Testing the Theory of Planned Behavior and Reasoned

Testing the Theory of Planned Behavior and Reasoned Action Approach: A Meta. Analytic Structural Equation Modeling Approach Martin S. Hagger John Curtin Distinguished Professor, Curtin University, Perth, Australia Finland Distinguished Professor (Fi. Di. Pro), University of Jyväskylä, Finland

Overview n n n Cumulative tests of the TPB using metaanalysis Meta-analytic structural equation modeling Basic examples – ‘univariate’ approach – TPB in exercise, with self-efficacy – TPB in chronic illness – TPB in health contexts, behavior as a moderator n Advanced examples – MASEM – RAA and TPB in health behaviors – Role of past behavior and habit n Implications for theory and practice

‘Testing’ the TPB n TPB tests involve testing key hypotheses: n Direct effects: – Intention→Behavior – PBC→Behavior – Attitudes→Intention – Subjective norms→Intention – PBC→Intention Indirect effects: – Attitudes→Intention→Behavior – Subjective norms→Intention→Behavior – PBC→Intention→Behavior Usually through regression/path analysis/SEM

Cumulative ‘Tests’ of the TPB n n n n What are the trends in effect sizes across research? Individual studies typically have relatively low statistical power Unresolved contradictory findings using significance tests (highly dependent on sample size) Introducing new studies does not necessarily resolve these issues Cumulative tests of the TPB aim to examine key TPB predictions across multiple samples One approach is to use multi-group SEM across studies (Cheung, 2010) Can be very cumbersome and lots of parameters

Cumulative ‘Tests’ of the TPB n n An alternative is to combine meta-analysis and path analysis/SEM Typically a two-stage approach is used: (1) Combine effect sizes among TPB constructs into a ‘pooled’ correlation matrix (2) Fit a structural equation model to the pooled matrix n Two methods: Univariate and MASEM

Cumulative ‘Tests’ of the TPB n Univariate approach: Stage 1 pooled correlations – Synthesizes correlation coefficients as if they were independent – Incomplete data sets handled by either listwise or pairwise deletion n Univariate approach: Stage 2 analysis – Fit the model to the meta-analytically derived pooled correlation matrix – The matrix treated as if it were a covariance matrix – Researchers usually use harmonic mean as a ‘conservative’ estimate of the sample sizes

Cumulative Evidence for the TPB Attitudes. 370 . 302 . 370 Subjective Norms. 038. 510 Intentions . 270 . 356 Behaviour . 281. 370 . 153 Self-Efficacy. 360 Indirect effects β) . 272. 123 Att Int Beh . 108 SN Int Beh. 014 Perceived PBC Int Beh. 097 Behavioral SE Int Beh. 100 Control Hagger, Chatzisarantis, & Biddle (2002) J. Sport Exerc. Psychol.

Cumulative Evidence for the TPB Attitudes . 819 . 196 . 953 . 320. 420 Subjective Norms . 160 . 250 . 388 Perceived Behavioral Control Intentions . 131 . 213 Behaviour Indirect effects β) Att Int Beh . 042 SN Int Beh . 034 PBC Int Beh . 083 Rich, Brandes, Mullan, & Hagger (2015) J. Behav. Med.

Cumulative Evidence. 298 for the TPB. 337 Attitudes Subjective Norms Perceived Behavioral Control . 399. 392. 355. 358. 227. 126. 119. 230. 315. 119 Intentions . 345. 362. 420. 354. 345. 306 Behaviours Risk. 297. 345. 325. 304. 258. 309 Mc. Eachen et al. (2011) Health Psychol. Rev. Detection Physical activity Dietary Safe sex Abstinence

Summary: Cumulative Tests of TPB Using Univariate Approach n n n Relative consistency in averaged effects across samples and behaviors Indirect effects supported High variability – all effects have substantive heterogeneity… …but they are non-zero Effect size is more relevant than statistical significance! Behavioral moderators seem not to resolve inconsistencies

Univariate vs. MASEM approaches n n Usually an estimate of the sample size is used – harmonic mean But sample size affects standard errors so the test statistics may be incorrect (Cudeck, 1989; Cheung, 2015) MASEM solves this problem by estimating a matrix of variance/covariance estimates Weight the elements of the pooled sampling matrix when estimating the models

The Reasoned Action Approach Hagger, Polet & Lintunen (2018) Soc. Sci. Med.

The Reasoned Action Approach Affective Attitudes Instrumental Attitudes Injunctive Norm Descriptive Norm . 296. 092 . 079 Intention . 151 . 370 Behavior . 032 Largest Indirect effects β) Autonomy . 318 AAtt Int Beh . 110 DN Int Beh . 056 Cap Int Beh . 118 Capacity Source: Hagger et al. (2018) Soc. Sci. Med.

TPB and Self. Control . 104 Attitudes. 126 Trait. 510 Self-Control . 088 Subjective Norms . 440. 174 Intentions . 426 Behaviour . 152 . 224 . 059 Perceived Behavioral Control Standardized Indirect effects β) . 089 Hagger, Hankonen, Kangro, Polet, Lintunen, Reis, & Hamilton (under review) Att Int Beh . 187 SN Int Beh . 074 PBC Int Beh . 064 SC TPB Vars Beh . 045 SC Beh (total) . 134

What’s the Value of These Syntheses? n n n Syntheses of TPB studies provide evidence for robustness (non-zero, heterogeneity) But most population correlation matrices are heterogeneous There is still value in testing whether the proposed model fits the ‘average’ (or ‘typical’) study Provides evidence of likely effect sizes Cumulative evidence may ‘signpost’ viable targets for intervention

Past Behavior in the TPB • Past behavior = frequency of doing the behavior in the past • Past behavior may model habits (Trandis, 1977; Ouellette & Wood, 1998) • Past behavior may also account for previous decision making (Ajzen, 2002) • Past behavior effects in TPB may test the sufficiency of theory (Ajzen, 1991) • Extinction of effects in social cognition theories may mean they are redundant as means to guide intervention

Effects of Past Behavior. 41 Past Behavior Attitudes Intentions . 30. 24 . 09 . 51. 39 Subjective Norms . 32. 55 . 22. 24 . 03. 04 Perceived Behavioral Control . 44 Source: Hagger, Chan, Protogerou, & Chatzisarantis (2016) Preventive Medicine Behavior

Past Behavior in the Reasoned Action Approach Affective Attitudes Instrumental Attitudes . 401. 290 Injunctive Norm . 245. 296. 072. 092 . 071. 079 . 265 Past Behavior . 338. 224. 398 Descriptive Norm Autonomy Intention . 090. 151 . 370. 175 Behavior . 022. 032 . 267. 318 . 412 Capacity Source: Hagger et al. (2018) Soc. Sci. Med.

What About Past Behavior and Habits? Social Cognitive Variables Past Behavior Intentions Habit (Self-Reported ? Automaticity) Health Behavior

People are Creatures of Habit!. 32 Past behavior . 21 Attitudes. 25 . 13. 18 Subjective Norms . 43 . 13 Intentions . 15 Behavior . 30. 26 . 19 Self. Reported Habit . 08. 31 Perceived Control Effect summary (β). 17 Source: PB Hab Beh . 10* PB Beh (total) . 49* PM . 20 Hagger, Protogerou, Mallia, Girelli, Zhang & Lucidi (submitted).

Summary • Syntheses of TPB using MASEM has shown robust, non-zero effects but high heterogeneity • Past behavior attenuates effects, but effects remain – most is sufficient in many cases • Behavior type does not seem to be a critical moderator – perhaps for norms • Future research should move beyond statistical significance – we have enough evidence now to have very precise ‘priors’ for large-scale replications • Future primary research should specify predicted effect sizes ranges rather than statistical significance

Thanks n n n n n Stuart J. H. Biddle Kim Brandes Derwin K. C. Chan Nikos L. D. Chatzisarantis Laura Girelli Kyra Hamilton Nelli Hankonen Eva-Maria Kangro Taru Lintunen Funding n Tekes n Business Finland n Health Collaborative Research Network n n n n n Fabio Lucidi Luca Mallia Barbara Mullan Juho Polet Cleo Protogerou Francis Ries Antonia Rich Chunqing Zhang Ministry of Education, Finland University of Rome “La Sapienza”