The Theory of Planned Behavior Within the Stages of the ...

STRUCTURAL EQUATION MODELING, 14(4), 649–670 Copyright © 2007, Lawrence Erlbaum Associates, Inc.

The Theory of Planned Behavior Within the Stages of the Transtheoretical Model: Latent Structural Modeling of Stage-Specific Prediction Patterns in Physical Activity Sonia Lippke Department of Health Psychology, Freie Universitaet Berlin

Claudio R. Nigg and Jay E. Maddock Department of Public Health Sciences, John A. Burns School of Medicine, University of Hawaii at Manoa

This is the first study to test whether the stages of change of the transtheoretical model are qualitatively different through exploring discontinuity patterns in theory of planned behavior (TPB) variables using latent multigroup structural equation modeling (MSEM) with AMOS. Discontinuity patterns in terms of latent means and prediction patterns for the different stage groups were examined. Adults (n D 3,462) were assessed on their physical activity stages of change and TPB variables. The TPB was separately examined within the five stage groups. The TPB measurement model fit was acceptable. Latent mean analyses with post-hoc contrast and MSEM indicated discontinuity patterns. Results underscore the qualitative differences between the stages that may guide further research and the design of interventions integrating the approaches.

Structural equation modeling (SEM) has been emphasized as a preferred statistical technique because it combines multiple regression and factor analysis

Correspondence should be addressed to Sonia Lippke, Freie Universitaet Berlin, Health Psychology, Habelschwerdter Allee 45 (PF 10), 14195 Berlin, Germany. E-mail: [email protected]

649

650

LIPPKE, NIGG, MADDOCK

procedures. With this approach it is possible to test measurement issues and theoretically based structures of variables with a large number of independent, mediator, and dependent variables (Burkholder & Harlow, 2003). SEM is ideal for testing theoretical models with complex variable relations (Hankins, French, & Horne, 2000) and modeling effects over time (Burkholder & Harlow, 2003). One further advantage of SEM is testing for multigroup invariance within the measurement and interrelations of the independent, moderating, and dependent variables (Byrne, 2004). In this study, multigroup structural equation modeling (MSEM) was used to test the combination of two theories of health behavior change to assess the advantages of this along with possible measurement issues, and to help direct more effective, theory-based interventions. Two different kinds of health behavior theories have been chosen with the study aim of integrating continuous and stage models.

CONTINUOUS MODELS OF HEALTH BEHAVIOR AND THE THEORY OF PLANNED BEHAVIOR In continuous models, individuals are placed along a continuum that reflects the likelihood of action. Influential predictor variables are identified and combined in one equation. The assumption is that one prediction equation meets all needs (one-size-fits-all; Kreuter, Strecher, & Glasman, 1999). The goal of an intervention based on this theoretically or empirically derived equation is to move the individual along this continuum toward action. Thereby, quantitative differences between persons are recognized but qualitative changes in the progress are not identified (Weinstein, 1993). One assumption is, for example, that the higher an intention the more likely the corresponding health behavior. One such continuous model is the theory of planned behavior (TPB; Ajzen, 1991). The TPB proposes that one’s intention to perform a behavior determines that behavior. Intention is influenced by three social-cognitive variables: attitude, subjective norm, and perceived behavioral control. Attitude is the positive or negative evaluation of the behavior. Subjective norm is the social pressure that individuals may perceive to perform the behavior. Perceived behavioral control (PBC) is the individual’s perception of his or her ability and controllability to perform a behavior. The TPB hypothesizes that PBC might have direct effects on intention and behavior, and that intention mediates the impact of attitude, subjective norm, and PBC to behavior. According to the TPB, the higher the intention, attitude, subjective norm, and PBC, the more likely is the behavior. Several meta-analyses and narrative reviews have demonstrated the TPB to be useful in general, and for the prediction of health behaviors in particular (for an overview see Conner & Sparks, 2005). The meta-analysis on physical activity by Hagger, Chatzisarantis, and Biddle (2002), found medium to large effect

PREDICTION PATTERNS

651

sizes for the intention-behavior, attitude-intention and PBC-intention relations, and a smaller effect for the subjective norm-intention relation. Two ideas led to the inclusion of a stage model into the investigation of the TPB: In the metaanalysis by Hagger et al. (2002) the included moderator groups (e.g., age) lacked homogeneity, and the question was whether stages would be moderators. The relations between the variables of the TPB in the different stage groups should be different and stage specific. The second idea is the precondition for more effective interventions that are tailored to an individual’s stage. Using the TPB, an intervention might be targeted to the TPB constructs important for different subgroups consisting of people in the same stage (Nigg, 2003).

STAGE THEORIES AND THE TRANSTHEORETICAL MODEL The main purpose of stage theories is to understand the mechanisms of behavior change. Stage models of health behavior change describe how individuals move through discrete stages while undergoing behavior change. According to such models, persons at different stages think and behave in different ways (Weinstein, Rothman, & Sutton, 1998). Stages are categories into which people can be classified according to presumptions of the stage theory. This approach attempts to explain several distinct steps along the way to action (Weinstein et al., 1998). With the stage construct, the dynamic nature of health behavior change is emphasized and made measurable. By identifying influences and factors that induce movement from one category to the next, the process of behavior change might be supported more effectively (Adams & White, 2003; Kreuter et al., 1999). Weinstein (1993) argued that different variables and processes could be important at different stages; therefore, various predicting rules are essential. Empirical findings support that one size does not fit all and that not one big bullet would predict all changes (Kreuter et al., 1999). In other words, there should be stage-specific effects of social-cognitive variables on behavior change. Therefore, a combination of continuous and stage assumptions may be adequate to model these stage-specific prediction patterns. The Transtheoretical Model (TTM; Prochaska & DiClemente, 1983) proposes five stages of change. The first stage is the precontemplation stage (PC) in which individuals do not consider any behavior change. In the contemplation stage (C), individuals consider performing the health behavior in question but have not yet decided to change. In the preparation stage (P), individuals prepare and plan the actual behavior change. In the action stage (A), a new goal behavior is initiated. When the behavior is performed and consolidated for a longer time, the maintenance stage (M) is reached (Prochaska & DiClemente, 1983). According to stage models, changes in a health behavior consist of movements through stages. As individuals move through these stages, social-cognitive

652


variables change their impact for progression and certain stage-specific barriers emerge. Individuals in the same stage show relatively small differences and relatively large differences with people in other stages (Weinstein et al., 1998). Several meta-analyses exist testing the TTM cross-sectionally by comparing the means of the stage groups. Rosen (2000) reanalyzed the processes of change and found for exercise that people in action and maintenance use all processes more than individuals in inactive stages. A meta-analysis on the TTM for physical activity by Marshall and Biddle (2001) also investigated processes of change as well as self-efficacy, pros and cons. The authors found that selfefficacy and pros significantly increased and cons decreased across the stages. However, only stage differences between PC and P as well as those between P and A were significant (Marshall & Biddle, 2001). Overall, both meta-analyses raise doubt that the stages are rather pseudo-stages of an underlying continuum. In general, critiques of stage models have questioned the existence of stages or whether stages are not just arbitrary divisions of an underlying continuum (Sutton, 2005; Weinstein et al., 1998). A stage model actually exists if, in different variables, discontinuity patterns are observable (Sutton, 2005). This would mean that there is a discontinuity in the degree to which variables act on different stages. Individuals at a particular stage should have different characteristics in comparison to those individuals in other stages. These discontinuity patterns may consist of mean differences in some stages and no mean differences in other stages, or an increase from one stage to the next and a decrease to the one thereafter (Weinstein et al., 1998). Nonlinear trends in cognitions at different stages would indicate such a discontinuity pattern, as found by Armitage, Povey, and Arden (2003) as well as Lippke and Plotnikoff (2006). Consequently for behavior change, the influence of certain variables is dependent on the stage in which a person is. As this has yet to be examined in longitudinal studies, this study has been performed to achieve that goal.

DISCONTINUITY OF PREDICTION BETWEEN STAGES Depending on the stage, different social-cognitive variables may be more or less influential. These different influences should be detected in discontinuity patterns (i.e., nonlinear trends across stages; Sutton, 2005; Weinstein et al., 1998). In particular, the assumptions of the TPB for each TTM stage are displayed in Figure 1 and are described in what follows. According to the TPB, intention mediates the influences of social-cognitive variables on behavior in all stages. Which social-cognitive variables have an impact depends on the stage in which a person is.

PREDICTION PATTERNS

653

FIGURE 1 Hypotheses on the interrelations between the variables at the different stages. Note. C D significant interrelation; 00 D no interrelation; Subj. Norm D subjective norm; PBC D perceived behavioral control. Reported in the order precontemplation, contemplation, preparation, action, and maintenance.

The TTM’s processes of change (POC), which are assumed to be the main predictors of behavior change, can be divided into cognitive POC and behavioral POC, which have stage-specific influences (Prochaska, DiClemente, & Norcross, 1992). The TTM hypothesizes that the cognitive processes are more relevant for stage transition in the inactive stages (Prochaska et al., 1992). Cognitive POC as pragmatic relief and emotional arousal are comparable to the TPB’s attitude as a motivating factor. Thus higher attitude should increase the intention to change behavior in the inactive stages of behavior change (PC, C, and P). Cognitive POC as social liberation share similarities with the TPB’s perceived subjective norm. Consequently, perceived social norm is assumed to be mainly important to initiate the changing process (PC and maybe C stage). According to the TTM, the behavioral POC are more relevant for behavior change in the active stages (Prochaska et al., 1992) because for people who have begun performing the behavior, other stimuli become more important. One component of the POC is stimulus control, which has similarities with the TPB’s PBC. Individuals have to perceive themselves to be able and in control to perform a behavior. However, this only helps to increase the intention if the decision to change is made (P, A, and M): Only if a person wants to change his or her belief in his or her own ability actually helps to make that change. Direct impact of PBC on behavior is only expected in the stages where the behavior is performed (A and M). Only one empirical study could be found testing the TPB constructs for prediction of stage transition. However, Courneya, Plotnikoff, Hotz, and Birkett (2001) could not find support for predicting stage progression with PBC.

654


Findings on attitude and intention revealed that both were predictive for most stage movements, with attitude being most reliably predictive in the PC stage and intention in the A and M stage. Subjective norm predicted forward movement out of the PC stage only (Courneya et al., 2001). No other study testing TPB’s variables for stage-specific prediction pattern could be found. Thus, this study aims to fill the gap, testing not only stage predictions, but linear patterns with stages as a moderator.

PURPOSE AND HYPOTHESES In general, discontinuity patterns should be indicated by significant differences in means and relations of the variables as described previously. No study was found testing the architecture of the TPB within the stages of TTM and, therefore, the study focuses on this issue. Furthermore, it attempts to elaborate the discontinuity patterns in latent means and to examine the hypothesized relations of constructs of the TPB variables within the stages of change of the TTM. At first, the invariant factorial structure of the psychometric instruments is tested. The first hypothesis is that the items comprising a particular instrument operate equivalently (are invariant) across the different stage groups, for (a) factor loadings and (b) covariances. Second, hypotheses on the discontinuity patterns in latent means are the following: (a) PBC, subjective norm, attitude, intention, and physical activity behavior will be lowest in the PC stage and highest in the M stage. The following discontinuity patterns are expected: (b) No significant differences in attitude, PBC, and behavior between individuals in PC and C; (c) the C stage and the P stage will be similar in terms of behavior; (d) the P stage and the A stage are hypothesized to be different in terms of intention, attitude, and PBC, as well as similar in subjective norms; and (e) the A stage and the M stage should be similar in subjective norms, attitudes, and intentions. Third, the stages should moderate the influence of the social-cognitive variables on intention and the translation of intention into behavior. This shall be indicated by unique (not invariant) regression paths in the specified causal structure of the TPB across the TTM stage groups over time. In particular, we specified 25 interrelations (standardized regression weights) to be positive or zero as follows: (a) in the PC stage and (b) in the C stage, intention is correlated with former attitude and subjective norm, and not to former PBC. In (c) the P stage intention is associated with PBC and attitude, and not with subjective norm. In (d) the A stage and (e) the M stage, intention is related to PBC, and PBC is correlated with physical activity. (f) In all stages intention and physical activity are associated. Overall, it should be demonstrated that stages are not just arbitrary distinctions of an underlying continuum.

PREDICTION PATTERNS

655

METHOD Participants and Procedure A longitudinal survey using random digit dialing of Hawaii’s noninstitutionalized adult population was conducted from February to April 2002 (Waksberg, 1978). Trained interviewers, assisted by a computer-aided telephone interview system, conducted a 20-min survey. The person aged 18 or older who had the last birthday was asked to complete the interview to provide randomization within the household. The University of Hawaii Committee on Human Subjects approved the procedures and informed consent was obtained over the telephone. Skip patterns and out-of-range responses were automatically controlled by the system. A total of 62,436 Hawaiian telephone numbers were called for the interviews. Of these 9,129 were nonresidential numbers and 34,039 had no answer, were disconnected, or relocated. Further 12,605 refused to participate in the study and 2,931 individuals were ineligible (under 18, language barrier, did not plan to be in Hawaii for 2 years). An additional 270 persons were excluded because of missing data. The final data set consisted of N D 3,462 individuals. These persons were contacted again after a 6-month period (Time 2). Of those, 2,390 took part at Time 2 (69%). All Time 1 individuals were contacted for a 12-month assessment (Time 3, independent of whether they took part at Time 2). At Time 3, 1,957 (57%) individuals provided data. Overall, 1,831 (53%) individuals provided data at all measurement points (61% female; age M D 46.43, SD D 16.32).

Measurements Participants were asked a series of demographic questions, including age, sex, height, weight, education attained, income level, marital status, ethnic identification, language spoken at home, and perceived health. Measurements were taken on physical activity, fruit and vegetable consumption, and tobacco usage. Only the physical activity-related measures are described in this study. Stage of change for physical activity. Participants were classified into one of five stages: (a) Precontemplation-Do not engage in regular physical activity and no intentions to do so in the next 6 months; (b) Contemplation—Do not engage in regular physical activity but intend to do so in the next 6 months; (c) Preparation—Do not engage in regular physical activity but intend to in the next month; (d) Action—Currently engaging in regular physical activity, but for less than 6 months; and (e) Maintenance—Currently engaging in regular physical activity for 6 months or more (Nigg, 2002).

656


Attitude was measured using a 7-point bipolar adjective scale suggested by Ajzen and Fishbein (1980). The statement that preceded the adjectives was “I think regular physical activity is: : : : ” Two items addressing the instrumental aspect of attitude (bad-good, foolish-wise) were combined to one scale. The two-item scale had good internal consistency (.759–.911), as reported for each stage group in Table 1. Subjective norm was measured by two items such as “Most people who are important to me think I should be physically active on a regular basis : : : ” and scored on a 5-point scale that ranged from 1 (disagree a lot) to 5 (agree a lot; Courneya, Nigg, & Estabrooks, 1998). Internal consistency was acceptable (.694–.795). PBC was measured by two questions such as “I have a lot of control over the number of times I am physically active” and scored on a 5-point scale that ranged from 1 (disagree a lot) to 5 (agree a lot; Courneya et al., 1998) and the intercorrelations of the two items are reported in Table 1. Although the reliability is only based on two items and seemed rather small (.440–.641), it is comparable to other studies (Rhodes, Matheson, & Blanchard, 2006). Intention was assessed using a scale adapted from Courneya et al. (1998). An example item is “I intend to be physically active at least four times a week : : : ” with responses on a 5-point scale that ranged from 1 (disagree a lot) to 5 (agree a lot). Internal consistency was moderately acceptable (.616–.776), as reported in Table 1. Physical activity behavior was assessed by two items from the International Physical Activity Questionnaire (IPAQ) short form and addressed time spent engaging in moderate and vigorous activity per week. High reliability and good validity of the IPAQ short form have been demonstrated in 12 countries in validating the subject measure with object data such as accelerometers (Craig et al., 2003). The low reliability coefficients found in this study (.238–.460, as

TABLE 1 Reliability Coefficients (Interitem Correlation) for the Scales in the Five Subsamples

Sample PC C P A M

Attitude (Time 1)

Subjective Norm (Time 1)

PBC (Time 1)

Intention (Time 2)

Physical Activity (Time 3)

.893 .911 .860 .759 .801

.780 .795 .694 .707 .756

.641 .572 .440 .554 .601

.776 .675 .709 .674 .616

.238 .388 .285 .460 .351

Note. PBC D perceived behavior control; PC D precontemplation; C D contemplation; P D preparation; A D action; M D maintenance. Time 2 was 6 months after Time 1. Time 3 was 12 months after Time 1.

PREDICTION PATTERNS

657

shown in Table 1) are not ideal, although they are comparable to those from other studies (Craig et al., 2003). The small interrelations between the two items apparently indicated that the two behaviors capture different aspects of behaviors, which are both relevant for describing physical activity. Although low reliabilities are comparable to other studies, they could have influenced the results. Using only one item could have caused different results. However, using two or more items had the advantage of latent analyses, and therefore this study remained using the two items. Data Analysis SEM with latent variables was employed to investigate the pattern of relations within the overall data set for several reasons. First, the underlying theoretical order among the factors and relations among predictors can be tested, and SEM has been suggested to be a good method to test the TPB (Hankins et al., 2000). Second, a multisample structural equation model analyzes invariances across the subsamples. Invariances may be analyzed in the measurement of the theoretical constructs, in the relations among theoretical constructs, in the regression paths in a specified causal structure, and in the latent means of constructs in a model. Third, if the independent variables in a regression analysis are moderately to highly interrelated, there may be multicollinearity problems. Finally, modeling with latent variables tests the relations among factors with reduced measurement error (Tabachnick & Fidell, 2006). This is especially important if scale reliabilities are moderate (Bentler, 1990). The AMOS Graphics was selected because this is a very advantageous method that is rarely used (Byrne, 2004). Multigroup structural equation modeling. A sequence of nested models ranged from an unconstrained multisample model with the parameters freely estimated across subsamples, to more parsimoniously nested models that include different levels of equality constraints (Kenny, 2002). The following models were estimated in this study: (a) Model 1: noninvariant, unconstrained model (unrestricted model); (b) Model 2: equal factor loading across the subsamples (measurement equivalent model); (c) Model 3: Model 2 constraints plus equal factor variance and covariances; (d) Model 4: Model 3 constraints plus equal regression paths; and (e) Model 5: Model 4 constraints plus equal factor residuals (fully constrained). The equality of variances and covariances is subsequently specific to Model 2 because the other constraints rely on assumptions of invariant measurements. Models 4 and 5 refer to the latent construct level. This level deals with more substantive hypotheses about how the subsamples may differ and are similar, respectively, in their perception of variable relations. Therefore, the most parsimonious model does not vary significantly from the unrestricted model when

658


examined in comparing the regression paths and the latent means (Byrne, 2001, 2004). Model fit. Structural equation models (see Figure 1) were analyzed with AMOS using maximum likelihood estimates for each subsample (Byrne, 2001). The overall fit of the resultant models was assessed using a number of goodnessof-fit indexes representing absolute, comparative, and residual aspects of fit, specifically ¦2 =df , Tucker-Lewis Index (TLI), comparative fit index (CFI), and root mean squared error of approximation (RMSEA). When the theory underlying the model indicates that a moderating relation among predictors may vary by specific population subgroups, as the stage groups, MSEM is preferable. A single ¦2 goodness-of-fit statistic evaluates a set of complex models, one for each group. To validate the usual assumptions that groups are equivalent, subsamples can be required to have identical estimates for all parameters (a fully constrained model). Differences among the groups can be evaluated for their appropriateness by freeing special parameters (allowing the groups to vary). The theoretical model is separately applied to each subgroup and then the invariance analyses can be set up. Before the invariance models are estimated, it must be established that the model without any invariances (i.e., a model that is different in each group) is reasonable. This model can be used as a basis of assessment of more constrained models. The constraints are placed in a sequence of nested models. To compare the models, the ¦2 difference test and the TLI can be used to test the equality constraints (Byrne, 2001; Kenny, 2002). If the difference between the ¦2 s is not statistically significant, then the statistical evidence points to no cross-group differences between the constrained parameters. The differences in the TLI up to .05 are considered trivial in practical terms. If the ¦2 difference is statistically significant and the TLI is .05 or larger, then the evidence of cross-group inequality exists (Byrne, 2001). For the test of significant regression paths and significant differences across the subgroups, one-tailed tests with p .05 were used, because directed hypotheses were stated (Figure 1). SEM was performed using AMOS 5. Reliability and dropout analyses were performed using SPSS 12.0.1.

RESULTS Distribution of Study Participants Across the Stages Participants who provided data at Time 1 (N D 3,462) had the following stage distribution: PC, n D 730 (20.7%); C, n D 157 (4.5%); P, n D 556 (15.8%); A, n D 240 (6.8%); and M, n D 1,779 (50.6%; see Table 2). The sample of participants with data at measurement points Time 1 and Time 3 and no missing

PREDICTION PATTERNS

659

TABLE 2 Stage Prevalences at Time 1 and Time 3 Stage at Time 3 Stage at Time 1 PC C P A M Total

PC

C

224 (11.6%) 24 (1.2%) 74 (3.8%) 25 (1.3%) 131 (6.8%) 478 (24.8%)

27 (1.4%) 16 (0.8%) 37 (1.9%) 10 (0.5%) 33 (1.7%) 123 (6.4%)

P 69 22 107 24 114 336

(3.6%) (1.1%) (5.6%) (1.2%) (5.9%) (17.4%)

A

M

No Participation at Time 3

20 (1.0%) 11 (0.6%) 34 (1.8%) 21 (1.1%) 55 (2.9%) 141 (7.3%)

79 (4.1%) 19 (1.0%) 82 (4.3%) 50 (2.6%) 618 (32.1%) 848 (44.0%)

311 65 222 110 828 1,536

Total 730 (21.8%) 157 (4.8%) 556 (17.3%) 240 (6.7%) 1779 (49.4%) 3462 (100%)

Note. PC D precontemplation; C D contemplation; P D preparation; A D action; M D maintenance. Time 3 was 12 months after Time 1.

values at the questions on the stage variables (n D 1,926) had the following stage distribution at Time 1: PC, n D 478 (24.8%); C, n D 123 (6.4%); P, n D 336 (17.4%); A, n D 141 (7.3%); and M, n D 848 (44%). A total of 1,536 dropped out and were therefore not classified (see Table 2). Individuals’ stages at Time 1 and Time 3, and whether they dropped out, are reported in Table 2. This pattern is significant, ¦2 (20, N D 3,462) D 530.09, p < .01. All cells were filled (see Table 2), so the precondition for investigating the TPB across the TTM stages—that individuals move forward and backward-was met. Participants in PC, P, and M at Time 1 were most likely to remain in the same stage at Time 3. The biggest fraction of individuals at Time 1 in the A stage moved to the M stage. Participants in C at Time 1 were most likely to move to PC, P, or M at Time 3. Dropout Analysis To test whether participants who provided data at all measurement points (n D 1,831) differed from individuals who dropped out at one measurement point, the two groups were compared with regard to stage, demographics, and socialcognitive variables at Time 1. Participants who provided data at all measurement points were older, reported slightly lower PBC and intention, and reported that they were less physically active than dropouts. Women, individuals with a higher income, and those with a college degree were more likely to participate (all p < .05). No differences in number of persons per household, body mass index, attitude, and subjective norm (all p > .20) and stage at Time 1 (p D .08) were found. Because the differences exist between persons who completed all measurement points and the ones who dropped out after answering the questionnaire at Time 1, the subsequent analyses report the results from the entire sample who answered the questions at Time 1. Multiple imputation methods dealt with all

660


missing values to prevent results from a selected sample. The full information maximum likelihood (FIML) estimation is such a theoretically based approach and therefore has several advantages over other missing treatment procedures (e.g., listwise or pairwise deletion) and imputation methods (e.g., listwise or pairwise estimates). The FIML estimation is based on computing the likelihood for the observed portion of all individuals’ data, and with that accumulating and maximizing it, which is provided by AMOS (cf. Byrne, 2001). All analyses were additionally run with only participants who had answered all measurement points. In the case where substantial differences emerged they are reported. MSEM To examine whether the single sample models were adequate, the structural model was separately tested in each group. Only if the hypothesized model is adequate in all subsamples can the model work well in the multisample analyses. Therefore, the hypothesized model shown in Figure 1 was tested in each sample. Goodness-of-fit indexes for the five subsamples are shown in Table 3. In practical terms, the hypothesized model represented the data well. Although the ¦2 was statistically significant in three groups, the other fit indexes showed good or moderate model fit with ¦2 /df smaller than 2.5, TLI and CFI of .98 or greater, and RMSEA smaller than .08. Post-hoc analyses and subsequent

TABLE 3 Goodness-of-Fit Indexes for the Five Stage Subsamples for the Whole Sample and the Subgroup of Individuals Who Participated at All Measurement Points (Time 1, Time 2, and Time 3) Sample PC C P A M

n

2

df

2 /df

p

TLI

CFI

RMSEA

730 393 157 86 556 312 240 121 1,779 878

39.466 29.011 52.432 51.985 44.860 50.742 65.281 72.380 27.841 35.065

27 27 27 27 27 27 27 27 27 27

1.462 1.074 1.942 1.925 1.661 1.879 2.418 2.681 1.031 1.299

.057 .360 .002 .003 .017 .004