Evaluation of the circumplex structure of the Activation Deactivation Adjective Check List before and after a short walk

Abstract

Background and purpose: The Activation Deactivation Adjective Check List (AD ACL; Thayer, 1989, The biopsychology of mood and arousal. New York: Oxford University Press) has been used in several studies to assess affective responses to bouts of physical activity. In recent years, researchers have suggested that the structure of this multidimensional measure should approximate a circumplex. This hypothesis was examined using a circumplex-specific confirmatory method.

Methods: Volunteers (n=165) completed the AD ACL before and after a short walk. The data were analyzed using Browne’s Circumplex models for correlation matrices. (Psychometrika, 57, 469–497) circular stochastic process model and CIRCUM software and the analyses were performed at the item level.

Results and conclusions: Before the walk, the circumplex provided a close fit to the data, whereas, after the walk, the fit was lower, but still reasonable. At neither time did items theorized to belong to one subscale become interspersed with items theorized to belong to an adjacent subscale. The AD ACL represents a satisfactory, albeit imperfect, option for the assessment of affective responses to physical activity from a circumplex perspective. In the future, closer fit to circumplex structure should be achieved by taking the specific structural postulates of the circumplex model into account from the beginning of the scale development process.

Introduction

The Activation Deactivation Adjective Check List (AD ACL; Thayer, 1989) has been used in several studies examining affective responses to bouts of physical activity during the past 20 years (e.g. Bird, 1981, Ekkekakis, Hall and Petruzzello, 1999, Ekkekakis, Hall, Van Landuyt, & Petruzzello, 2000, Hall, Ekkekakis and Petruzzello, 2002, Jerome, Marquez, McAuley, Canaklisova, Snook and Vickers, 2002, Oweis and Spinks, 2001, Saklofske, Blomme, and Kelly,1992, Tate and Petruzzello, 1995, Thayer, 1987a, Thayer, 1987b, Thayer, Peters, Takahashi, & Birkhead-Flight, 1993, Van Landuyt, Ekkekakis, Hall and Petruzzello, 2000). Yet, as is the case with many of the self-report instruments used in this area of research, the structure of this multidimensional measure in the particular domain of physical activity has not been investigated. The necessity of such investigations was underscored recently, following suggestions from both general psychology (Feldman Barrett and Russell, 1999, Russell and Feldman Barrett, 1999, Schimmack and Reisenzein, 2002, Yik, Russell and Feldman Barrett, 1999) and exercise psychology (Ekkekakis, Hall, Van Landuyt, & Petruzzello, 2000, Ekkekakis and Petruzzello, 2002, Hall, Ekkekakis and Petruzzello, 2002, Van Landuyt, Ekkekakis, Hall and Petruzzello, 2000) that the structure of the AD ACL should conform to the affect circumplex described by Russell, 1978, Russell, 1980) and Tellegen and his associates (Tellegen, 1985, Watson and Tellegen, 1985, Zevon & Tellegen, 1982). On the basis of several considerations derived from an analysis of what was termed the ‘affect measurement conundrum’ in exercise psychology, Ekkekakis and Petruzzello (2002) proposed that the circumplex model could provide a useful conceptual and measurement framework for investigating affective responses to acute exercise. They further noted that the AD ACL, although not originally designed and validated as a measure of the circumplex dimensions, could be used in this role. Given the paucity of multi-item instruments specifically designed as measures of the circumplex dimensions in the literature, an empirical evaluation of the degree to which the structure of the AD ACL approximates a circumplex is important to investigators interested in assessing affective responses to exercise from a circumplex perspective. Thus, the purpose of the present study was to evaluate whether a circumplex provides an adequate fit to AD ACL data collected before and after a short bout of walking by taking advantage of a recently developed, circumplex-specific statistical modeling method (Browne, 1992) and software (Browne, 1995).

The AD ACL has been used in conjunction with a variety of conceptual frameworks since its inception. It is, therefore, important to review the history of the measure and explain how a measure that was initially intended to assess the levels of an ‘activation continuum’ came to be viewed as a potentially useful measure of the dimensions of the circumplex model of affect.

The AD ACL was initially developed as a measure of a bipolar activation continuum ranging from extreme excitement to deep sleep (Thayer, 1967). An item intercorrelation matrix containing 28 items considered to be indices of activation and 21 items considered to be ‘nonactivation mood-descriptive’ items, derived from Nowlis’ (1965) mood measure, was factor-analyzed using the centroid method of extraction followed by a varimax rotation. Four of the 16 factors that emerged ‘loaded mainly with activation adjectives’ (Thayer, 1967, p. 665). These four factors were labeled ‘General Activation’ (lively, active, full-of-pep, energetic, peppy, vigorous, activated), ‘High Activation’ (clutched-up, jittery, stirred-up, fearful, intense), ‘General Deactivation’ (at-rest, still, leisurely, quiescent, quiet, calm, placid), and ‘Deactivation-Sleep’ (sleepy, tired, drowsy). Thayer’s (1967) initial interpretation of these factors was that they ‘roughly approximate four points on a hypothetical activation continuum’ (p. 668). Consequently, one of the first applications of the AD ACL attempted to demonstrate the advantages of this instrument over the Manifest Anxiety Scale (Taylor, 1953) as a measure of drive (Thayer & Cox, 1968), and subsequent applications focused on demonstrating relationships between the AD ACL and composites of psychophysiological indices of activation (Thayer, 1970, Thayer, 1971).

As early as 1970, however, Thayer started reassessing his belief in a single activation continuum. This led to the formulation of a model consisting of two bipolar activation dimensions. Dimension A was characterized by energy-sleep and Dimension B was characterized by tension-placidity (Thayer, 1978a, Thayer, 1985). The relationship between these two dimensions was theorized to vary as a function of one’s position along a continuum of energy expenditure. Thus, at high levels of energy expenditure, when the individual is experiencing high tension or high energy, the relationship between Dimensions A and B is theorized to be negative (i.e. increases in energy are associated with decreases in tension and vice versa). Conversely, at moderate and low levels of energy expenditure, the relationship between the two dimensions is theorized to be either close to zero or positive (i.e. energy and tension increase and decrease concurrently). This complex pattern of relationships led Thayer to propose that, although the model is mainly defined by the two bipolar activation dimensions, four separate unipolar factors, representing the two high and the two low poles of Dimensions A and B, may also be identified.

This revision of theoretical assumptions led to further investigation of the structure of the AD ACL (Thayer, 1978b), with exploration in two new directions. First, oblique instead of orthogonal rotations were selected, to account for the hypothesized possible relationships between factors. Second, a second-order factor analysis was conducted to examine whether the four first-order factors could be organized into two bipolar higher-order dimensions. The results of this analysis were consistent with the theoretical postulates. Specifically, General Activation and Deactivation-Sleep were negatively correlated (–0.49) as were High Activation and General Deactivation (–0.41). The second-order factor analysis showed that General Activation and Deactivation-Sleep loaded on one bipolar factor (with loadings of 0.74 and –0.63, respectively), whereas High Activation and General Deactivation formed another bipolar factor (with loadings of 0.65 and –0.55, respectively). A revised version of the AD ACL was thus developed by applying the same factor-analytic procedures to a sample of 20 adjectives (five from each first-order factor). With the exception of the items calm and at-rest which loaded on both the General Deactivation and the High Activation factor (in opposite directions), the resultant four-factor solution exhibited fairly simple structure. The correlation between General Activation and Deactivation-Sleep was –0.58 and the correlation between High Activation and General Deactivation was –0.50. The other factor intercorrelations were lower, ranging between –0.22 and 0.36. A second-order factor analysis showed that General Activation and Deactivation-Sleep formed one bipolar factor (with loadings of 0.76 and –0.73, respectively), whereas High Activation and General Deactivation formed another bipolar factor (with loadings of –0.68 and 0.69, respectively). Additional evidence for the structural validity of the AD ACL was provided in 1986 (Thayer, 1986).

A change in the terminology used to describe the components of the AD ACL was proposed in the late 1980s (Thayer, 1989). Specifically, Dimension A was renamed ‘Energetic Arousal’ (EA) and its two opposite poles were named Energy (formerly General Activation) and Tiredness (formerly Deactivation-Sleep). Dimension B was renamed Tense Arousal (TA) and its two opposite poles were renamed Tension (formerly High Activation) and Calmness (formerly General Deactivation). The new terminology will be used in the remainder of this article.

Although arousal formed the core of Thayer’s theoretical model, the implications of the model clearly extend beyond the domain of arousal. In fact, Thayer considered arousal a basic element of mood and behavior in general. Thus, the AD ACL has commonly been used as a measure of mood in recent years. This was justified by Thayer’s observation that each pole of the two activation dimensions of his model are usually charged with either positive or negative ‘hedonic tone’ (otherwise referred to as affective valence or pleasure–displeasure). Specifically, Energy and Calmness are typically associated with positive hedonic tone, whereas Tension and Tiredness are typically associated with negative hedonic tone.

Following these observations, it became clear that the bipolar dimensions of EA and TA were essentially compatible with two-dimensional models originally developed to describe affect and mood, such as Russell, 1978, Russell, 1980 affect circumplex and Tellegen and coworkers’ two-dimensional model of mood (Tellegen, 1985, Watson and Tellegen, 1985, Zevon & Tellegen, 1982). According to Thayer (1989), although the dimensions derived by other investigators have been labeled differently, ‘the dimensions themselves are descriptively very similar to Energetic and Tense Arousal’ (p. 61; also see pp. 133–134, 164). As noted earlier, this issue has resurfaced recently, as some authors proposed that the structure of the AD ACL can be represented as an affective circumplex (Feldman Barrett and Russell, 1999, Matthews, Jones, and Chamberlain, 1990, Russell and Feldman Barrett, 1999, Yik, Russell and Feldman Barrett, 1999.

Circumplex models posit that affective states are systematically interrelated and their relationships can be parsimoniously modeled by as few as two basic dimensions (see Larsen & Diener, 1992, for a review). These dimensions are theorized to be orthogonal and bipolar. The relationships among affective states can, therefore, be represented as a circle, with experientially similar states being close together on the perimeter of the circle and experientially antithetical states being located across from each other. Although there is some consensus on these basic postulates, there continues to be disagreement on a number of fronts. Perhaps the most controversial issue involves the determination of the most conceptually appropriate rotation of this two-dimensional system (note that any rotation is defensible from a statistical standpoint). Russell and his associates (Feldman Barrett and Russell, 1999, Russell, 1978, Russell, 1980, Russell, 1989, Russell, 1997, Russell and Feldman Barrett, 1999) have supported an unrotated version, with one dimension representing affective valence (pleasure versus displeasure) and the other representing activation (low versus high). Tellegen, Watson, and their associates (Tellegen, 1985, Watson and Clark, 1997, Watson and Tellegen, 1985, Watson, Wiese, Vaidya and Tellegen, 1999, Zevon & Tellegen, 1982), on the other hand, have argued in favor of a 45° rotation of this dimensional system. This results in two bipolar and orthogonal dimensions that combine valence and activation. Specifically, one dimension, labeled Positive Activation (PA; formerly Positive Affect) ranges from activated pleasant affect to unactivated unpleasant affect. The other, labeled Negative Activation (NA; formerly Negative Affect) ranges from activated unpleasant affect to unactivated pleasant affect (for a more detailed explanation and elaboration, see Ekkekakis & Petruzzello, 2001, pp. 213–221).

From a conceptual standpoint, the dimensions of EA and TA in Thayer’s model and the dimensions of PA and NA in Tellegen and Watson’s model are compatible (Thayer, 1989, Watson, Wiese, Vaidya and Tellegen, 1999, Yik, Russell and Feldman Barrett, 1999). Therefore, the AD ACL should be expected to conform to the two-dimensional PA-NA model described by Watson and Tellegen (1985; also see Tellegen, 1985). Initial factor-analytic data presented by Watson and Tellegen (1985) were consistent with this idea. Moreover, Nemanick and Munz (1994) have suggested that the AD ACL may provide a more complete assessment of the theoretical space defined by PA and NA, compared to the Positive and Negative Affect Schedule (PANAS), which is the measure of PA and NA developed by Watson, Clark, and Tellegen (1988). This is because, as others have pointed out (Egloff, 1998, Larsen & Diener, 1992, Mossholder, Kemery, Harris, Armenakis and McGrath, 1994), the PA and NA scales of the PANAS only assess the high-activation poles of the theoretically bipolar PA and NA dimensions, thus covering only one half of the PA–NA space (see Watson & Clark, 1997, for a response).

Although rudimentary analyses, based mainly on simple correlations and factor analyses, support the validity of the AD ACL as a measure of the two-dimensional affective space, these methods of analysis have been criticized on several grounds (Fabrigar, Visser and Browne, 1997, Sjöberg, Svensson and Persson, 1979, van Schuur & Kiers, 1994). Browne (1992) has proposed a solution based on a non-standard covariance structure model developed specifically for the circumplex (see Gurtman & Pincus, 2003, Tracey, 2000, for overviews of circumplex-specific methodologies). This model postulates that the pattern of correlations among variables can be represented as an ordering of the variables along the circumference of a circle. The model assumes that (a) the variance in observed scores on each measured variable is composed of common (among two or more variables) and unique variance, the latter likely to be due in part to measurement error, (b) the circumplex pattern of correlations refers to the correlations among common rather than observed scores, (c) common scores can be represented as points along the circumference of a circle, and (d) the correlation between two common score variables should be a function of the angle between the common score variables on the perimeter of the circle (Browne, 1992, Fabrigar, Visser and Browne, 1997). For example, in the theoretical case of error-free data, the angle of separation between two common score variables correlated at 0.707 would be 45° (i.e. a value estimated by the inverse cosine of 0.707). This can be represented as a Fourier series correlation function (Browne, 1992). It should be clear that empirical data may define an infinite number of possible correlation functions and that the vast majority of them will deviate from a perfect circumplex pattern. The range of possible correlation functions increases as one specifies more free parameters (correlation function weights) in the correlation function equation. Within certain constraints imposed to ensure that the solution will make conceptual sense (see Fabrigar et al., 1997, for a review), the model determines the one solution that best fits the observed data.

Although the mathematical aspects of the model were presented in the early 1990s (Browne, 1992), applications did not start to emerge until a computer program, named CIRCUM (Browne, 1995), was developed as an extension of another program, named AUFIT (Browne & Du Toit, 1992), which was designed for testing non-standard covariance structure models. CIRCUM yields indices of the goodness of fit of the model to observed data (using the Pearson product-moment item inter-correlation matrix as input). Following the postulates of Browne’s stochastic process model described above, the goodness of fit indices represent a measure of the extent to which the structure of the data is consistent with a model in which correlations among variables are a function of distance (angle) on the perimeter of a circle. The indices of fit provided by the program include the χ² and Steiger’s (1990) root mean square error of approximation (RMSEA), a measure of ‘badness of fit’ or, more specifically, a ‘measure of discrepancy per degree of freedom’ (Browne & Cudeck, 1992, p. 238). Note, however, that ‘there is nothing about the model that limits the researchers to use these indices—nearly any index of model fit used for more standard covariance structure models could be used to assess the fit of the circular stochastic process model with a Fourier series correlation function’ (Fabrigar et al., 1997, p. 195). In the present study, we based our interpretation on the RMSEA because of the well-documented advantages of this index, namely insensitivity to sample size and model complexity, robustness to non-normality, and sensitivity to model misspecification (Fan, Thompson and Wang, 1999, Fan and Wang, 1998, Hutchinson and Olmos, 1998, Olsson, Foss, Troye and Howell, 2000, Olsson, Troye and Howell, 1999, Sugawara and MacCallum, 1993). Following Browne and Cudeck (1992), a RMSEA value of about 0.05 or less is interpreted as indicative of close fit of the model in relation to the degrees of freedom, whereas a value of about 0.08 or less is interpreted as indicative of reasonable fit. Supporting evidence for these interpretations has been provided by the Monte Carlo study of Hu and Bentler (1999), who characterized a RMSEA value ‘close to 0.06’ as indicative of ‘relatively good fit’.

Furthermore, CIRCUM yields several additional useful parameter estimates, including the polar angles of common score variables (i.e. location on the circle in relation to a reference variable, whose position is set to 0°), estimates of the communality of each measured variable (i.e. the proportion of variance estimated to represent common variance), and the minimum common score correlation (i.e. the correlation between variables that are 180° apart). Confidence intervals for these estimates are also provided.

To date, only one study has used the CIRCUM to test the hypothesis that the circumplex can fit data collected with the AD ACL. Yik et al. (1999) reported that the angular locations of the Energy, Tension, Tiredness, and Calmness factors were 56° (49° to 62°), 136° (130–143°), 238° (231–245°), and 302° (296–308°), respectively (the estimates in parentheses represent the limits of the confidence intervals). These locations were relative to the location of an external variable, Pleasure, whose location was set to 0°. However, it should be noted that the model examined by Yik et al. included, in addition to the AD ACL scales, several other variables that were also theorized to fit within the circumplex structure, so the overall indices of goodness of fit reported by Yik et al. are not informative of the degree to which the structure of the AD ACL alone conforms to a circumplex. Furthermore, this analysis was done at the level of scale scores, not at the level of individual items, and the observed AD ACL scores that were used to derive the input correlation matrix were based on a composite of response formats and not on the standard AD ACL response format.

Thus, the purpose of the present study was to evaluate the fit of the circumplex, as specified in Browne’s stochastic process model, to AD ACL data collected before and after a short bout of walking at a self-selected pace. Previous studies using a similar stimulus have shown significant changes in AD ACL scale scores from before to after the activity (e.g. Ekkekakis, Hall, Van Landuyt, & Petruzzello, 2000, Saklofske, Blomme, and Kelly,1992, Thayer, 1987a, Thayer, 1987b, Thayer, Peters, Takahashi, & Birkhead-Flight, 1993). It was of interest, therefore, to examine the structure of the AD ACL both before and after the bout of walking.