7/27/2019 j.1539-6924.2010.01406.x

1/7

Risk Analysis, Vol. 30, No. 7, 2010 DOI: 10.1111/j.1539-6924.2010.01406.x

Changes in Risk Perceptions Prospectively Predict Changes

in Self-Reported Speeding

Stephen L. Brown

Risk perception theories posit that changes in risk perception prompt subsequent changesin risk behavior. Prospective studies using observations made at three time-points offer thecapacity to test this hypothesis by observing sequential changes in both risk perceptions andbehavior. A telephone survey was administered by random-digit dialing to 255 adult Aus-tralian drivers at baseline (T1), 6 weeks (T2), and 14 weeks (T3). During weeks 25, a risk-perception-based anti-speeding mass media campaign was conducted. The survey assessedrisk perception, operationalized as the proportion of time that driving at 70 km/h (43 mph)was perceived to be dangerous, and self-reported speeding behavior, defined as the frequencyof respondents driving 5 km (3 mph) faster than the legal speed limit in built up areas. HigherT2 risk perception predicted lower T3 self-reports of speeding after controlling T1 risk per-ception and T1 and T2 self-reported speeding. This can be interpreted as changes in riskperceptions between T1 and T2 predicting changes in speeding between T2 and T3. Furtheranalyses showed that increases in risk perception predicted lower subsequent self-reportedspeeding changes, but decreases in risk perception were unrelated to those changes. Riskperception changes were unrelated to recall of exposure to the media campaign. These find-ings support a dynamic view of the relationship between risk perception and self-reportedbehavior, and that risk perception theories can be applied to speeding.

KEY WORDS: Media campaign; risk perception; road safety; speeding

1. INTRODUCTION

When applied to health- and safety-related be-havior, risk perception theories propose that peo-ple move toward the initiation of behavioral changeafter they perceive that current behaviors carry aburden of risk.(1) The risk perception literature isextensively used as a basis for mass-reach interven-tions to control risk-taking, and meta-analyses andreviews show prospective links between risk per-ceptions and behavioral changes over a range ofhealth behaviors.(24) However, these links are of-ten weak and may not apply to all health and safety

Address correspondence to Stephen L. Brown, School of DentalSciences, University of Liverpool, Pembroke Place, Liverpool,UK L3 5PS; tel: 44 (0) 151 706 5108; fax: 44(0) 151 706 5809;slbrown@liverpool.ac.uk.

behaviors.(4) As risk perception theories are used tounderpin expensive and strategically important pub-lic health campaigns, it is crucial to develop a moresecure empirical underpinning for the idea that in-ducing increases in risk perception leads to positivebehavioral changes.

Road safety is one key public health area towhich risk perception theory has been applied. Dur-

ing 2007, 41,059 people were killed and 2.49 millioninjured in road crashes in the United States(5) and anestimated 1.2 million worldwide during 2004.(6) Case-control studies show that vehicles involved in crashesare likely to have been traveling at higher speedsthan other vehicles using the same roads under thesame conditions.(7) Speed control is now a major roadsafety priority throughout the world.(6)

1092 0272-4332/10/0100-1092$22.00/1 C 2010 Society for Risk Analysis

7/27/2019 j.1539-6924.2010.01406.x

2/7

Risk Perception and Speeding 1093

Some cross-sectional studies show support forthe predicted negative association between risk per-ception and speeding,(811) although others find noassociation.(1214) To some extent, this ambiguity isnot surprising, as interpretation of cross-sectional

designs is compromised by the inability to demon-strate sequential associations between risk percep-tion and behavior. Weinstein(15) shows that cross-sectional correlations can be interpreted as measuresof how accurately people perceive the riskiness ofcurrent behavior. Thus, unless a completely new be-havioral pattern is advocated, and no baseline forthat behavior exists, it is not possible to use a cross-sectional design to test the proposition that risk per-ceptions precede behavior.

Two observation designs use baseline risk per-ception to predict either changes of behavior or sub-sequent behavior, controlling baseline observations.These designs represent a standard approach in riskperception research because they allow sequentialinterpretations. However, two observation designstreat risk as a static entity, which fails to fully testthe prediction that changes in risk perception pre-cede behavioral changes. Weinstein(15) advocates athree-observation design, where risk perception ismeasured before and directly after any event thatmay change it. Behavior should be assessed beforethe event, contemporaneously with the first measureof risk perception, and a sufficient length of time af-ter the event to enable the postevent risk perception

to cause behavioral change. The timing of the secondbehavior measure allows the actor time to act uponthe risk perception change.

Particular importance lies in the timing of thepostevent risk perception measure. Provided thatactors perceive risk accurately, reciprocal relationsbetween risk perception and behavior means thatany change in risk perception that causes behavioralchange will, in turn, be modified by the change it cre-ates. The greater the time-lag between the event andthe second risk perception measure, the greater theprospect that this will occur. This will lead to an un-

derestimation of the relationship between the initialchange in risk perception and any change in behaviorit causes.

Weinsteins(15) recommendation is suited to well-controlled intervention contexts where the timingof observations and events can be conducted withprecision. Achieving this precision in large-scalesurveys of the driving population is obviously moredifficult. This problem may be circumvented by al-tering the way that risk perception is measured.

Weinstein and colleagues(16) operationalize risk per-ception in terms of actors perceptions of the like-lihood and severity of negative outcomes that arecontingent upon behavior (usually phrased as thecontinuity of current behavior). Using this definition,

risk perception is a property of the interaction be-tween risk inherent in a specific behavior and the fre-quency with which that behavior is engaged. Theserisk perception measures will be responsive to anybehavioral changes that they cause. Mills, Reyna,and Estrada(17) suggest that the responsiveness ofrisk perceptions to behavior can be reduced by mea-suring risk perceptions at a hypothetical level (e.g.,How risky is it to exceed speed limits by X?).These measures allow prediction of behavior,(9,16,17)

but do not incorporate elements of current behavior.Thus, they are less likely to respond to behavioralchanges.

Operationalizing speeding as the perceived dan-ger in exceeding speed limits (How often is itdangerous to drive at 70 km/h (43 mph) in a60 km/h (38 mph) zone), this report describes athree-observation prospective study of the tempo-ral relationship between changes in speeding-relatedrisk perception and subsequent changes in self-reported speeding. A baseline (T1) survey of thedriving population was conducted before a mass-media campaign emphasizing speeding-related crashrisk. A second survey was conducted after the cam-paign ended (T2), with a third conducted eight weeks

after T2 (T3). This provides the opportunity to linkany campaign-induced risk perceptions (T2) to laterself-reported speeding (T3). Evidence that risk per-ception influences self-reported speeding requiresthat T2 risk perception negatively predicts T3 self-reports of self-reported speeding behavior, control-ling T1 self-reported speeding and risk perceptionsand T2 self-reported speeding.

2. METHODS

The study was conducted in Adelaide, South

Australia, from early April to mid June 1998. A me-dia campaign, using the slogan Think about the Im-pact, was conducted on prime-time popular televi-sion, radio, and roadside billboards beginning a weekbefore the Easter weekend. Messages stressed thedanger inherent in small violations of the speed limit,the social unacceptability of speeding, and disputedcommon risk-mitigating excuses.(9) Consequences ofspeed-related crashes were portrayed, but highly dis-tressing crash or medical imagery was avoided. A

7/27/2019 j.1539-6924.2010.01406.x

3/7

7/27/2019 j.1539-6924.2010.01406.x

4/7

Risk Perception and Speeding 1095

Campaign recall provides a measure of attention andmessage processing, where greater specificity of re-call suggests greater attention and more elaboratemessage processing.(24) Three measures of media ex-posure were used at T2 only. First, participants were

asked whether they were aware of the existence ofa media campaign. Then all participants were askedto spontaneously recall message themes. A specificmessage score was compiled by the number of thefollowing campaign themes recalled: the campaignslogan, small violations of the speed limit being dan-gerous, the social unacceptability of speeding, andno excuses for speeding. This scale ranged 04, withhigher scores denoting more items recalled. A non-specific message score was compiled by drivers re-membering more general themes: speeding is danger-ous, speeding endangers others, and do not speed(range 03).

Specific and general media recall both showedsome skewness and kurtosis (skew = 1.24, kurtosis= 1.48 and skew = 1.90, kurtosis = 2.70, respec-tively). General media recall was transformed by tak-ing the square root (new mean = 0.27, SD = 0.53,skew = 1.5, kurtosis = 0.75). The transformed vari-able was used in inferential analyses. The skewnessand kurtosis on specific media recall proved resis-tant to transformation and the untransformed vari-able was used.

2.3. ProcedureInterviewers explained confidentiality and

anonymity rights to respondents. Interviewers firstrequested an unprompted response, and, where theresponse did not exactly fit the category system, theyprompted the respondent using the category system.A small trial of the questionnaire was conductedto ensure that participants understanding of itemmeanings were consistent with the researchers.

2.4. Analysis Plan

Hierarchical multiple regression was used to as-sess the extent to which T12 risk perception changespredict T23 self-reported speeding changes. T3 self-reported speeding was predicted from T2 risk per-ception, controlling age, gender, T1 and T2 self-reported speeding, and T1 risk perception. Statisticalcontrol of the T1 risk perception effectively yields aT12 change score associated with the T2 risk per-ception variable. In a similar way, controlling T1 andT2 self-reported speeding yields the equivalent of a

Table I. Means SDs and Frequencies of Study Variables(N = 255)

PossibleRange T1 T2 T3

Self-reportedspeeding 03 2.62 (0.96) 2.75 (0.83) 2.64 (0.87)

Riskperception

04 2.03 (0.83) 1.66 (0.85) 1.68 (0.83)

Higher scores indicate greater speeding and risk perceptions.

T23 change score. Thus, the T2 residual is used topredict the T3 self-reported speeding residual. Vari-ables were entered at four steps: first age, gender, andT1 self-reported speeding; second T1 risk perception;third T2 self-reported speeding; and fourth T2 riskperception. A significant change in R2 after the final

step would indicate that T12 changes in risk percep-tion predict T23 changes in self-reported speeding.

3. RESULTS

Means, SDs, and frequencies of untransformedvariables are shown in Table I. Repeated mea-sures ANOVAs were conducted to detect temporalchanges and, where significant, post hoc tests used toidentify the nature of those changes. T2 observationsof self-reported speeding were slightly higher than T1and T3 (F(1,153) = 3.61, 2(partial) = 0.03, p < 0.05).

Risk perception km fell from T1 to T2, with T3 obser-vations remaining at the T2 level (F(1,253) = 36.88,2(partial) = 0.21 p < 0.01).

Correlations between variables at the three timepoints are presented in Table II. As expected, riskperception and self-reported speeding were nega-tively correlated in cross-section at T1 and T2, andprospectively from T12, T13, and T23. Neitherspecific (mean = 0.50, SD = 0.66) nor generic (mean= 0.27, SD = 0.53) media recall, nor campaignawareness (82%, n = 209, awareness) were asso-ciated with self-reported speeding or risk percep-

tion. This suggests that the media campaign had littleeffect on study variables.The regression equation significantly predicted

T3 self-reported speeding (R2(adjusted) = 0.40,p < 0.01). Changes in R2 and their significance andstandardized betas for all steps are shown in TableIII. The significant 2.4% increase in explained vari-ance after the addition of T1 risk perception showsprospective prediction of self-reported speedingfrom a static risk perception variable. The significant

7/27/2019 j.1539-6924.2010.01406.x

5/7

1096 Brown

Table II. Correlations Between Study Variables (N = 255)

2 3 4 5 6 7 8

T11 Self-reported speeding 0.27 0.56 0.28 0.02 0.03 0.01 0.49

2 Risk perception 0.27

0.51

0.00 0.05 0.08 0.31

T23 Self-reported speeding 0.31 0.04 0.01 0.04 0.58

4 Risk perception 0.05 0.05 0.05 0.35

5 Media recall specific 0.46 0.35 0.046 Media recall generic 0.14 0.067 Campaign awareness 0.10

T38 Self-reported speeding

Table III. Standardized Betas and Change in R2 After theAddition of Each Variable in Hierarchical Regression Predicting

T3 Speeding

Step 1 Step 2 Step 3 Step 4

Age 0.20 0.19 0.14 0.14

Gender 0.06 0.04 0.01 0.01T1 speeding 0.45 0.40 0.21 0.20

T1 risk perception 0.17 0.12 0.06T2 speeding 0.40 0.39

T2 risk perception 0.13

R2 0.273 0.024 0.106 0.010

F 31.40 0.024 43.65 4.38

df 3,250 1,249 1,248 1,247

1.0% increase after the entry of T2 risk perceptiondemonstrates that T12 changes in risk perceptionpredict T23 changes in self-reported speeding.

From both theoretical and intervention view-points, it is important to understand whether par-ticipants who increased, decreased, or were stableon T12 risk perceptions made greater changes inT23 self-reported speeding. A change score wascalculated and partitioned into three dummy bino-mial variables representing: downward changes inT12 risk perception (n = 96) versus remaining staticor upward movement; stasis (138) versus upward or

downward; and upward (21) versus static or down-ward. The previous regression was repeated threetimes, using each binomial variable at the fourth stepinstead of T2 risk perception. This procedure waspreferred to a single analysis using dummy codingbecause the primary interest was in the extent towhich each dummy variable could improve predic-tion of T3 self-reported speeding, not the extent towhich each does so independently of the others. Inother words, the three-analysis procedure maximizes

individual standardized s because it eliminates vari-ance shared between dummy variables. The upwardmovement binomial significantly improved explainedvariance (R2 = 0.010, = 0.10, F(1,247) = 4.25,p < 0.05), predicting decreased T3 self-reportedspeeding. The other binomials did not improve ex-plained variance in T3 self-reported speeding (down-ward R2 = 0.006, = 0.01, p = 0.107; staticR2 = 0.00, = 0.00, F(1,247) = 2.60, p = 0.739).

4. DISCUSSION

Risk perception theories posit that risk percep-tion is a dynamic process, where changes in riskperception lead to subsequent behavioral changes.

There are few studies in any health or safety domainthat explicitly link changes in risk perceptions to fu-ture behavioral changes. Also, there is little prospec-tive evidence linking speeding-related risk percep-tions to lower self-reported speeding. The findingsof this study showed that changes in self-reportedspeeding between T23 were predicted by preced-ing T12 changes in the proportion of time thatdrivers considered it dangerous to be traveling at70 km/h in a 60 zone. Follow-up analyses suggestthat upward movement in T12 risk perceptions mayhave been the active component in predicting self-

reported speeding changes, but decreases had norelation to future self-reported speeding. Althoughan anti-speeding media campaign was conducted be-tween T12, perceptions of danger fell during thistime, and risk perception changes were not linked toindicators of campaign recall.

The use of a three-observation design allowsdemonstration of a dynamic risk perception process,where upward changes in perceived risk precede,and may cause, reductions in risk behavior. Although

7/27/2019 j.1539-6924.2010.01406.x

6/7

Risk Perception and Speeding 1097

assumptions about this process underlie all theoriesof risk perception, this is one of the few studies to usea three-observation design to demonstrate this. Thissupports the prediction that increases in risk percep-tion precede positive health behavior changes. One

downside of the findings is that changes in risk per-ception were not linked to the risk-perception-basedmedia campaign; thus it cannot be claimed that astructured intervention was responsible for risk per-ception changes.

These findings are in line with the outcomesof several cross-sectional studies into the nature ofthe relationship between risk perception and self-reported speeding.(811) The prospective design pro-vides stronger support for the application of riskperception theories to speeding. Although changesin risk perceptions were not linked to the mediacampaign in this study, risk-perception-based anti-speeding campaigns have been shown to changetarget risk perceptions at a group level.(11) Thus, al-though we did not show such a link, further workmight demonstrate that risk perceptions help to me-diate relationships between anti-speeding campaignsand speeding reductions.

It is interesting that risk perception scores de-clined from T12 and remained at the T2 level atT3. It is unlikely that the campaign caused this pat-tern, as media exposure variables were not relatedto risk perceptions or self-reported speeding. Thismay be an artifact, perhaps created by repeated mea-

surement or related transient environmental issues.The first interpretation appears to be unlikely, as self-reported speeding estimates did not decline in a sim-ilar way. Also, there did not appear to be differen-tial correlations between T1 and T2 risk perceptionsand other study variables as would be expected froman artifact of multiple measurement. The cause ofthis reduction is unknown. Transient environmentalfactors, such as the weather, may affect risk percep-tions regarding speeding. None were noted duringthe study, although it was conducted during autumn,where seasonal patterns may lead drivers to expect

greater rainfall, making speeding more dangerous.Although we have argued that self-reportedspeeding is closely associated with observed speed-ing, the study must still be interpreted in the lightof commonly understood limitations associated withbehavioral self-reports. Also, risk was measured asthe proportion of the time the respondent estimatesthat exceeding at 60 km/h speed limit is consideredto be dangerous. This measure conceptualizes risk interms of a drivers hypothetical, rather than actual,

engagement in speeding. The current study wouldhave benefited from comparison between this andmore commonly used measures of outcome likeli-hood, the potential severity of those outcomes, andemotional representations of risk. Also, the design

deliberately covers a reasonably short period of time,and it may be unwise to draw implications pertain-ing to the permanence of any effects observed in thisstudy. Finally, the design of this study allows conclu-sions to be drawn about sequential relations betweenrisk perceptions and self-reported speeding. How-ever, sequence should not be confused with cause,and the study does not provide conclusive evidencethat risk perceptions cause behavioral changes.(15)

This study attempted a reasonably strict test ofrisk perception theory, finding prospective links be-tween increases in risk perception and self-reportedspeeding reductions. This provides stronger evidencelinking risk perceptions to self-reported speedingthan previous cross-sectional studies. Links were notfound between risk perception changes and the anti-speeding media campaign, and further research mayhelp to tie changes in risk perceptions more firmly tostructured interventions.

ACKNOWLEDGMENTS

The assistance of Transport SA (known as theSouth Australian Department of Transport duringthe conduct of the research), and particularly Mr.John Walker, in the collection of the data containedin this report is gratefully acknowledged.

REFERENCES

1. Weinstein ND, Nicolich M. Correct and incorrect interpreta-tions of correlations between risk perceptions and risk behav-iors. Health Psychology, 1993; 12:235245.

2. Brewer NT, Chapman GB, Gibbons FX, Gerrard M, McCaulKD, Weinstein ND. Meta-analysis of the relationship betweenrisk perceptions and health behavior: The example of vaccina-tion. Health Psychology, 2007; 26:136145.

3. Harrison JA, Mullen PD, Green LW. A meta-analysis of stud-ies of the health belief model in adults. Health Education Re-search, 1992; 7:107116.

4. Van Der Pligt J. Perceived risk and vulnerability as predictors

of precautionary behaviour. British Journal of HealthPsychol-ogy, 1998; 3:114.5. National Highway Traffic Safety Administration. 2007 Traf-

fic Safety Annual AssessmentHighlights. Washington, DC:NHTSA, DOT HS 811 017, 2008.

6. World Health Organization. World Report on Road TrafficInjury Prevention. Geneva: WHO, 2004.

7. Kloeden CN, MacLean AJ, Moore VM, Ponte G. TravellingSpeed and the Risk of Crash Involvement. Canberra, Aus-tralia: Federal Office of Road Safety, CR172, 1997.

8. Adams-Guppy JR, Guppy A. Speeding in relation to percep-tions of risk, utility and driving style by British company cardrivers. Ergonomics, 1995; 38:25252535.

7/27/2019 j.1539-6924.2010.01406.x

7/7

1098 Brown

9. Brown SL, Cotton A. Risk-mitigating beliefs, risk estimatesand self-reported speeding in a sample of Australian drivers.Journal of Safety Research, 2003; 34:16.

10. Fildes BN, Rumbold G, Leening A. Speed Behaviourand Drivers Attitudes to Speeding. Melbourne, Australia:Monash University Accident Research Centre, Report No. 16,1991, 1991.

11. Rundmo T, Iversen H. Risk perceptions and driving behav-ior among adolescents in two Norwegian counties before andafter a traffic safety campaign. Safety Science, 2004; 42:121.

12. Machin MA, Sankey KS. Relationships between youngdrivers personality characteristics, risk perceptions and driv-ing behavior. Accident Analysis and Prevention, 2008; 40:541547.

13. McKenna FP, Horswill MS. Risk taking from the participantsperspective: The case of driving and accident risk. Health Psy-chology, 2006; 25:163170.

14. Ulleberg P, Rundmo T. Personality, attitudes and risk per-ception as predictors of risky driving behavior among youngdrivers. Safety Science, 2003; 41:427443.

15. Weinstein ND. Misleading tests of health behavior theories.Annals of Behavioral Medicine, 2007; 33:110.

16. Weinstein ND, Kwitel A, McCaul KD, Magnan RE, Ger-

rard M, Gibbons FX. Risk perceptions: Assessment and re-lationship to influenza vaccination. Health Psychology, 2007;26:146151.

17. Mills B, Reyna VF, Estrada S. Explaining contradictory rela-tions between risk perception and risk taking. PsychologicalScience, 2008; 19:429433.

18. Aberg L, Larsen L, Glad A, Beilinsson L. Observed ve-hicle speed and drivers perceived speed of others. Ap-plied Psychology: An International Review, 1997; 46:287302.

19. West RJ, French D, Kemp R, Elander J. Direct observation ofdriving, self-reports of driver behavior and accident involve-ment. Ergonomics, 1993; 36:557567.

20. Slovic P, Peters E. Risk perception and affect. Current Direc-tions in Psychological Science, 2006; 15:322325.

21. Underwood J, Dillon J, Farnsworth B, Twiner, A. Readingthe road: The influence of age and sex on child pedestriansperceptions of road risk. British Journal of Psychology, 2007;98:93110.

22. Rosenbloom T, Nemrodov D, Ader B-E, Ehud E. Fear anddanger appraisals of a road-crossing scenario: A develop-mental perspective. Accident Analysis and Prevention, 2008;40:16191626.

23. Groeger JA, Chapman PR. Judgement of traffic scenes: Therole of danger and difficulty. Applied Cognitive Psychology,1996; 10:349364.

24. Mehta A, Purvis SC. Reconsidering recall and emotion inadvertising. Journal of Advertising Research, 2006; 46:4956.