Pse Reed2006 - [PDF Document] (2024)

7/27/2019 Pse Reed2006

1/38

Psychology of Sport and Exercise 7 (2006) 477514

The effect of acute aerobic exercise on positive activatedaffect:

A meta-analysis

Justy Reeda,, Deniz S. Onesb

aDepartment of Health, Physical Education, and Recreation,Chicago State University, 9501 South King Drive,

Chicago, IL 60628, USAbDepartment of Psychology, Elliot Hall, 75E. River Road, University of Minnesota, Minneapolis, MN 55455,USA

Received 27 November 2004; accepted 23 November 2005

Available online 9 March 2006

Abstract

Objective: The purpose of this meta-analysis was to examine theeffect of acute aerobic exercise on self-

reported positive-activated affect (PAA). Samples from 158studies from 1979 to 2005 were included

yielding 450 independent effect sizes (ESs) and a sample size of13,101.

Method: Studies were coded for moderators related to assessmenttime, exercise variables such as intensity,duration, and dose(combination of intensity and duration), and designcharacteristics. The analysis

considered multiple measures of affect and corrected forstatistical artifacts using Hunter and Schmidt

[(1990). Methods of meta-analysis: Correcting error and bias inresearch findings. Newbury Park: Sage;

(2004). Methods of meta-analysis: Correcting error and bias inresearch findings (2nd ed.). Thousand Oaks:

Sage] meta-analytic methods.

Results: The overall estimated mean corrected ES( dcorr) andstandard deviation (SDcorr) were .47 and .37,

respectively. Effects were consistently positive (a) immediatelypost-exercise, (b) when pre-exercise PAA

was lower than average, (c) for low intensity exercise o1539%oxygen uptake reserve (%VO2R), (d) for

durations up to 35 min, and (e) for low to moderate exercisedoses. The effects of aerobic exercise on PAA

appear to last for at least 30 min after exercise beforereturning to baseline. Dose results suggest the

presence of distinct zones of affective change that moreaccurately reflect post-exercise PAA responses thanintensity orduration alone. Control conditions were associated with reductionsin PAA ( dcorr :17,SDcorr .25).

ARTICLE IN PRESS

www.elsevier.com/locate/psychsport

1469-0292/$ - see front matter r 2006 Elsevier Ltd. All rightsreserved.

doi:10.1016/j.psychsport.2005.11.003

Corresponding author. Fax: +1 773 995 3644.

E-mail address: [emailprotected] (J. Reed).

7/27/2019 Pse Reed2006

2/38

Conclusion: The typical acute bout of aerobic exercise producesincreases in self-reported PAA, whereas

the typical control condition produces decreases. However, largeSDcorr values suggest that additional

variables, possibly related to individual differences, furthermoderate the effects of exercise on PAA.

r 2006 Elsevier Ltd. All rights reserved.

Keywords: Exercise; Affect; Well-being; Meta-analysis;Doseresponse

Introduction

The majority of research on the exerciseaffect relationship hasexamined the impact of exercise

on negative psychological states (e.g., Gauvin & Spence,1996; McAuley & Rudolph, 1995). Since

1981, more than 70 reviews have been published and the vastmajority of them, narrative and

quantitative, have found that exercise reduces self-reportedanxiety and depression (e.g., Biddle,2000; Brosse, Sheets, Lett,& Blumenthal, 2002; Landers, & Arent, 2001). However,health is not

merely the absence of disease and negative affect, but acondition of physical and psychological

well-being as well (McAuley, 1994; USDHHS, 1996, p. 141).

There is ample evidence to support the practical importance ofwell-being, defined as positive

mood, engagement, life satisfaction, and meaning (Seligman,2002). For example, well-being and

positive mood predict job satisfaction and productivity (George& Brief, 1992; Hersey, 1932;

Miner, 2001), and marital satisfaction (Rogers & May, 2003).Higher well-being and positive

mood also correlate with better physical (Ostir, Markides, Black& Goodwin, 2000) and mental

health (Diener & Seligman, 2004), lower all-cause mortality(Fiscella & Franks, 1997), greater

longevity and lower rates of non-fatal heart attack (Kubzansky,Sparrow, Vokonas, & Kawachi,

2001), lower physiological stress reactivity (Smith, Ruiz, &Uchino, 2001), and improved immunefunction (Cohen, Doyle, Turner,Alper, & Skoner, 2003). In addition, studies show thatpeople

who report higher well-being exercise more compared to thosereporting lower well-being (e.g.,

Lox, Burns, Treasure, & Wasley, 1999). In sum, people withhigher levels of well-being are

healthier and function more effectively (Diener & Seligman,2004).

This study addresses the affective component of well-being. Forthe purpose of this paper,

affect is defined as the quality of a subjective experiencerelative to the two independent

dimensions of valence and activation or what Russell describesas core affect (Russell, 2003). For

most researchers, affect includes self-reportable states such ashappiness, elation, tension, and

relaxation (Russell & Carroll, 1999). Self-reported affectcan be described as a circumplex formed

by two dimensions of activation (activateddeactivated) andvalence (positivenegative). Thismeta-analysis specifically examinespositive activated affective states as defined by the upperright

quadrant of Fig. 1. Subscales from self-report instruments suchas the Energy subscale of the

ActivationDeactivation Adjective Checklist (AD-ACL; Thayer,1996) and the Positive Affect

subscale of the Positive and Negative Affect Schedule (PANAS;Watson, Clark & Tellegen, 1988)

have been shown to occupy this quadrant (see Yik, Russell, &Feldman Barrett, 1999). Although

this area of the circumplex and its associated affectiveconstructs has been named pleasant

activated (Yik et al., 1999), positive activation (Watson,Wiese, Vaidya, & Tellegen, 1999),

activated pleasant (Larsen & Diener, 1992), and energy(Thayer, 1996), we refer to these affective

states as positive activated affect (PAA).

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514478

7/27/2019 Pse Reed2006

3/38

Why study PAA in the context of exercise? First, theoretically,there is evidence that changes in

self-reported PAA could be a function of an evolutionaryadaptive behavioral facilitation system

(BFS; Depue & Iacono, 1989; Depue, Luciana, Arbisi, Collins& Leon, 1994) that mediates goal-

directed approach behaviors (Depue & Collins, 1999; Depue,et al., 1994; Tomarken & Keener,

1998). Dopamine pathways are associated with the BFS (Depue& Collins, 1999) and recent data

suggest that DNA sequence variations in dopamine receptor genesare related to self-reportedphysical activity levels (Simonen etal., 2003).

Second, PAA appears to facilitate and reward behaviors mediatedby the BFS (Watson,

2000) whereas low levels of PAA are associated with depressedmood (Mineka, Watson, &

Clark, 1998). From an evolutionary viewpoint, exercise might beconsidered a behavioral

means to obtain resources such as food and should increasefeelings of vigor and energy, thereby

providing both a reward and incentive to repeat the activity.Indeed, narrative reviews con-

clude that exercise improves PAA (e.g., Ekkekakis &Petruzzello, 1999; Gauvin & Spence,

1996) and individual studies report that these improvements aremore consistent than changes

in depression and anxiety (e.g., Gauvin, Rejeski, & Norris,1996; Thayer, 1996; Watson,

2000).Third, several investigators have suggested that affectivechanges related to exercise are an

important part of exercise adherence (Sallis & Hovell, 1990;Thayer, 1996; Wankel, 1993). The

psychological states associated with exercise adherence maytherefore be related to changes in

positive and not just negative affect. Unfortunately, there arevery few guidelines available to

assist health professionals in planning and prescribing physicalactivity to increase the adoption

and maintenance of exercise programs, largely due to the lack ofdata available to develop them

(Dishman, 2001).

The purpose of this meta-analysis is to provide data on theeffect of acute aerobic exercise

on PAA relative to three main questions. First, how doesbaseline PAA influence the magnitude

ARTICLE IN PRESS

(-) Valence (+)

(+)

Activation

(-)

Positive

Activated

Negative

Deactivated

Positive

Deactivated

Negative

Activated

Fig. 1. A circumplex model of self-reported affect. Adapted fromYik et al. (1999).

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 479

7/27/2019 Pse Reed2006

4/38

of change in post-exercise PAA? Second, how do various levels ofexercise intensity, dura-

tion, and combinations of intensity and duration (dose) affectthe magnitude of change in

post-exercise PAA? Third, what is the magnitude of PAA changeacross various post-exercise

assessment times? That is, how long do the acute effects ofaerobic exercise on PAA last?This study improves and extendscurrent knowledge and provides quantitative answers not

found in prior narrative reviews. To this end, we tested thefollowing potential moderator

variables.

Potential moderators

Baseline PAA

Acute studies (e.g., Focht, 2002; Reed, Berg, Latin, & LaVoie, 1998; Rejeski, Gauvin, Hobson,& Norris, 1995), chronicstudies (e.g., Blumenthal, Emery, & Rejeski, 1988; Simons&

Birkimer, 1988; Wilfley & Kunce, 1986), and quantitativereviews (e.g., Craft & Landers, 1998;

North, McCullagh, & Tran, 1990) show that participants withless positive or more negative pre-

exercise affect report greater post-exercise improvementcompared to those with higher baseline

scores. The relation appears to hold for active and sedentaryparticipants (e.g., Reed et al., 1998),

but has yet to be quantified for PAA across various self-reportscales. From a theoretical and

practical standpoint, Thayer (1996) proposed that exerciseserves as a self-regulatory strategy

to improve low mood and energy and delay the urge to snack orsmoke (Thayer, Peters,

Takahashi, & Birkhead-Flight, 1993). Also, based on the lawof initial value (Wilder, 1957) those

with less positive baseline affect and energy might be expectedto improve more because they have

more room for improvement. Thus, we hypothesize that there wouldbe larger effects for lowerbaseline scores.

Exercise intensity

Ekkekakis and Petruzzello (1999) reviewed 31 studies and found ageneral inverse relation

between intensity and post-exercise PAA. Walking increases PAA(e.g., Ekkekakis, Hall, Van

Landuyt, & Petruzzello, 2000; Thayer, 1987a) while maximalexercise produces deceases (e.g.,

Pronk Crouse, & Rohack, 1995). On the other hand, othershave suggested that optimal

benefits occur following moderate, but not low or high intensityexercise (Berger & Motl,

2000). These suggestions point to the notion of either acritical intensity, i.e., threshold stimulus(Raglin & Morgan,1985) or inverted-U (Ojanen, 1994) relation between intensity andaffective

benefit. However, high-intensity exercise may result inaffective improvement by way of a

rebound effect whereby the post-exercise affective staterebounds from an affective

decline during exercise (see Bixby, Spalding, & Hatfield,2001). Theoretically, intensity appears

to be a key variable moderating post-exercise affective response(Ekkekakis, 2003) and is of

practical importance as a potential exercise-related barrier tothe adoption and maintenance of

exercise (Dishman, 2001). Because the intensity literatureappears mixed, it is tentatively

hypothesized that there will be a general inverse relationbetween intensity and post-exercise PAA

improvement.

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514480

7/27/2019 Pse Reed2006

5/38

Exercise duration

Exercise of at least 20 min has been proposed for PAAimprovement (Berger & Motl, 2000).

However, individual studies (e.g., Blanchard, Rodgers, Courneya,& Spence, 2002; Petruzzello &Landers, 1994; Rudolph &Butki, 1998) and quantitative reviews (e.g., North, et al.,1990;

Petruzzello, Landers, Hatfield, Kubitz, & Salazar, 1991) donot appear to support this threshold

duration. Shorter bouts (10 min) might result in better exerciseadherence than longer bouts

(e.g., Jakicic, Wing, Butler, & Robertson, 1995) and lack oftime is a barrier to exercise

participation (Trost, Owen, Bauman, Sallis, & Brown, 2002).The importance of shorter bouts

for adherence and psychological benefit should therefore beemphasized, but the effects of

duration on post-exercise PAA have not been quantified acrossstudies. Therefore, we

hypothesized that there will be no differential effect ofexercise time on PAA across typical

exercise durations (e.g., 1540 min).

Exercise dose

Dose is the product of exercise frequency, intensity, andduration (Kesaniemi, et al., 2001). For

acute exercise, dose is the product of intensity and duration.Establishing a doseresponse effect is

one form of evidence for evaluating whether the exerciseaffectrelationship might be interpreted

as causal (Mondin et al., 1996). Practically, exercising at thedose likely to improve affect may lead

to better exercise adherence (Dishman, 1995). An important pointabout dose is that the

strenuousness of any exercise bout is a function of intensityand duration (McArdle, Katch, &

Katch, 2001, p. 195) and therefore conclusions about exercisedose and affective change based on

either intensity or duration alone may be misleading (He, 1998).Generally, studies employing lowdoses (e.g., Thayer, 1987a) findshort-term increases in PAA and those examining extreme bouts

show decreases (e.g., Hassmen & Blomstrand, 1991). Based onthis information, we hypothesize

that there will be a general inverse relationship between doseand post-exercise PAA

improvement.

Post-exercise assessment time

Post-exercise assessment time is an important consideration. Forexample, when affect is

measured only once following a delay of several minutes, theresults are limited because any

immediate effects of the exercise bout may have dissipated(Ekkekakis & Petruzzello, 1999).Increases in PAA often peakwithin 5 min (e.g., Ekkekakis, et al., 2000; Petruzzello, Hall,&

Ekkekakis, 2001; Petruzzello, Jones, & Tate, 1997; Steptoe& Bolton, 1988; Steptoe, Kearsley &

Walters, 1993a), and remain significantly elevated abovebaseline for 20 (e.g., Bixby et al., 2001;

Van Landuyt, Ekkekakis, Hall, & Petruzzello, 2000) to 30 min(e.g., Steptoe & Bolton, 1988;

Focht & Hausenblas, 2001), or longer (e.g., Daley &Welch, 2004). With the exception of extreme

exercise, in which PAA is typically reduced after exercise(e.g., Acevedo, Gill, Goldfarb, & Boyer,

1996), the literature supports a pattern of initial improvementfollowed by gradually decreasing

effects. The consistency of this pattern has been noted for bothpositive and negative affect

(Ekkekakis & Petruzzello, 1999), but the magnitude of thispattern of change for PAA remains

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 481

7/27/2019 Pse Reed2006

6/38

unknown. Therefore we expected to find the largest effectswithin the first 5-min post-exercise and

lower effects thereafter.

Study quality and source

Study quality (internal validity) and source (published vs.unpublished) warrant examination.

While Eysenck (1994) and Slavin (1986) contend studies withmethodological faults should not be

meta-analyzed, others disagree (Dickersin & Berlin, 1992;Glass, 1983; Hunter & Schmidt, 2004,

p. 382). Excluding poorer quality and unpublished studies mayproduce inflated results since

journals tend to accept methodologically superior studies andstudies with larger effects (e.g.,

Cook et al., 1992; Greenland, 1998; Hunter & Schmidt 1990,p. 509). Meta-analyses on exercise

and negative affect have found either larger effects for studieswith moderate levels of internal

validity (North et al., 1990), or have failed to finddifferences between validity levels (Long &

Stavel, 1995; Petruzzello, et al., 1991). Findings are alsoequivocal for source. Differences betweenpublished and unpublishedstudies have been shown in some meta-analyses (Craft &Landers,

1998; North et al., 1990; Petruzzello et al.), but not in others(Arent, Landers, & Etnier, 2000;

Long & Stavel, 1995). The conflicting results and lack ofcomparative data for PAA preclude

justification of a formal hypothesis for these two potentialmoderators.

Method

Description of the database

Searches were performed for studies on mood or affect inrelation to aerobic exercise to includeactivities such as aerobicdance, walking, jogging, running, swimming, and cycling.Relevant

English-language studies from 1979 to December 2005 wereidentified using computer databases

(PsychINFO, ERIC, Medline, SPORTDiscus, World Cat, Pub Med, andDissertation Abstracts

International), manual searches of narrative reviews publishedbetween 1980 and 2003,

quantitative reviews (e.g., McDonald & Hodgdon, 1991; North,et al., 1990; Petruzzello et al.,

1991), and books (e.g., Biddle, 2000; Seraganian, 1993; Thayer,1996; Watson, 2000). Reference

lists of published articles, theses, and dissertations wereexamined for additional studies. We

contacted the authors of all studies with missing informationfor the calculation of effect sizes

(ESs), the standardized unit of analysis in meta-analysis. Atotal of 47 authors were contacted.

Twenty-six authors responded, the others either did not respondor could not be located. Of thosewho responded, 13 provided therequested data.

We included studies that assessed affect with scales consideredrepresentative of PAA. To avoid

confounding theoretically separate constructs, we did notinclude scales representative of positive

deactivated affect (lower right quadrant in Fig. 1). The rightnow time set represented

85.35% of the samples. Although time set was not reported in14.65% of the samples, studies

using extended time sets such as the today or past month wereexcluded. Table 1 lists the

instruments and subscales contributing to the meta-analysis.

The psychometric work of Yik et al. (1999) serves as a logicalanchor for the inclusion of the

affect subscales in Table 1. Briefly, Yik et al., usingstructural equation modeling, found that

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514482

7/27/2019 Pse Reed2006

7/38

ARTICLE IN PRESS

Table 1

Mood instruments, subscales and effect sizes (ESs) contributingto the meta-analysis

Instrument name (Reference) Subscale ESs

ActivationDeactivation Adjective Checklist Energy 137

(AD-ACL; Thayer, 1986)

Brunel University Mood Scale Vigor 9

(BRUMS; Terry, Lane, & Fogarty, 2003)

Exercise-Induced Feeling Inventory Positive Engagement 72

(EFI; Gauvin & Rejeski, 1993) Revitalization

Eigenzustands-Skala Activation 5

(The Self-State Scale; Nitsch, 1976)

Modified Morris Mood Questionnaire Positive Mood 6

(Williamson et al., 2001)

Mood Adjective Checklist Pleasantness 4

(MACL; Nowlis, 1965) Activation

Positive and Negative Affect Schedule Positive Affect 84

(PANAS; Watson et al., 1988)

Profile of Mood States Bi-Polar Energetic-Tired 6

(POMS-BI; Lorr & McNair, 1988)

Profile of Mood States Long Form Vigor 15

(POMS-LF; McNair et al., 1992)a

Profile of Mood States Short Form Vigor 23

(POMS-SF; McNair et al., 1992)b

Stress Arousal Checklist Arousal 2

(SACL; Makay, Cox, Grenville, & Lazzerini, 1978)

Subjective Exercise Experiences Scale Positive Well-Being140

(SEES; McAuley & Courneya, 1994)

UWIST Mood Adjective Checklist Energetic Arousal 2

(UMACL; Matthews et al., 1990)

Visual Analogue Mood Scale Joy 6

(VAS; Folstein & Luria, 1973) Euphoria

Note: For instruments with two subscales, ESs were computed foreach and the average of the two used in the meta-analysis. ThePOMS-Bi, SACL, UMACL, AD-ACL (bipolar version) and VAS are scoredusing bipolar format. We

assumed that bipolar terms acted as reciprocal pairs such that adecrease in unpleasant deactivated states (e.g., tired,

drowsy) resulted in a corresponding increase in pleasantactivated states (e.g., active, lively) allowing for comparable

ESs with the other unipolar scales in the database. Bipolarscales comprised 6% of the total number of ESs.aAdditional versionsof the POMS included: Profile of Mood States-A (POMS-A; Terry etal., 1999); Profile of Mood

States-C (POMS-C; Terry et al., 1996); Profile of MoodStates-Japanese Version (Yokoyama et al., 1990).bThe analysis alsoincluded the POMS-SF versions of Grove and Prapavessis (1992) andShacham (1983).

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 483

7/27/2019 Pse Reed2006

8/38

FeldmanBarrett and Russells Activated (Feldman Barrett &Russell, 1998), Watson, Clark, and

Tellegens Positive Affect (PA; Watson et al., 1988), ThayersEnergy (Thayer, 1986), and Larsen

and Dieners Activated Pleasant (Larsen & Diener, 1992), allfell within the quadrant of the affect

circumplex we call PAA (see Yik et al. (1999, Fig. 6). Wedetermined the number of affect termsfrom these subscales thatmatched terms in the subscales of the instruments listed in Table 1and

found the following: ActivationDeactivation Adjective Checklist(AD ACL; Thayer, 1986), 5 of

5, Brunel University Mood Scale (BRUMS; Terry et al., 2003), 4of 4, Exercise-Induced Feeling

Inventory (EFI; Gauvin & Rejeski, 1993), 3 of 6 (2 termsfrom Positive Engagement and 1 term

from Revitalization), Eigenzustands-Skala (The Self-State Scale;Nitsch, 1976), 5 of 6, Modified

Morris Questionnaire (Williamson, Dewey, & Steinberg, 2001),6 of 8, Mood Adjective Checklist

(MACL; Nowlis, 1965), 7 of 8, Positive and Negative AffectSchedule (PANAS; Watson et al.,

1988), 10 of 10, Profile of Mood States Bipolar (POMS-BI; Lorr& McNair, 1988), 5 of 7, Profile

of Mood States Long Form (POMS-LF; McNair, Lorr, &Droppleman, 1992), 6 of 8, Profile of

Mood States Short Form (POMS-SF; McNair et al., 1992), 5 of 5,Stress Arousal Checklist(SACL; Makay, Cox, Burrows, &Lazzerini, 1978), 6 of 8, Subjective Exercise Experiences Scale

(SEES; McAuley & Courneya, 1994), 1 of 4, UWIST MoodAdjective Checklist (UMACL;

Matthews, Jones, & Chamberlain, 1990), 4 of 4, and VisualAnalogue Mood Scale (VAS; Folstein

& Luria, 1973), 3 of 3. For other versions of the POMS wefound: POMS-A ( Terry, Lane, Lane, &

Keohane, 1999), 4 of 4, POMS-C (Terry, Keohane, & Lane,1996), 4 of 4, Japanese POMS

(Yokoyama, Araki, Kawakami, & Takesh*ta, 1990), 6 of 8,POMS-SF (Grove & Prapavessis,

1992), 5 of 6, and POMS-SF (Shacham, 1983), 5 of 6.

Two subscales deserve further justification. First, althoughonly the terms upbeat, refreshed, and

revived of the 6 terms from the EFI Positive Engagement andRevitalization subscales were

matches, Rejeski, Reboussin, Dunn, King, and Sallis (1999) arguefor the inclusion of these

subscales in the PAA quadrant (see Rejeski, et al., 1999, p.98). To strengthen this argument,Positive Engagement andRevitalization correlate with PANAS PA at r :69 (po:001) andr :58(po:001), respectively (see Gauvin & Rejeski, 1993). Second,for the SEES, while only theterm strong on the Positive Well-Being(PWB) subscale was a match, we believe the other terms,

great, positive, and terrific, qualify as positively activated.In support of our position that PWB

should occupy the PAA quadrant, McAuley and Courneya (1994)found that PWB correlated

with PANAS PA (r :71, po:01). Finally, Lox, Jackson, Tuholski,Wasley, and Treasure (2000)found high correlations between the SEESPWB and EFI Revitalization (r :81, po:01), SEESPWB and EFI PositiveEngagement (r :78, po:01), and EFI Revitalization andPositiveEngagement (r :86, po:01) indicating a substantial amountof common variability between

these subscales and suggesting a common underlying construct.Thus, based on logical, semantic,and statistical grounds, webelieve the subscales in Table 1 are representative of PAA.

We excluded studies where investigators introduced confoundersbefore the post-exercise

assessment such as mental arithmetic (e.g., Szabo et al., 1993),or manipulation of efficacy

expectations (e.g., McAuley, Talbot, & Martinez, 1999).Table 2 lists studies of potential relevance

that were eventually excluded. Studies with similar authors wereevaluated for sample overlap.

When sample overlap occurred, studies with more complete datawere included. Duplicate articles

were excluded.

The database included 158 studies and 450 independent ESs with asample size of 13,103 (total

number of ESs was 711). Study year ranged from 1979 to 2005 (M1997:10; SD 5.20). Age

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514484

7/27/2019 Pse Reed2006

9/38

ARTICLE IN PRESS

Table 2

Relevant studies not included in the meta-analysis

Study Reason for exclusion

Allen and Desmond (1987) Dependent t test; prepost correlationnot reported.

Annesi (2002a) Insufficient data for ES calculation.

Barabasz (1991) POMS Vigor not reported.

Berger et al. (1988) POMS Vigor nor reported.

Berger and Owen (1988) POMS Vigor means and SDs notreported.

Berger and Owen (1992b) POMS Vigor not reported.

Bird (1981) Insufficient data for ES calculation.

Boutcher and Landers (1988) POMS Vigor not reported.

Butki et al. (2003) Affect assessed only post-exercise.

Courneya and McAuley (1993) Affect assessed onlypost-exercise.

Daniel et al. (1992) Dependent t test; prepost correlation notreported.

Fallon and Hausenblas (2005) Positive affect notassessed.Fillingim et al. (1992) Post-exercise POMS confounded byimagery

manipulation.

Friedman and Berger (1991) POMS Vigor means and SDs notreported.

Gurley, Neuringer, and Massee (1984) Dependent t test; prepostcorrelation not reported.

Hobson and Rejeski (1993) Post-exercise PANAS confounded bymental stressor.

Hochstetler et al. (1985) Affect assessed only pre-exercise.

Jin (1989) POMS means and SDs not reported.

Jin (1992) POMS means and SDs not reported.

Jerome et al. (2002) Post-exercise affect confounded byefficacy

manipulation.

Johnson (1994) POMS Vigor not reported.

Kell et al. (1993) Insufficient data for ES calculation.

Kerr and Svebak (1994) SACL SDs not reported.

Kerr and Vlaswinkel (1993) SACL SDs not reported.

Kilpatrick, Hebert, Bartholomew, Hollander, and

Stromberg (2003)

Affect assessed only post-exercise.

LaCaille et al. (2004) Affect assessed only post-exercise.

Laguna and Dobbert (2002) Affect assessed onlypost-exercise.

Lane et al. (2005) Post-exercise BRUMS not reported

Lichtman and Poser (1983) POMS Vigor SDs not reported.

McAuley et al. (1999) Post-exercise affect confounded byefficacy

manipulation.

McAuley et al. (2000) Insufficient data for ES calculation.

McGowan et al. (1985) Affect assessment confounded by mentalstressor

McIntyre et al. (1990) PANAS SDs not reported.Moore (1997)Post-exercise affect confounded by distraction and

mastery information.

Morris and Salmon (1994) Positive affect SDs not reported.

Naruse and Hirai (2000) Prepost UWIST MACL scores notreported.

Nelson (1994) Positive affect not reported.

Otto (1990) Post-exercise affect assessment confounded byinduced

mood.

Oweis and Spinks (2001) AD-ACL means and SDs not reported.

Pernell (1997) Unable to determine exercise type.

Parente (2000) Insufficient data for ES calculation.

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 485

7/27/2019 Pse Reed2006

10/38

was reported in 91.80% of the studies with a mean age of 24.47(SD 11.64). Gender wasreported in 97.00% of the studies. Maleparticipants comprised 22.40%, female participants

34.90%, and mixed gender 39.70% (mixed gender was defined ashaving less than 75% of either

gender). Participant source was provided in 97.80% of thestudies. College students represented

62.40% of samples, community samples 19.00%, athletes 7.80%,clinical participants 3.90%,

and mixed samples of faculty, staff and students, 4.70%. Studiesconducted in the US represented

74.68% of the ESs, those from the UK, 10.73%, Canada, 4.29%,Australia and Japan, 3.43%,

Sweden, 1.72, Korea, 1.29, and Estonia, Greece, and Norway.43%.

Coding

Baseline PAA

We examined the influence of pre-exercise affect using thefollowing method. First, we copied

affect scale names, pre-exercise means, and sample size valuesfrom the original database to a

separate database then sorted by affect scale name. Because someaffect scales had a relatively

small number of pre-exercise means, we increased the number ofpre-exercise means for these

scales by adding pre-exercise means from studies not included inthe meta-analysis and from

normative data reported in test manuals. Distribution samplesizes ranged from 180 for the

ARTICLE IN PRESS

Table 2 (continued)

Study Reason for exclusion

Prusaczyk et al. (1992) Unable to determine when affect wasassessed relative to

exercise bout.

Rehor et al. (2001) Insufficient data for ES calculation.

Rejeski et al. (1991) Post-exercise POMS confounded withintroduction of

stress task.

Rejeski et al. (1991) AD-ACL SDs not reported.

Rocheleau et al. (2004) Positive affect not assessed.

Rosenfeld (1998) Prepost SEES means and SDs not reported.

Rudolph and McAuley (1996) Affect assessed onlypre-exercise.

Sakolfske et al. (1992) AD-ACL SDs not reported.

Steinberg et al. (1997) Dependent t test; pre-post correlationnot reported.

Steptoe and Bolton (1988) POMS Vigor SDs not reported.

Steptoe and Cox (1988) POMS Vigor SDs not reported.Steptoe etal. (1993b) Post-exercise POMS Vigor means and SDs notreported.

Szabo et al. (1993) Affect assessment confounded by mentalstressors.

Takenaka (1993) Insufficient data for ES calculation.

Thayer (1987a) AD ACL SDs not reported.

Thayer et al. (1993) Dependent t test; prepost correlation notreported.

Tredway (1978) Affect scale SDs not reported.

Turnbull and Wolfson (2002) Prepost POMS means and SDs notreported.

Tuson et al. (1995) Insufficient data for ES calculation.

Vasilaros (1988) POMS Vigor SDs not reported.

Watson (1988) Temporal sequence of exercise and affect notgiven.

Note: Contact the first author at [emailprotected] to obtain acomplete list of all studies reviewed.

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514486

7/27/2019 Pse Reed2006

11/38

POMS-LF (McNair et al., 1992) to 6 for the MACL (Nowlis, 1965)and the VAS (Folstein &

Luria, 1973). Next, in order to compare ESs associated withpre-exercise means from different

affect scales, we standardized the pre-exercise means for eachaffect scale by computing separate

sets ofz-scores with the sample-size weighted mean and SD fromthe respective scale. Scores frominstruments with more than oneversion or alternative scoring methods were converted to the

same scale before calculating z-scores. Each z score was thenplaced in the original database in a

new column next to the appropriate pre-exercise mean. Zscoreswere then sorted from low to high

and divided into three groups: less than .5z, .5z to .5z, andgreater than .5z. This procedureallowed for a comparison of ESsbetween samples of participants having average pre-exercise

affect scores either in the lower third (less than .5 z), middlethird (.5 z to .5 z), or upper third(greater than .5 z) of thedistribution for the affect scale on which they were assessed.

Exercise intensity and duration

Intensity was coded using the classification system of theAmerican College of Sports Medicine(ACSM). In this system,intensity can be classified as a percentage of oxygen uptakereserve

(%VO2R), a relative measure of intensity, which permitsconsistent coding of intensity whether

expressed as percent oxygen uptake (%VO2max), heart rate, orperceived exertion (Howley, 2001).

When necessary, we converted %VO2max to %VO2R using theappropriate equations (see Swain

& Leutholtz, 1997; Swain, Leutholtz, King, Haas, &Branch, 1998). The database allowed for the

formation of low (1539% VO2R), moderate (4059% VO2R), and high(6085% VO2R)

categories based on these guidelines (see Howley, 2001). Wecoded intensity as moderate for the 5

studies in the database where participants self-selectedexercise intensity. Duration was coded in

minutes of exercise, excluding the warm-up and cool-down.

Exercise doseWe quantified dose as the product of the relativeexercise intensity and duration. The following

four dose categories were formed from natural gaps in dosevalues: (a) low, 1030 min low

intensity to 720 min moderate intensity, (b) moderate, 3040 minmoderate intensity to 2030 min

high intensity, (c) high, 6090 min moderate intensity to 4060min high intensity, (d) very high,

1801400 min moderate to 300 min of high intensity exercise.

Post-exercise assessment time

Post-exercise assessment times were coded for all ESs. We sortedpost-exercise times and

created intervals based on natural breaks in the coded values:02, 510, 1530, and 401440 min

post-exercise. Additional ESs were available for this moderatorbecause all individual ESs foreach time interval were enteredinstead of being averaged as in the overall analysis.

Study quality and source

Based on questions concerning the use of a control group,randomization procedures, etc., we

coded the number of threats to internal validity thatinvestigators attempted to control, a higher

number indicating greater internal validity and study quality.The following threats were

considered: history, maturation, testing, statisticalregression, selection bias, experimental

mortality, compensatory rivalry, resentful demoralization,Hawthorne effect, demand character-

istics, halo effect, and expectancy effect (see Cook &Campbell, 1979; Thomas & Nelson, 2001).

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 487

7/27/2019 Pse Reed2006

12/38

Due to limited drop out rates, experimental mortality wascontrolled in 86% of the studies.

Maturation and selection bias were controlled in 63% of thestudies, followed by testing

(62.30%), history (51.30%), statistical regression (50.60%),compensatory rivalry, resentful

demoralization, Hawthorne effects and demand characteristics(13.60%), and halo andexpectancy effects (2.60%). To addresspublication bias, we coded source as an unpublished

masters thesis or doctoral dissertation, or as a publishedjournal article or abstract. It should be

noted that the database contained only two abstracts both fromthe same author who we

corresponded with concerning this information. Thus, we feelconfident the abstract data does not

bias the results of the source moderator analysis.

Coder reliability

The first author recoded 12 randomly selected studies 2 weeksafter the coding phase for data

relevant to the results (moderator variables, reliabilitycoefficients, sample sizes, and ESs). Datathat could be codedwithout error such as publication dates were not included toavoid

overestimating coder reliability (Kuncel, Hezlett, & Ones,2001). The agreement rate was 99.10%

(464 agreements out of 468 items). Two discrepancies involvedestimating the length of the warm-

up and cool-down resulting in exercise durations that differedby about 2 min between the recoded

and original values. An internal validity disagreement resultedfrom not recoding a threat

controlled for in one of the studies (the Hawthorne effect). Areliability difference occurred in a

study using the Exercise-Induced Feeling Inventory (EFI; Gauvin& Rejeski, 1993). This

inventory required Mosiers formula (Hunter & Schmidt, 2004,p. 438) to estimate reliability from

two subscales. The recalculated reliability differed triviallyfrom the original (87 vs. .89). Coding

errors were therefore minor and agree with prior research on thereliability of meta-analytic data

(Zakzanis, 1998).

Analysis

The data were analyzed with the Hunter and Schmidt (1990, 2004)meta-analytic method. This

method employs a random effects approach, which, unlike thefixed-effects model, allows for the

possibility that population effects vary from to study to study(Hedges, 1992). Because varying

population effects appear to be the rule rather than theexception for most real-world data, the

random-effects model is preferred to fixed-effects models andprocedures (Field, 2003; Hunter &

Schmidt, 2000). The Hunter and Schmidt method corrects ESs forthe attenuating effects of errorsuch as measurement unreliabilityand provides an estimate of the amount of ESvariance due to

sampling error and other artifacts.

Calculation of ESs

Mean values and pooled standard deviations (SDp) were used tocalculate ESs. The use of SDpresulted in Cohens d statistic (Cohen,1977). For within-subjects study designs, the following

within-subjects ESs were calculated: prepost treatment, prepostcontrol, pre-treatment vs. pre-

control and post-treatment vs. post-control. Forbetween-subjects study designs both within and

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514488

7/27/2019 Pse Reed2006

13/38

between-subjects ESs were calculated. The between-subject ESsincluded pre-treatment vs. pre-

control and post-treatment vs. post-control and thewithin-subjects ESs included prepost

treatment and prepost control. For studies having both withinand between ESs, the average

was entered to maintain statistical independence (Hunter &Schmidt, 2004, p. 431). Three r-valueswere converted to ds usingformulas described by Hunter and Schmidt (2004, Chapter 7).

Within ESs comprised 75.33% of the total number of ESs, between17.68%, average of

within and between 6.57%, and r-values .42%, respectively.Effect size calculations based on

paired t and F ratios from studies without the necessary meansand SDs were not included

because r-values were not reported in these studies. Calculationof ESs for paired-sample data

without appropriate r-values results in an upward bias of theeffect (Dunlap, Cortina, Vaslow, &

Burke, 1996).

Each ES was weighted by the study sample size and corrected formeasurement error using

the affect scale internal consistency reliability (a) reportedin the study. When the reliability was

not reported, the internal consistency reliability either fromthe validation study for the affectscale or from the appropriatetest manual was used. Descriptive statistics for the internal

consistency reliabilities for the main meta-analyses arepresented in Table 3. Further correc-

tions were made for small sample size bias, unequal sample size,and treatment by subject

interaction (Hunter & Schmidt, 2004, pp. 266, 279, 282).Sampling error variance was calcula-

ted separately for within and between ESs (Hunter & Schmidt,2004, p. 305, 370) and the aver-

age sampling error variance of a within and between ES wasentered when appropriate for a

particular study.

Most data sets contain errors, which can arise from varioussources such as transcriptional or

computational mistakes (Gulliksen, 1986). Some of these dataerrors are likely to be outliers

(Schmidt, et al., 1993) and any meta-analysis containing a largenumber of ESs should be

examined for outliers to eliminate cases of bad data (Hunter& Schmidt, 1990, p. 262). Unlikeother approaches, the Hunterand Schmidt method employs a random effects model that focuses

on the accurate estimation of the standard deviation ofpopulation ESs (SDcorr) because they play

an important role in the interpretation of the results of themeta-analysis (Hunter & Schmidt,

1990, p. 453). Unfortunately, the presence of a single outliercan bias the estimation of SDcorr by

producing variance above that predicted by sampling error andother artifacts (Schmidt, et al.,

1993). Therefore, we employed Tukeys (1977) method to identifyand omit outlier ESs prior to

the meta-analyses. This method is roughly equivalent to removingvalues greater than 2.5 SDs

from the mean, a common benchmark for outliers (Kirk, 1995, p.169).

ARTICLE IN PRESS

Table 3Descriptive statistics for internal consistencyreliabilities (a) for the main meta-analyses

Artifact distribution N K Mean a SD a Meanffiffiffi

a

pSD

ffiffiffi

a

p

Pre-exercise vs. pre-control 2666 66 .89 .02 .95 .01

Pre-post control 1666 66 .89 .04 .94 .02

Pre-post exercise 8362 230 .89 .03 .94 .01

Note: N, total number of participants; K, number of reliabilityvalues in the distribution; Mean a, average alpha

reliability; SD a, standard deviation of alpha reliabilities;Meanffiffiffi

a

p, average square root of alpha reliabilities; SD

ffiffiffi

a

p,

standard deviation of the square root of alphareliabilities.

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 489

7/27/2019 Pse Reed2006

14/38

Meta-analyses

We computed the following for all meta-analyses (includingmoderator analyses): total sample

size (N), number ofESs (K), mean sample-size weighted observedES( dobs), dobs 95% confidenceinterval (95% CI), mean sample-sizeweighted corrected ES ( dcorr), corrected standard deviation

(SDcorr), residual standard deviation (SDres), percent of dobsvariance due to sampling error

(%Vare), 90% credibility interval (90% CrI), and dcorr fail-safeN ( dfs). The dcorr fail-safe N

estimates the number of unlocated ESs with null results neededto reduce dcorr to the lowest

critical ES considered practically or theoretically important(Hunter & Schmidt, 2004, p. 500;

Orwin, 1983). The critical ESwas set at .20, considered a smalleffect (Cohen, 1988). The dcorr and

SDcorr were interpreted as best estimates of the populationparameters.

We employed a random effects model for the calculation of the95% CI, which provided an

estimate of the sampling error in dobs. The 90% CrI represents adistribution containing 90% of

the true population values of d. The SDcorr and the 90% CrI wereused to identify moderators.The width of the 90% CrI depends onSDcorr. If SDcorr is large relative to dcorr and the 90% CrI

includes zero, dcorr is the mean of several populationparameters, indicating the presence of

moderators. If the 90% CrI does not include zero, dcorrestimates a single population parameter

and moderators are not operating (Whitener, 1990). This intervalalso determines whether dcorr is

a generalizable effect. When the 90% CrI does not include zero,the magnitude of ESs may vary,

but 90% of the true ESs will retain a positive sign (or anegative sign for negative ESs) and

generalize across settings (Ones, Viswesvaran, & Schmidt,1993).

Overall meta-analyses

The first meta-analysis compared exercise and control groupsprior to treatment using between

and within ESs. We tested pre-activity equivalence because largedifferences may confoundconclusions about moderator variablesassociated with exercise groups. For this meta-analysis,

positive ESs indicated greater average PAA in exercise samples.The second meta-analysis tested

pre- to post-changes for control samples using within ESs.Attention controls such as lecture

contributed 90.62% and activity controls such as very lightexercise 6.25% of the control ESs,

respectively. Control type was not reported for 3.13% of thecontrol ESs. The third meta-analysis

examined prepost changes in PAA across exercise samples usingwithin, between, and average of

within and between ESs, the majority being within ESs. PositiveESs reflected increased PAA

relative to baseline. For all analyses, the average ESwasentered for studies with multiple post- or

post-exercise assessments to ensure independence and we used thestudy sample sizes as the weight

for the average ES (Hunter & Schmidt, 2004, p. 432).

Moderator analyses

As a first step, a moderator variable correlation matrix wasgenerated to explore the possibility

that moderators were substantially intercorrelated, making theresults difficult to interpret. Next,

the appropriate ESsubgroups for each moderator variable weremeta-analyzed. Moderators were

explored by examining differences between dcorr values andchanges in the SDcorr across

moderator subgroupings (Hunter & Schmidt, 2004, p. 293). The90% CrI values were used to

determine the generalizability of a subgroup effect, or tosuggest the presence of additional

unexamined moderators.

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514490

7/27/2019 Pse Reed2006

15/38

Results

Overall meta-analyses

Pre-test differences between exercise and control groups werevery small, dcorr :05(SDcorr .10). This indicates that on average,there was only .05 of a SD difference in self-reported PAA betweenexercise and control samples prior to experimental conditions. Itappears

safe to assume equivalence between groups on the dependentvariable at pre-test. The mean

corrected ES of.17 (SDcorr .25) for the control groupmeta-analysis indicated that controlconditions are associated withsmall decreases in PAA. The dobs for exercise groups was .45(95%

CI: .40 to .50) and the dcorr was .47 (SDcorr .37), which is arobust, moderate ES, nearly fourtimes that associated with controlsamples. The SDcorr suggests the presence of moderators and

the 90% CrI included zero indicating that the effects ofexercise on PAA do not generalize. Fail-

safe Ns of 48297 suggest good tolerance to availability bias.That is, 48 additional ESs with adcorr of .40 for pre-exercise vs.pre-control, and 117 additional ESs with a dcorr of .40 forprepost

control would have to be found and included in the analysis toincrease the current dcorr values to

.20, the critical ESwe set based on Cohens (1988) criterion. Forprepost exercise, 297 additional

ESs with a dcorr of .00 would have to be found and included toreduce the current dcorr to .20.

Results are presented in Table 4.

Moderator analyses

Results for moderator correlations are presented first, followedby baseline PAA, exercise

characteristics, study quality, and source. Table 5 displays themoderator correlations and

correlations between moderators and corrected ESs (dcorr).Tables 6 and 7 display results forbaseline PAA, exercisecharacteristics, post-exercise assessments time, study quality, andsource.

Moderator correlations

The majority of correlations were small to negligible,suggesting moderator variables were

unrelated. As anticipated, however, duration and dose weremoderately correlated indicating the

general influence of duration on dose. The number of threatscontrolled was inversely related to

ARTICLE IN PRESS

Table 4

Overall meta-analyses of the effects of acute exercise onpost-exercise positive activated affect

Analysis N K dobs 95% CI dcorr SDcorr SDres %Vare 90% CrIdfs

Pre-exercise vs. pre-control 2582 64 .05 .02 to .12 .05 .10 .1088.28 .08 to .18 48Prepost control 1602 63 .17 .23 to .10 .17 .25.25 18.54 .49 to .14 117Prepost exercise 8094 223 .45 .40 to .50.47 .37 .35 17.64 .01 to .94 297

Note: N, total sample size; K, number of ESs; dobs, meansample-size weighted observed ES; 95% CI, dobs 95%

confidence interval; dcorr, mean sample-size weighted correctedES; SDcorr, sample-sized weighted corrected standard

deviation; SDres, residual standard deviation; %Vare, percent ofdobs variance due to sampling and measurement error;

90% CrI, 90% credibility interval; dfs, dcorr fail-safe N withdcritical :20, dunlocated :00 for dcorr values above .20anddunlocated :40 for dcorr values below .20. Boldface entries arebest estimates of the population mean ES.

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 491

7/27/2019 Pse Reed2006

16/38

duration and dose. This shows that studies with greater internalvalidity (e.g., Petruzzello & Tate,

1997; Van Landuyt et al., 2000), typically conducted in a labsetting, tended to employ shorter

bouts, while most of the higher dose longer duration studieswere conducted outside a laboratory

with less experimental control (e.g., Odagiri, Shimomitsu,Iwane, & Katsumura, 1996). Mostmoderators were inverselyrelated to dcorr with an exception being the number of internalvalidity

threats controlled suggesting a trend toward higher ESs instudies with greater internal validity.

Baseline PAA

We had hypothesized larger effects for lower baseline scores.Individuals reporting less

favorable pre-exercise affect experienced greater affectiveimprovement than those reporting a

more favorable affect (e.g., Parfitt, Rose, Markland, 2000;Nabetani & Tokunaga, 2001). The dcorrfor the lower third of thedistribution was .63 (SDcorr .37), nearly twice the magnitude oftheother two categories and the 90% CrI did not include zero,indicating that lower than average

baseline scores produce generalizable increases in PAA. Thesmaller effects associated with higherbaseline scores were notconfounded by studies examining extreme bouts. The 90% CrI forthe

middle and upper third of the distribution included zeroindicating the presence of additional

moderators. The findings show that when baseline affect is belowaverage, aerobic exercise results

in consistent, generalizable increases in PAA, in line with ourhypothesis.

Exercise intensity

We had tentatively hypothesized a general inverse relationbetween intensity and post-exercise

PAA improvement. The dcorr of .57 (SDcorr .33) for low intensity(e.g, Ekkekakis et al., 2000;Thayer, 1987b) was nearly twice thatof moderate (e.g., Parfitt & Gledhill, 2004; Reed et al.,1998)

ARTICLE IN PRESS

Table 5

Intercorrelations between moderators and moderators andcorrected ESs

Moderator 1 2 3 4 5 6 7 8

1. Baseline PAA .06 .11 .12 .01 .02 .09 .20(167) (195) (166)(196) (197) (200) (200)

2. Exercise intensity .05 .03 .10 .04 .04 .11(180) (180) (179)(185) (185) (185)

3. Exercise duration .69 .20 .24 .08 .40(180) (205) (216) (216)(216)

4. Exercise dose .21 .26 .08 .44(179) (180) (180) (180)

5. Post-exercise assessment time .01 .09 .14(218) (218)(218)

6. Threats controlled .03 .28

(227) (227)7. Publication status .11

(230)

8. Corrected ES (dcorr)

Note: All moderators were correlated using actual coded valuessuch as time (min) for duration, except for publication

status where published 1 and unpublished 2. Values inparentheses are sample sizes.

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514492

7/27/2019 Pse Reed2006

17/38

and high (e.g., Springer, Bartholomew, & Loukas, 2003;Steptoe, et al., 1993a) lending support to

the tentative hypothesis of larger effects for lower intensityaerobic exercise. Based on the 90%

credibility intervals, effects for low intensity also generalizeacross settings while those for

moderate and high intensity do not. Large SDcorr values leaveroom for additional moderators,

especially for moderate and high intensities.

One possible moderator for high-intensity aerobic exercise isfitness level. Ekkekakis and

Petruzzello (1999) have suggested that affective benefitsfollowing high-intensity exercise occur

ARTICLE IN PRESS

Table 6

Moderator analyses for baseline PAA, exercise characteristics,and post-exercise assessment times

Analysis N K dobs

95% CI dcorr

SDcorr

SDres

%Vare

90% CrI dfs

Baseline PAA

o.5 z 2861 75 .60 .51 to .69 .63 .37 .35 18.97 .16 to 1.10 162.5z to .5 z 2159 66 .33 .25 to .41 .34 .30 .29 24.24 .04 to .72 474.5z 1686 49 .39 .30 to .48 .39 .32 .32 23.45 .02 to .79 48

Exercise intensity

Low 748 23 .54 .40 to .69 .57 .33 .31 22.71 .16 to .98 42

Moderate 2732 91 .34 .26 to .43 .35 .39 .38 14.70 .14 to .8570High 1679 60 .31 .20 to .43 .31 .42 .42 12.48 .22 to .84 29Notreported 2578 41 .49 .39 to .59 .53 .31 .29 21.69 .13 to .92 67

Exercise duration (min)

7 to 15 1317 33 .55 .41 to .68 .56 .38 .37 16.51 .07 to 1.055620 to 28 2063 69 .44 .36 to .52 .46 .31 .30 25.20 .07 to .8691

30 to 35 1940 62 .55 .46 to .64 .57 .33 .32 19.94 .15 to .99113

40 to 60 1323 35 .36 .24 to .48 .37 .31 .30 30.15 .02 to .7629475 331 12 .72 1.11 to .33 .72 .66 .66 5.28 1.56 to .13 55Notreported 1102 13 .36 .27 to .45 .38 .14 .13 41.72 .21 to .56 12

Exercise dose

Low 2447 73 .44 .36 to .52 .45 .30 .29 26.51 .06 to .84 92

Moderate 2146 82 .46 .37 to .54 .46 .34 .34 21.07 .02 to .91108

High 279 15 .09 .07 to .26 .09 .27 .27 36.31 .25 to .43 8Veryhigh 216 6 .98 1.31 to .66 .98 .37 .37 15.99 1.45 to .51 35Notreported 2754 41 .48 .38 to .58 .52 .32 .30 20.30 .11 to .92 65

Post-exercise assessment timea

0 to 2 3512 87 .60 .51 to .68 .61 .40 .39 13.89 .10 to 1.12179

5 to 10 3184 110 .41 .34 to .48 .43 .35 .33 17.10 .03 to .8812515 to 30 1744 57 .27 .17 to .37 .27 .34 .34 19.28 .16 to .712140 to 1440 677 29 .09 .03 to .21 .10 .31 .29 25.01 .28 to .4815Not reported 1033 14 .52 .40 to .64 .58 .21 .19 34.16 .32 to .8527

Note: N, total sample size; K, number of ESs; dobs, meansample-size weighted observed ES; 95% CI, dobs 95%

confidence interval; dcorr, mean sample-size weighted correctedES; SDcorr, sample-sized weighted corrected standard

deviation; SDres, residual standard deviation; %Vare, percent ofdobs variance due to sampling and measurement error;

90% CrI, 90% credibility interval; dfs, dcorr fail-safe N withdcritical :20, dunlocated :00 for dcorr values above .20anddunlocated

:40 for dcorr values below .20. Boldface entries are bestestimates of the population mean ES.

aPost-exercise assessment time in minutes (min).

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 493

7/27/2019 Pse Reed2006

18/38

only for fit participants. To test this claim, we recodedhigh-intensity ESs for fitness level (active

vs. sedentary) and found a difference in the predicteddirection: active dcorr :48 (SDcorr .39),sedentary dcorr :14(SDcorr .30). In sum, low-intensity effects were generalizable andlargerthan moderate and high-intensity exercise. Some caution iswarranted in interpreting the low-

intensity subsample results because the smaller K can result ingreater sampling error and a less

stable ES estimate.

Exercise duration

We had hypothesized that there would be no differential effectof exercise time on PAA across

typical exercise durations (e.g., 1540 min). All dcorr valuesfor bouts from 7 to 60 min were within

.20 of a SD, providing support for the hypothesis of nodifferential effect of exercise duration on

post-exercise PAA. Additional moderators may be operating due tothe relatively large SDcorrvalues. Generalizable effects were foundfor bouts ranging from 7 to 35 min, although the lower

bound 90% CrI values for bouts less than 30 min were close tozero. Bouts from 40 to 60 min

produce increases, (e.g., Bodin & Hartig, 2003; Janal, Colt,Clark, & Glusman, 1984; McInman &

Berger, 1993), but effects do not generalize. Exercise durationslonger than 75 min likely result in

decreased PAA (e.g., Hassmen & Blomstrand, 1991), but thesmaller K for this subsamplewarrants some caution in theinterpretation of the results.

The 3035 min duration produced the largest effect ( dcorr :57,SDcorr .33). To assess theextent to which higher ESs associatedwith studies using lower intensity bouts influenced this

generalizable result, we correlated ESs and intensity for thissubsample. The correlation was small

and in the opposite direction of the expected confoundinginfluence: r (60) .15, p .25, addingto the strength of thegeneraliziblity of this finding.

Ekkekakis and Petruzzello (1999) also point out that durationeffects may be difficult to

substantiate because many studies assess affect several minutesafter exercise, potentially allowing

duration effects to subside. To address this, we also analyzedduration using the same time

ARTICLE IN PRESS

Table 7

Moderator analyses for study quality and source information

Analysis N K dobs

95% CI dcorr

SDcorr

SDres

%Vare

90% CrI dfs

Threats controlled

12 2687 67 .28 .17 to .38 .30 .45 .42 5.13 .28 to .88 3334 126725 .47 .35 to .60 .49 .25 .24 43.61 .17 to .81 37

56 3202 107 .49 .43 to .55 .50 .29 .28 31.11 .14 to .87 162

710 489 16 .47 .23 to .71 .49 .48 .46 12.83 .12 to 1.09 23

Source

Unpublished 1173 44 .26 .14 to .38 .26 .37 .37 15.62 .22 to .7414Published 7082 184 .45 .39 to .51 .47 .39 .37 18.59 .03 to .97252

Note: N, total sample size; K, number of ESs; dobs, meansample-size weighted observed ES; 95% CI, dobs 95%

confidence interval; dcorr, mean sample-size weighted correctedES; SDcorr, sample-sized weighted corrected standard

deviation; SDres, residual standard deviation; %Vare, percentofdobs variance due to sampling and measurement error;

90% CrI, 90% credibility interval; dfs, dcorr fail-safe N withdcritical :20, dunlocated :00 for dcorr values above .20anddunlocated :40 for dcorr values below .20. Boldface entries arebest estimates of the population mean ES.

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514494

7/27/2019 Pse Reed2006

19/38

intervals with only the 87 ESs measuring affect within 2 minpost-exercise and found comparable

results: ESs ranged from .60 to .37 for bouts from 7 to 15 and4060 min, respectively, and all 95%

confidence and 90% credibility intervals were comparable.

Exercise dose

We had hypothesized a general inverse relation between dose andpost-exercise PAA

improvement. The dcorr values of .98 (SDcorr .37) for very high,.46 (SDcorr .34) formoderate, and .45 (SDcorr .30) for low doses,supports the hypothesis of an inverse relationbetween exercise doseand ES. The lower 90% credibility values for low and moderate dosesare

close to zero, indicating that while not likely to be negative,effects may be small in some

instances. Moderators influence high dose effects due to thelarge SDcorr and the 90% CrI

straddling zero. Very high doses are associated with reducedpost-exercise PAA that generalizes

across settings. That is, in most situations, very high doses ofaerobic exercise result in at least

temporary reductions in PAA below baseline scores (e.g., Hassmen& Blomstrand, 1991; Knapiket al., 1991). Note that the high andvery high dose subsample analyses are based on small sample

sizes (K) and should be interpreted with some caution.

Post-exercise assessment time

We had expected to find the largest effects within 5 minpost-exercise and lower effects

thereafter. PAA peaked shortly after exercise then declined, aresult in line with the hypothesis.

The reduction in dcorr from .61 (02 min) to .10 (401440 min) andthe associated change in SDcorracross intervals suggest amoderating effect of post-exercise assessment time. The immediatepost-

exercise increase generalized, but the 90% credibility intervalsfor the other subsamples indicate

that a declining pattern across the time intervals examinedmight not hold in all settings (e.g., Cox,

Thomas, & Davis, 2001; Tate, 1991) and may be dependent onunexamined moderators. The401440 min subsample results are based arelatively small K and should interpreted with some

caution. In sum, the results show that post-exercise assessmenttime is an important consideration

in the interpretation and comparison of these studies.

Study quality and source

There was no hypothesis for these two potential moderators. Apositive relationship emerged

between ESs and the number of threats controlled suggesting atrend toward greater effects for

studies attempting to control more threats to internal validity(e.g., Jarvekulg & Viru, 2002; Van

Landuyt et al., 2000). It should be noted, however, that thepotential moderating effect of study

quality is tempered by the almost complete overlap of the 90%credibility intervals. This resultindicates that population ESsacross the categories occupy nearly the same range of values.The

results for the 34 and 710 threats controlled subsamples shouldbe interpreted with some

caution because of the relatively small K in thesecategories.

There were 184 and 44 ESs associated with published andunpublished studies, respectively. Thedcorr for published studiesof .47 (SDcorr .39) was nearly double that for unpublished studiesat.26 (SDcorr .37). However, the overlapping 90% CrI indicate thatthe potential moderatingeffect of source is not as apparent at thepopulation level. The fail-safe N ( dfs) for unpublished

studies also points to the possibility of a file drawer effectinfluencing this moderator analysis.

That is, there is a reasonable possibility of 14 additionalunlocated theses or dissertations with

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 495

7/27/2019 Pse Reed2006

20/38

dcorr of .00 in file drawers that, if added to the analysis,would reduce the dcorr to the critical

value of .20.

Discussion

The results indicate that, in general, exercise is associatedwith increased PAA. This

improvement is nearly one half of a SD higher after exercisethan before ( dcorr :47). Thetypical control condition producesdecreases in PAA of about one fifth of a SD ( dcorr :17).Onaverage, exercise and control groups do not report differentpre-test levels of PAA ( dcorr :05).These findings provide supportfor the favorable effects of exercise on positively activated

affective states.

Studies with individuals having pre-exercise scores in the lowerthird of the distribution were

associated with greater increases in PAA than those in themiddle and upper thirds. Thedcorr forlower pre-exercise scores wasnearly twice that of the other two levels and generalizesacross

participants and settings. This difference held for observed andcorrected ESs, supporting the

hypothesis that baseline scores moderate PAA (see Rejeski etal., 1995). An important implication

of this finding is the practical application of exercise as aself-regulatory strategy to improve

feelings of energy and positive affect (Thayer, 1996).

Our results challenge suggestions of a moderate intensitythreshold for optimal PAA benefits

because moderate and high-intensity exercise produced similar,non-generalizable post-exercise

improvements that were smaller on average than those forlow-intensity exercise. For low-

intensity exercise, reviewers often cite Thayer (1987a) whofound that short bouts of brisk walking

significantly increased feelings of energy. Recently, Ekkekakiset al. (2000) replicated these

findings across different measures and exercise settings. Ourdata agree: low-intensity exerciseproduced generalizableimprovements, a finding with implications for exercise adherence.Overall,

the results agree with the narrative literature forlow-intensity exercise. We found little evidence to

support a moderate intensity threshold or inverted-Uhypothesis.

Exercise of at least 2030 min has been suggested as a durationthreshold for improvement of

PAA (Berger & Motl, 2000). This meta-analysis suggests thatshorter and longer bouts, even up to

60 min, result in affective improvement while decreases arelikely for durations longer than 75 min,

results that remained even when the effects were controlled forpost-exercise assessment time. In

agreement with quantitative reviews on negative affect (e.g.,Craft & Landers, 1998; Landers,

1997; North et al., 1990; Petruzzello et al., 1991), our resultsdo not provide evidence to support a

threshold hypothesis for typical exercise duration.In line withpublic health recommendations regarding dose to improve exerciseadherence (e.g.,

USDHHS, 1996) low doses were generally associated with improvedPAA. Identical results were

found for moderate doses, which may consist of 30 min of higherintensity exercise, as defined in

this meta-analysis. Low and moderate doses represent optimalzones for PAA change because

improvements tend to generalize. High doses represent anunstable zone. This level resulted in

nearly null effects on average, but may produce improvements ordecrements depending on

moderators such as fitness level (see Ekkekakis &Petruzzello, 1999) or other individual

differences related to the preference and tolerance of variousexercise intensities (see Ekkekakis,

Hall, & Petruzzello, 2005b). Another explanation for theinstability at high doses may be related

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514496

7/27/2019 Pse Reed2006

21/38

to the method of defining exercise intensity. For example,higher exercise intensities (e.g., X70%

VO2max) associated with higher doses may force many less fitparticipants to rely more heavily on

anaerobic processes, resulting in greater physiological andperceptual discomfort during, and less

positive affect post-exercise, compared to more fit individuals,even though the relative intensityof the exercise stimulus is thesame for all participants (see Bixby et al., 2001; Ekkekakis&

Petruzzello, 1999). This potential metabolic responsevariability within samples may have

produced smaller ESs in samples with a greater percentage ofparticipants exercising above the

aerobicanaerobic transition and larger ESs for samples with agreater percentage of participants

exercising at or below the aerobicanaerobic transition.Individual metabolic response variability

may therefore explain some of the instability around the nulleffect for high doses. A shift to

defining exercise intensity relative to the aerobicanaerobictransition may help clarify affective

responses to higher doses. Very high doses, bouts outside thenorm of a typical exercise session,

represent an aversive zone. These bouts are associated withsubstantial physiological and

psychological fatigue resulting in short-term affectivedecrements in nearly all situations.Dose results point to theimportance of understanding affective changes as a function of

intensity and duration together (He, 1998). In contrast to thegenerally positive effects found for

intensity and duration separately, a clear interaction effectoccurred with dose whereby corrected

ESs dropped dramatically from low and moderate to high doses.Inspection of the database

revealed that the decline in ESmagnitude started at 40 min, asintensity increased from ESs coded

as moderate intensity at 40 min (upper end of moderate dose) tothose coded as high intensity at

40 min (lower end of high dose), a result not apparent fromduration or intensity effects alone.

Explanations for this obvious drop in post-exercise PAA may berelated to peripheral and central

factors associated with fatigue such as changes in blood glucose(Coyle, Coggan, Hemmert, & Ivy,

1986), or brain serotonin (Davis & Bailey, 1997). Thepsychological consequences of these

changes may also vary with fitness level, but the significanceof this apparent transition requiresfurther investigation.

Post-exercise time points produced a pattern of immediateincrease followed by progressively

smaller effects. At the population level, the immediatepost-exercise effect generalizes, a result

consistent with narrative reviews (Ekkekakis, 2003; Ekkekakis& Petruzzello, 1999). However,

others have found improvements lasting anywhere from 1 (e.g.,Cox et al., 2001; Daley & Welch,

2004) to 4 h post-exercise (e.g., Thayer, 1987a). Our dataclarify these conclusions by showing that

sustained increases do not appear to generalize acrossparticipants and settings. For example,

Petruzzello et al. (2001) found that treadmill running producedimmediate increases in PAA for

high and low to moderately fit individuals, but affect remainedelevated above baseline during

30-min of recovery only in the high-fit group. The moderatorsassociated with different patterns ofpost-exercise affectiveresponse should be a continued area of examination in futureresearch.

Publication status results agree with some meta-analyses in theexercise psychology literature

(e.g., Craft & Landers, 1998; Petruzzello et al., 1991), butnot others (e.g., Arent et al., 2000; Long

& Stavel, 1995). Methodological weaknesses are likelyreasons for lack of publication. However,

similar to North et al.s (1990) meta-analysis, we found norelationship between the number of

threats controlled (internal validity) and publication status,indicating that the source bias in our

results is more likely related to the size of the effect thanstudy quality per se. This argues for the

importance of including both published and unpublished studiesto avoid positively biased

conclusions in meta-analyses (Begg, 1994).

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 497

7/27/2019 Pse Reed2006

22/38

The dual-mode model of exercise and affect (Ekkekakis, 2003)offers some theoretical

clarification for the intensity results. The model proposes thatintensity is a key variable that

produces changes in the salience of cognitive and physiologicalfactors that influence the pattern

of affective response during and after exercise. Cognitivefactors are proposed to dominate at lowand moderate intensities. Athigher intensities, hedonic tone (valence) decreases asphysiological

cues increase due to rising blood lactate levels. The model alsosuggests that low, moderate, or

high intensities can result in post-exercise affectiveimprovement. At low intensities due to

cognitive factors and mild increases in activation that may beperceived as pleasant, and at higher

intensities due to an affective opponent process (Solomon,1980), possibly driven by

endogenous opiates that quickly reverse the negative affectivevalence reported during exercise.

The results concerning exercise intensity from the presentmeta-analysis fit the model, as all

levels were associated with positive effects, especially for lowintensity, which generalized and was

larger on average than moderate or high intensities. Anadditional albeit very speculative

explanation for the low-intensity effects involves the notion ofan activity-stat, an innatebiological mechanism that promotes andrewards habitual spontaneous activity such as brisk

walking or play behavior (Rowland, 1998). The affective changesrelated to these activities may

serve as a reward to promote a balance between energy intake andenergy expenditure through

daily physical activity. Several lines of genetic researchappear to support the concept of biological

control of physical activity (e.g., Perusse, Tremblay, Leblanc,& Bouchard, 1989). If shown to

exist, such a system may help further clarify the generalizableincreases in PAA for low-intensity

exercise.

Moderate and high-intensity effects did not generalize,indicating the presence of unexamined

moderators. A possible moderator of moderate intensity exerciseis self-efficacy, a construct

shown to correlate with affective valence during moderateexercise (e.g., Ekkekakis, Hall, &

Petruzzello, 1999a). Blood lactate often begins to rise insedentary individuals during moderateexercise, increasing themetabolic strain and effort perception. The interpretation of thischallenge

may depend on self-efficacy with higher efficacy related toaffective improvement and lower

efficacy with affective decline during exercise. According tothe dual-mode model, either response

can lead to improved post-exercise affect: those who reportimproved valence during exercise via

cognitive factors and those reporting decreased valence viaphysiological factors related to an

opponent process (Solomon, 1980). However, because Solomon(1980) hypothesized that the

strength of the opponent process increases with repeatedexposure to the stimulus, for sedentary

participants with little prior exercise experience who reportnegative affect during exercise, the

opponent process may be weak, resulting in decreasedpost-exercise affect. Thus, some of the

unexplained post-exercise variability associated with moderateintensity effects may be due tothe heterogeneity of affectiveresponses during moderate exercise.

From the conclusions of narrative reviews (Ekkekakis &Petruzzello, 1999) and the results of

this meta-analysis, fitness level appears to moderatepost-exercise PAA for high-intensity exercise.

One explanation for the moderating effect of fitness, based onthe dual-mode model, may be the

perception of fatigue related to the increased salience ofphysiological cues such as respiration and

blood lactate. Unfit individuals may be more likely than fitindividuals to experience fatigue

associated with these physiological changes. Because fatigue (anunpleasant deactivated affective

state) exhibits bipolarity with PAA (see Yik et al., 1999),higher feelings of fatigue may result in

lower reported PAA. Thus, fitness level, possibly due todifferences in self-reported fatigue,

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514498

7/27/2019 Pse Reed2006

23/38

appears to be a moderator of post-exercise PAA particularly forhigh-intensity exercise. Taken as

a whole, our intensity results appear consistent with thedual-mode model.

A practical implication of this meta-analysis involves thedevelopment of guidelines for

improving exercise adherence and maintenance of health-relatedquality of life. Dishman (2001)notes the lack of data regarding thedose of physical activity for the development of these

guidelines. Feelings of energy and vigor are important aspectsof health-related quality of life

(Wilson & Cleary, 1995) and they tend to be good predictorsof health over time ( Dixon, Dixon, &

Hickey, 1993). An important question, however, has been theeffect that exercise has on these

positive affective states. This study offers quantitativeevidence that low to moderate doses

represent zones that improve energy and vigor across a range ofpersonal and situational

variables. The present results, the consistency with whichexercise is associated with increased

positive affect (e.g., Steinberg et al., 1998; Watson, 2000),and the tendency for people to choose

activities associated with positive affective experiences(Emmons & Diener, 1986), all point to the

potentially important role of PAA in exercise adherence.Thismeta-analysis also adds to the literature in several ways. First,findings are based on 158

studies and 450 independent ESs. A greater number of ESsimproves the validity of the meta-

analysis by increasing the accuracy of population estimates andenhancing the statistical power of

the moderator analyses (Hunter & Schmidt, 2004, Chapter 9).Second, we examined only PAA

rather than combine activated and deactivated positive affect,the objective being to eliminate the

confounding relative to the different post-exercise patterns ofchange in each of these distinct

affective states (Ekkekakis, 2003). Third, prior reviews havenot considered the influence of

statistical artifacts related to methodological imperfections inresearch studies. This is a limitation

as failure to correct for measurement error in the dependentvariable results in attenuated mean

ESs, while failure to correct for sampling error results inartificial variation in ESs not due to true

moderator variables (Hunter & Schmidt, 2004, Chapter3).There are a couple of limitations that deserve mention. First,this meta-analysis did not consider

affective responses during exercise. This poses a conceptuallimitation because affective changes

from pre to post-exercise are often not linear and the dynamicchanges that take place during

exercise allow for a more complete picture of the overallaffective response. Second, in some of the

moderator analyses, the number of ESs was small enough towarrant caution concerning the

stability of the estimates of effect. For example, Field (2001)has shown that both fixed and

random effects methods do not control the Type I error rate inestimates of the mean uncorrected

ES in meta-analyses with 15 or fewer studies. Any meta-analysisis limited by the number of

available studies in a given research domain which hasimplications for second-order sampling

error in meta-analyses (Hunter & Schmidt, 2004, p. 399400).For this reason, the results of someof the moderator analysesshould be considered preliminary. Even with this limitation,however, a

meta-analysis based on a theoretical framework can provide validconclusions, including accurate

estimates of the magnitude of an effect and the examination ofpotential moderator variables,

procedures that overcome problems associated with otherapproaches to understanding the data,

including the traditional narrative review.

Future studies using theory-driven approaches (e.g., Ekkekakis,2003) that test self-efficacy,

fitness, and other individual difference variables are needed assuch variables may account for a

sizeable portion of the true variance in post-exercise affectiveresponses. A very important

unanswered question is how these individual differences interactand change with repeated bouts

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514 499

7/27/2019 Pse Reed2006

24/38

of acute exercise. From a meta-analytic standpoint, becausemoderators only describe the

conditions under which ESs vary, additional theory-drivenresearch will not only increase the

number of studies for future meta-analyses, but provideadditional meaning to the results as well.

Affective responses during exercise also appear to play animportant role in the overallrepresentation of the affectivedynamics of acute exercise, but relatively few studies arecurrently

available.

In summary, the results indicate that increased PAA can last upto 30 min post-exercise.

Furthermore, the effects remain positive across personal andsituational settings immediately

post-exercise, when pre-exercise PAA is lower than average, forlow intensities, for durations up

to 35 min, and for intensity/duration combinations (doses)ranging from low to moderate as

defined in this meta-analysis. We did not find evidencesupporting a threshold hypothesis for

intensity or duration. In contrast, dose results suggest thepresence of distinct zones of affective

change that more accurately reflect post-exercise responsesacross the literature than intensity or

duration alone. In this regard, the dose analysis is unique andinformative. Beyond thesegeneralizable effects, however, largeSDcorr values suggest that additional variables, possibly

related to individual differences, moderate the effects ofexercise on PAA.

References

References marked with an asterisk indicate studies included inthe meta-analysis.

Acevedo, E. O., Gill, D. L., Goldfarb, A. H., & Boyer, B. T.(1996). Affect and perceived exertion during a two-hour

run. International Journal of Sport Psychology, 27, 286292.

Allen, M. E., & Desmond, C. (1987). Naloxone blocking ofrunning-induced mood changes. Annals of Sports Medicine,3(3),190195.

Annesi, J. J. (2002a). Effects of differing durations andintensities of cardiovascular exercise on aversion and feeling

states in new women exercisers. Perceptual and Motor Skills, 94,735738.

Arent, S. M., Landers, D. M., & Etnier, J. L. (2000). Theeffects of exercise on mood in older adults: A meta-analytic

review. Journal of Aging and Physical Activity, 8, 407430.

Barabasz, M. (1991). Effects of aerobic exercise on transientmood state. Perceptual and Motor Skills, 73, 657658.

Begg, C. B. (1994). Publication bias. In H. Cooper, & L. V.Hedges (Eds.), The handbook of research synthesis

(pp. 399409). New York: Russell Sage Foundation.

Berger, B. G., Friedman, E., & Eaton, M. (1988). Comparisonof jogging, the relaxation response, and group

interaction for stress reduction. Journal of Sport &Exercise Psychology, 10, 431447.

Berger, B. G., & Motl, R. W. (2000). Exercise and mood: Aselective review and synthesis of research employing the

Profile of Mood States. Journal of Applied Sport Psychology, 12,6992.

Berger, B. G., & Owen, D. R. (1988). Stress reduction andmood enhancement in four exercise modes: Swimming, body

conditioning, Hatha yoga, and fencing. Research Quarterly forExercise and Sport, 59(2), 148159.

Berger, B. G., & Owen, D. R. (1992b). Preliminary analysisof a casual relationship between swimming reduction:

Intense exercise may negate the effects. International Journalof Sport Psychology, 23, 7085.

Biddle, S. J. H. (2000). Emotion, mood and physical activity. InS. J. H. Biddle, K. R. Fox, & S. H. Boutcher (Eds.),

Physical activity and psychological well-being (pp. 6387).London: Routledge.

Bird, E. I. (1981). Sensitivity of the Activation-DeactivationCheck List to activation during physical activities.

Psychological Reports, 49, 375380.

* Bixby, W. R., Spalding, T. W., & Hatfield, B. D. (2001).Temporal dynamics and dimensional specificity of the

affective response to exercise of varying intensity: Differingpathways to a common outcome. Journal of Sport &

Exercise Psychology, 23, 171190.

ARTICLE IN PRESS

J. Reed, D.S. Ones / Psychology of Sport and Exercise 7 (2006)477514500

7/27/2019 Pse Reed2006

25/38

* Blanchard, C. M., Rodgers, W. M., Courneya, K. S., &Spence, J. C. (2002). Moderators of the exercise/feeling-state

relationship: The influence of self-efficacy, baseline, andin-task feeling states at moderate and high intensity exercise.

Journal of Applied Social Psychology, 32(7), 13791395.

Blumenthal, J. A., Emery, C. F., & Rejeski, W. J. (1988).The effects of exercise training on psychosocial functioningaftermyocardial infarction. Journal of Cardiopulmonary Rehabilitation,8, 183193.

* Bodin, M., & Hartig, T. (2003). Does the outdoorenvironment matter for psychological restoration gained through

running? Psychology of Sport and Exercise, 4, 141153.

Boutcher, S. H., & Landers, D. M. (1988). The effects ofvigorous exercise on anxiety, heart rate, and alpha activity of

runners and nonrunners. Psychophysiology, 25(6), 696702.

Brosse, A. L., Sheets, E. S., Lett, H. S., & Blumenthal, J.A. (2002). Exercise and the treatment of clinical depression in

adults. Recent findings and future directions. Sports Medicine,32, 741760.

Butki, B. D., Baumstark, J., & Driver, S. (2003). Effects ofa carbohydrate restricted diet on affective responses to acute

exercise among physically active participants. Perceptual andMotor Skills, 96, 607615.

Cohen, J. (1977). Statistical power analysis for the behaviorsciences (rev. ed.). New York: Academic Press.

Cohen, J. (1988). Statistical power analysis for the behavioralsciences (2nd ed.). Hillsdale, NJ: Erlbaum.

Cohen, S., Doyle, W. J., Turner, R. B., Alper, C. M., &Skoner, D. P. (2003). Emotional style and susceptibility tothecommon cold. Psychosomatic Medicine, 65, 652657.

Cook, T. D., & Campbell, D. T. (1979).Quasi-experimentation: Design and analysis issues for fieldsettings. Boston:

Houghton Mifflin.

Cook, T. D., Cooper, H., Cordray, D. S., Hartmann, H., Hedges,L. V., Light, R. J., Louis, T. A., & Mosteller, F.

(1992). Meta-analysis for explanation: A casebook. New York:Russell Sage Foundation.

Courneya, K. S., & McAuley, E. (1993). Efficacy,attributional, and affective re