Propensity score methods are common for estimating a binary treatment effect when treatment assignment is not randomized. When exposure is measured on an ordinal scale (i.e. low–medium–high), however, propensity score inference requires extensions which have received limited attention. Estimands of possible interest with an ordinal exposure are the average treatment effects between each pair of exposure levels. Using these estimands, it is possible to determine an optimal exposure level. Traditional methods, including dichotomization of the exposure or a series of binary propensity score comparisons across exposure pairs, are generally inadequate for identification of optimal levels. We combine subclassification with regression adjustment to estimate transitive, unbiased average causal effects across an ordered exposure, and apply our method on the 2005–2006 National Health and Nutrition Examination Survey to estimate the effects of nutritional label use on body mass index.
Food and Drug Administration. Education Act of 1990. Public law1990; 101: 104–104.
2.
Food and Drug Administration. Regulatory Impact Analysis of the Final Rules to Amend the Food Labeling Regulations. Federal Register. 1993.
3.
VariyamJNCawleyJ. Nutrition labels and obesity, Cambridge, MA: National Bureau of Economic Research, 2006.
4.
LoureiroMLYenSTNaygaRMJr. The effects of nutritional labels on obesity. Agric Econ2012; 43: 333–342.
5.
DrichoutisACNaygaRMJrLazaridisP. Can nutritional label use influence body weight outcomes?Kyklos2009; 62: 500–525.
6.
RoystonPAltmanDGSauerbreiW. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med2006; 25: 127–141.
7.
ChamblissHO. Exercise duration and intensity in a weight-loss program. Clin J Sport Med2005; 15: 113–115.
8.
PuetzTWFlowersSSOConnorPJ. A randomized controlled trial of the effect of aerobic exercise training on feelings of energy and fatigue in sedentary young adults with persistent fatigue. Psychother Psychosom2008; 77: 167–174.
9.
United States Public Health Service Office of the Surgeon General et al. Physical activity and health: a report of the surgeon, Darby, PA: DIANE Publishing, 1996.
10.
U.S. Food and Drug Administration. How to understand and use the nutrition facts label. http://www.fda.gov/Food (2013, accessed 19 September 2013).
NeuhouserMLKristalARPattersonRE. Use of food nutrition labels is associated with lower fat intake. J Am Diet Assoc1999; 99: 45–53.
15.
SatiaJAGalankoJANeuhouserML. Food nutrition label use is associated with demographic, behavioral, and psychosocial factors and dietary intake among African Americans in North Carolina. J Am Diet Assoc2005; 105: 392–402.
16.
RosenbaumPRRubinDB. The central role of the propensity score in observational studies for causal effects. Biometrika1983; 70: 41–55.
17.
JoffeMMRosenbaumPR. Invited commentary: propensity scores. Am J Epidemiol1999; 150: 327–333.
18.
ImaiKVan DykDA. Causal inference with general treatment regimes. J Am Stat Assoc2004; 99: 854–866.
19.
RubinDB. For objective causal inference, design trumps analysis. Ann Appl Stat2008; 2: 808–840.
20.
ZanuttoELuBHornikR. Using propensity score subclassification for multiple treatment doses to evaluate a national antidrug media campaign. J Educ Behav Stat2005; 30: 59–73.
21.
Splawa-NeymanJDabrowskaDSpeedT, et al.On the application of probability theory to agricultural experiments. Essay on principles. Section 9. Stat Sci1990; 5: 465–472.
22.
HollandPW. Statistics and causal inference. J Am Stat Assoc1986; 81: 945–960.
23.
RubinDB. Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol1974; 66: 688–688.
24.
RubinDB. Randomization analysis of experimental data: the Fisher randomization test comment. J Am Stat Assoc1980; 75: 591–593.
25.
StuartEA. Matching methods for causal inference: a review and a look forward. Stat Sci2010; 25: 1–1.
26.
ImbensGW. The role of the propensity score in estimating dose-response functions. Biometrika2000; 87: 706–710.
27.
LechnerM. Identification and estimation of causal effects of multiple treatments under the conditional independence assumption, Heidelberg: Springer, 2001.
28.
McCullaghP. Regression models for ordinal data. J R Stat Soc B1980; 42: 109–142.
29.
CaliendoMKopeinigS. Some practical guidance for the implementation of propensity score matching. J Econ Surv2008; 22: 31–72.
30.
RubinDB. Using propensity scores to help design observational studies: application to the tobacco litigation. Health Serv Outcomes Res Methodol2001; 2: 169–188.
31.
LuBZanuttoEHornikR, et al.Matching with doses in an observational study of a media campaign against drug abuse. J Am Stat Assoc2001; 96: 1245–1253.
32.
RubinDThomasN. Matching using estimated propensity scores: relating theory to practice. Biometrics1996; 52: 249–264.
33.
DehejiaRHWahbaS. Causal effects in non-experimental studies: re-evaluating the evaluation of training programs, Cambridge, MA: NBER, 1998.
34.
ImbensGRubinDB. Causal inference in statistics and the social sciences, Cambridge, UK: University Press, 2013.
35.
CochranWGRubinDB. Controlling bias in observational studies: a review. Sankhyā1973; Series A: 417–446.
36.
RosenbaumPRRubinDB. Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc1984; 79: 516–524.
37.
AustinPC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med2009; 28: 3083–3107.
38.
RubinDB. Using multivariate matched sampling and regression adjustment to control bias in observational studies. J Am Stat Assoc1979; 74: 318–328.
39.
LuncefordJKDavidianM. Stratification and weighting via the propensity score in estimation of causal treatment effects: a comparative study. Stat Med2004; 23: 2937–2960.
40.
AbadieAImbensGW. Large sample properties of matching estimators for average treatment effects. Econometrica2006; 74: 235–267.
41.
DuJ. Valid inferences after propensity score subclassification using maximum number of subclasses as building blocks, Cambridge, MA: Department of Statistics, Harvard University, 1998.
42.
BenjaminDJ. Does 401 (k) eligibility increase saving?: evidence from propensity score subclassification. J Public Econ2003; 87: 1259–1290.
43.
LechnerM. Program heterogeneity and propensity score matching: an application to the evaluation of active labor market policies. Rev Econ Stat2002; 84: 205–220.
44.
McCaffreyDFGriffinBAAlmirallD, et al.A tutorial on propensity score estimation for multiple treatments using generalized boosted models. Stat Med2013; 32: 3388–3414.
45.
FengPZhouXHZouQM, et al.Generalized propensity score for estimating the average treatment effect of multiple treatments. Stat Med2012; 31: 681–697.
46.
LittleRJ. Missing-data adjustments in large surveys. J Bus Econ Stat1988; 6: 287–296.
47.
KangJDYSchaferJL. Demystifying double robustness: a comparison of alternative strategies for estimating a population mean from incomplete data. Stat Sci2007; 22: 523–539.
48.
HuberMLechnerMWunschC. The performance of estimators based on the propensity score. J Econ2013; 175: 1–21.
49.
KilpatrickRDGilbertsonDBrookhartMA, et al.Exploring large weight deletion and the ability to balance confounders when using inverse probability of treatment weighting in the presence of rare treatment decisions. Pharmacoepidemiol Drug Saf2013; 22: 111–121.
50.
StuartEARubinDB. Best practices in quasi-experimental designs. Best Pract Quant Methods2008, pp. 155–176.
51.
LewisJEArheartKLLeBlancWG, et al.Food label use and awareness of nutritional information and recommendations among persons with chronic disease. Am J Clin Nutr2009; 90: 1351–1357.
52.
RubinDB. Bayesian inference for causal effects: the role of randomization. Ann Stat1978; 6: 34–58.
53.
D’AgostinoRB. Tutorial in biostatistics: propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med1998; 17: 2265–2281.
54.
WooldridgeJM. Inverse probability weighted estimation for general missing data problems. J Econ2007; 141: 1281–1301.
55.
RubinDB. Multiple imputation after 18+ years. J Am Stat Assoc1996; 91: 473–489.
56.
AhmedAHusainALoveTE, et al.Heart failure, chronic diuretic use, and increase in mortality and hospitalization: an observational study using propensity score methods. Euro Heart J2006; 27: 1431–1439.
57.
JolliffeI. Principal component analysis, New York: Wiley Online Library, 2005.
58.
WaernbaumI. Model misspecification and robustness in causal inference: comparing matching with doubly robust estimation. Stat Med2012; 31: 1572–1581.
59.
RosenbaumPR. Design sensitivity in observational studies. Biometrika2004; 91: 153–164.
60.
DanielsMJHoganJW. Missing data in longitudinal studies: strategies for Bayesian modeling and sensitivity analysis2008; Vol. 109.Boca Raton, FL: Chapman and Hall/CRC.
61.
HosmanCAHansenBBHollandPW, et al.The sensitivity of linear regression coefficients confidence limits to the omission of a confounder. Ann Appl Stat2010; 4: 849–870.
62.
Liu T and Hogan JW. Inference about ATE from observational studies with continuous outcome and unmeasured confounding. arXiv preprint arXiv:13036165. 2013.
63.
HernánMAAlonsoALoganR, et al.Observational studies analyzed like randomized experiments: an application to postmenopausal hormone therapy and coronary heart disease. Epidemiology2008; 19: 766–779.
64.
PearlJBareinboimE. External validity: from do-calculus to transportability across populations, UCLA Department of Computer Science: DTIC Document, 2012.
65.
JohnsonMLCrownWMartinBC, et al.Good research practices for comparative effectiveness research: analytic methods to improve causal inference from nonrandomized studies of treatment effects using secondary data sources: the ISPOR Good Research Practices for Retrospective Database Analysis Task Force ReportPart III. Value Health2009; 12: 1062–1073.
66.
RubinDB. On the limitations of comparative effectiveness research. Stat Med2010; 29: 1991–1995.
67.
MyersRH. Classical and modern regression with applications1990; Vol. 2.Belmont, CA: Duxbury Press.
