale

Summary Statistics (from 24HR) Women (n=27)Men (n=14) Energy (kcal/d)1491 (62)1971 (600) Fat (g/d)51.9 (17.3)65.1 (23.8) % Fat31.3 (5.5)29.4 (4.6) Calcium (mg/d)668 (277)785 (218) Cholesterol (mg/d)185 (65)261 (195) * Tabulated values are mean (standard deviation)

WATCH – External Validation Study – Summary Statistics Variable% or Mean/SD % Married74.1%64.3% % White81.5%85.7% Age (years)51.3/ /13.9 Social Desirability Score20.5/ /5.9 Social Approval Score48.4/ /9.1 Total energy (k/cal)1490/ /599 Total fat (g)51.9/ /23.8 Female Male (N=27) (N=14)

Statistical Model Nutrient test method = Nutrient comparison method + + BMI + Covariates Social Approval + Social Desirability

WATCH – External Validation Study – Regression Results (Pre) 24-HR Score SD Score Total fat (g) n.s. (0.06) Saturated fatty acids (g) n.s. (0.04) Total Energy (kcal) b = SE b = P-value = n.s. (0.06)

24-HR Score SD Score Total fat (g) n.s. (0.06) Saturated fatty acids (g) n.s. n.s. WATCH – External Validation Study – Regression Results (Post) Total Energy (kcal) b = SE b = P-value = n.s. (0.09)

Saturated fatty acids-0.85(0.45)-0.29 (0.67) (0.21) 0.22 (0.49) Total energy (kcal) (30.31) (32.00) (14.35) (32.97) Fat (% energy) -0.33(0.23) -0.13(0.20)-0.01(0.22) (0.21) Cholesterol-7.17(2.55) (7.56)-3.68(2.13) 1.48 (6.00) Results of General Linear Models to Assess Gender Differences In Social Desirability Bias Pre Measurements Post Measurements Women Men Women Men b SE b b SE b b SE b b SE b Dependent Variable Total fat(g) (1.57) -1.11(1.79) (0.64) (1.43)

Pre Measures Post Measures Dependent Variable b Se b b SE b Fat Quartile Quartile Quartile Quartile Total Energy Quartile Quartile Quartile Quartile Results of nutrient quartile stratification to assess variation in social desirability bias

Slide 20  A downward bias in the 7DDR due to social desirability was observed in women.  It amounted to about 700 kcal/day across the interquartile range of social desirability scores for women WATCH – External Validation Study Conclusions:

 The overall purpose of the Worcester Area Trial for Counseling in Hyperlipidemia was:  To evaluate the effectiveness of a physician-delivered nutrition intervention counseling program in reducing dietary fat and LDL WATCH Study Hebert JR, Ma Y, Clemow L, Ockene IS, Saperia G, Stanek EJ, Merriam PA, Ockene JK. Gender differences in social desirability and social approval bias in dietary self report. Am J Epidemiol 1997; 146:

 Over 8,000 patients screened to determine eligibility  Age years  No prior drug treatment and RD referral  First fingerstick for testing cholesterol  Upper 25% of cholesterol distribution  Second fingerstick  Consent Data Collection in WATCH Study: 1,278 patients One year Single 24HR 7DDR 2 lipid profile BMI(kg/m 2 ) SD and SA measures March 1995 Baseline Single 24HR 7DDR 2 lipid profiles BMI(kg/m 2 )

Variable RespondersNon-responder P-value * MEAN (SD) MEAN (SD) Age (Years) (10.63)48.29(10.45)0.06 Baseline BMI (5.29)29.51 (5.90)0.05 Baseline Blood Cholesterol (mg/dl) (46.85) (56.59) Baseline LDL Cholesterol (mg/dl) (39.39) (43.19)0.01 # of extra lipid measures from Fallon 2.15 (1.26) 1.92 (1.22)0.01 *P value based on Chi-square for categorical variables and two sample t-test the two group means Responders Vs. Non-Responders

Male (n=325) Female (n=434) MEAN SD# MEAN SD# P- value* 7DDR Nutrients at Baseline Total Energy (kcal/d) Total Fat (g/d) % Calories from Fat Saturated Fat (g/d) % Calories from Saturated Fat *P value based on two sample t-test the difference of two group means Male and Female Comparison

Results of GLM for Males, WATCH Study Independent Variables: Dependent Variables: 24HR SD SCORE AP SCORE BMI Total Fat (g/d) b (p)(0.0003)(0.24)(0.01)(0.04) Total Saturated Fat (g/d) b (p)(0.0001)(0.23)(0.001)(0.05) Total Energy (kcal/d) b (p)(0.0001)(0.38)(0.01)(0.01)

Results of GLM for Females, WATCH Study Independent Variables: Dependent Variables: 24HR SD SCORE AP SCORE BMI Total Fat (g/d) b (p)(0.0002)(0.07)(0.98)(0.18) Total Saturated Fat (g/d) b (p)(0.0001)(0.05)(0.78)(0.24) Total Energy (kcal/d) b (p)(0.0002)(0.02)(0.99)(0.22)

 In females: SDSCORE 75th percentile=23, 25th percentile=15, 8 points x 19.16=153.3 kcal energy underreport, or a 6.2g fat underreport  In males: APSCORE 75th percentile=41, 25th percentile=32, 9 points x 21.50=193.5 kcal overreport, or a 10.9g fat overreport Some Perspectives

WATCH Social Desirability Study Conclusion  Social desirability was associated with a downward bias in dietary fat and energy intake in females  Social approval was found to be related to over-reporting energy and fat intake in males  Further studies are needed to establish models to adjust for the bias

The Treatwell 5-a-Day Study  Multi-ethnic sample of community health center workers  Representing three control sites  Multiple 24HR as “relative criterion”  Uses three methods for comparison, including Harvard/Channing FFQ  Predominantly (~85% women) Hebert JR, Peterson KE, Hurley TG, Stoddard AM, Cohen N, Field AE, Sorensen G. The effect of social desirability trait on self-reported dietary measures among multi-ethnic female health center employees. Ann Epidemiol 2001; 11:

Black (n=23) Hispanic (n=31) Variable:bSE b b Total Energy Intake (kcal/d) Total Fat Intake (g/d) The Treatwell 5-a-Day Study – Social Desirability Results by Ethnicity, Women Only White (n=30) bSE b

Non-Professional (n=52) Professional (n=39) Variable:bSE b b Total Energy Intake (kcal/d) Total Fat Intake (g/d) Fruit (servings/d) - FFQ Fruit (servings/1000kcal/d) The Treatwell 5-a-Day Study – Results by Occupational Category, Women Only p-value for Ho:Ho:  coll <0.005 <0.05 ns

Less Than College (n=52) College Degree or More (n=39) p-value for H o :  coll Total Energy Intake (kcal/d) Total Fat Intake (g/d) Fruit (servings/d) - FFQ Fruit (servings/1000kcal/d) bSE b b Variable: <0.001 ns The Treatwell 5-a-Day Study – Results by Education, Women Only

The Treatwell 5-a-Day Study – Conclusions  The FFQ also appears to be biased by social desirability in women, but …..  the critical factor determining the bias is education which is …..  more important than occupational category or ethnicity/race.  As in the WATCH study, bias is oriented toward fat/energy intake

The Energy Study, Worcester, MA  First such study to focus on the most widely used FFQ (NCI/WHI) First study to focus on these biases employing stable isotope methods for comparison (TEE from DLW)  Hebert JR, Ebbeling CB, Matthews CE, Ma Y, Clemow L, Hurley TG, Druker S. Systematic errors in middle-aged women's estimates of energy intake: Comparing three self-report measures to total energy expenditure from doubly labeled water. Ann Epidemiol 2001; (In Press):

Overview of Study days Doubly-Labeled Water Metabolic Period Baseline questionnaires  Demographic data (education)  Social desirability (Marlowe-Crowne Scale, 33-item, true/false) Food frequency questionnaire  (WHI)

Married White Pre-menopausal Bachelors Degree or more Employed Full Time Professional, Managerial Work Current Smoker 79.6 Sedentary Description of the Study Population, The Energy Study (N=73) n%

Interquartile RangeMeanStandard DeviationMinimum25%75%Maximum Age (years) Body Mass (kg) BMI (kg/m 2 ) Fat-Free Mass (kg) Social Desirability Score Further description of the Study Population, The Energy Study (N=73)

TEE from DLW (kcal/d) Hour Recall-Derived Data (7-day average) Energy Intake (kcal/d) Food Quotient Day-0 Administration Day-14 Administration FFQ Energy (kcal/d) 2102 Interquartile RangeMeanStandard DeviationMinimum25%75%Maximum Further description of the Study Population, The Energy Study (N=73)

All Education Levels: Whole Sample (n=73) (-65.7, -7.5) Excluding “Outliers ” (n=69) (-34.7, 13.1) High Education (college +) Whole Sample (n=33)-73.3 (-113., -32.9) (-63.6, -0.2) Excluding “Outliers ” (n=31) Social Desirability Bias (kcal/day/point) by Education Level (FFQ-Derived Energy Intake Versus TEE from DLW, Beginning of Metabolic Period), The Energy Study (N=73).

All Education Levels: Whole Sample (n=73) (-34.7, 13.1) Excluding “Outliers ” (n=72)-13.7 (-35.8, 8.4) High Education (college +) Whole Sample (n=33)-21.8 (-53.5, 9.9) Social Desirability Bias (kcal/day/point) by Education Level (FFQ-Derived Energy Intake Versus TEE from DLW, End of Metabolic Period), The Energy Study (N=73).

Social Desirability Bias Whole Sample (n=75) High (n=33) Low (n=42) Education Beginning End Bias (kcal/day/point)

Revisiting WATCH --- Why?  Is there an effect of education when cut at college+?  What happens with these biases after an intervention? Hebert JR, Ma Y, Ebbeling CB, Matthews CE, Ockene IS. Self-report data. Compliance in Healthcare and Research. Armonk, NY: Futura, 2001:

Social Approval Bias in Males, by Education, WATCH Study, Worcester, Massachusetts, < College (n=150) Social Approval Score BMI Baseline Total Energy (kcal/day)29.8 (0.003)29.1 (0.07) Total Fat (g/day)1.63 (0.004)1.60 (0.07) Total SFA (g/day)0.59 (0.003)0.53 (0.09) One-year< College (n=112) Total Energy (kcal/day)36.7 (0.0003)53.4 (0.001) Total Fat (g/day)1.50 (0.004)1.72 (0.04) Total SFA (g/day)0.41 (0.02)0.57 (0.05)

Social Approval Bias in Males, by Education, WATCH Study, Worcester, Massachusetts,  College (n=70) Social Approval Score BMI Baseline Total Energy (kcal/day)8.6 (0.49)48.6 (0.05) Total Fat (g/day)0.58 (0.39)3.87 (0.05) Total SFA (g/day)0.26 (0.26)1.34 (0.07) One-year  College (n=56) Total Energy (kcal/day)19.9 (0.14)33.7 (0.14) Total Fat (g/day)1.05 (0.11)0.70 (0.52) Total SFA (g/day)0.25 (0.18)0.15 (0.63)

 College (n=220) Social Approval Score BMI Baseline Total Energy (kcal/day) Total Fat (g/day) Total SFA (g/day) One-year  College (n=172) Total Energy (kcal/day) Total Fat (g/day) Total SFA (g/day) Social Desirability Score (0.14) (0.34) (0.45) -3.6 (0.77) (0.43) -0.2 (0.97) (0.95) 0.03 (0.76) 11.0 (0.07) 0.36 (0.32) 0.14 (0.21) 6.9 (0.43) 0.25 (0.61) 0.05 (0.75) 11.1 (0.32) 0.36 (0.58) 0.21 (0.32)-0.15 (0.52) Social Approval and Social Desirability Bias in Females, by Education, WATCH Study, Worcester, Massachusetts,

 College (n=64) Social Approval Score BMI Baseline Total Energy (kcal/day) Total Fat (g/day) Total SFA (g/day) One-year  College (n=53) Total Energy (kcal/day) Total Fat (g/day) Total SFA (g/day) Social Desirability Score (0.04) (0.07) (0.01) -9.5 (0.54) (0.80) (0.86) -2.9 (0.72) (0.49) (0.52) -5.1 (0.61) (0.67) (0.74) 19.7 (0.11) 1.42 (0.05) 0.41 (0.07) 35.5 (0.04) 1.98 (0.02) 0.75 (0.01) Social Approval and Social Desirability Bias in Females, by Education, WATCH Study, Worcester, Massachusetts,

WATCH Study Conclusions:  Education modifies the effect of the social desirability and social approval  The effects differ by gender  There appears to be a differential effect of the intervention on the bias according to gender and education

The Role of Social Desirability in Epidemiologic Confounding SD ScorePsychologic Predispositions Physiologic Responses (e.g., Immune Function) Disease True Diet Reported Diet

Total Fat and Saturated Fat Change in fat intake (% total energy) ±0.38 ±0.14 ±1.36 ±0.49 ±1.39 ±0.50 Total Fat Saturated Fat (n=645) Never Referred< 3 Sessions  3 Sessions WATCH Nutritionist Intervention:

Total Cholesterol and LDL Changes in serum cholesterol (mmol/L) ±0.03 ±0.12 ±0.11 ±0.13 ±0.11 TC LDL-C (n=555) Never Referred< 3 Sessions  3 Sessions WATCH Nutritionist Intervention:

Actual Changes in Total Cholesterol vs. 7DDR - Predicted Values Change in Total Serum Cholesterol (mg/dL) Actual Keys Prediction Hegsted Prediction WATCH Nutritionist Intervention:

Variable  P * Self-reported data Fat intake (% energy) Body weight (kg) Measured data Serum LDL-C (mmol/L) Body weight (kg) Table 4. Effects of social desirability on self-reported and measured change scores, WATCH Study, Worcester, Massachusetts, * P-value for the test of H 0 :=0