Download - Deterrence theory: Certainty of PUNISHMENT in an experimental context · Deterrence theory postulates that certainty, severity and celerity of punishment reduce deviant behavior.

Transcript

Deterrence theory: Certainty of PUNISHMENT in an

experimental context

R eb ecca L O B M A N N , H an s-P eter K R U G E R

C en ter fo r traffic safety , P sycho log isches Institu t der U niversita t W urzburg ,

R oentgering 11, D -97070 W urzburg , G erm any

A B S T R A C T

D eterrence theory postu lates that certain ty , severity and celerity o f pun ishm ent reduce dev ian t

behavior. W ith d runk driv ing certain ty o f detection is in fact a tw o-step-process: first, the

d river has to be stopped, second, the police has to notice his drunkness. T herefore , in

discussing certain ty o f detection as a counterm easure against driv ing under the in fluence o f

alcohol (D U I), frequency and effic iency o f con tro lling bo th have to be taken in to account. In

an experim enta l con tex t the relationship betw een the tw o factors w as analysed. S ubjects

(n=20) p layed a com pete tive gam e, w here frequency and efficiency o f con tro lling cou ld be

varied. C onditions o f 20% and 80% probability fo r each facto r resulted in 4 treatm ent groups

to w hich subjects w ere random ly assigned. A fter the gam e subjects estim ated p robab ility o f

detection and ra ted the ir com petence in the gam e. Subjects overestim ated low probab ilites o f

detection . A lso, estim ation o f subjective risk w as h igher w hen con tro ls w ere m ore efficient.

IN T R O D U C T IO N

O ur in terv iew s w ith d rink ing and non-drink ing drivers (n=2368) (see K ruger and L obm ann ,

th is volum e) have revealed som e in teresting relationsh ips betw een factors o f deterrence and

DU I: K now ledge about the law and severeness o f punishm ent do not p red ic t d runk d riv ing ,

bu t subjective p robab ility o f detection does have an effect if exposure to traffic is in troduced

as a m oderato r variab le (O nly m uch driv ing drivers have undergone enough police con tro ls to

base their estim ation on experience). W hen subjects w ere asked »Is a d river exceed ing the

legal B A C lim it detected w hen stopped by the police?» M uch driv ing drink ing drivers

estim ated effic iency o f police con tro ls substantially low er than m uch driv ing sober contro ls.

- 94 9 -

T he drink ing d river m ay thus draw the fo llow ing conclusion from his experience w ith police

contro ls: »1 d id drink, the police does stop people, bu t they d id not notice m y drunkness. I

w ill go on driv ing drunk.» T he sober d river on the o ther hand w ill p robably experience the

re lie f o f negative reinforcem ent: » If I had had alcohol they w ould have detected m e. I w ill go

on driv ing sober.» T he low efficiency o f police contro ls w as dem onstrated by ano ther study

at our institu te (K azenw adel & V ollrath, 1995). C om paring norm al detection rates w ith the

proportion o f drink ing drivers found if every partic ipant o f an accident w as b rea th tested

K azenw adel and V ollrath calculated a detection probability o f only 54.8% for B A C s o f

0 .08% and above.

E vidently , the tw o-step process o f detection and its perception by the d river need m ore

attention. T herefore, an experim ental context w as designed in o rder to analyse basic law s o f

hum an percep tion in th is field. T w o questions had to be answ ered:

• H ow is the ob jective risk o f being detected represented in subjective percep tion?

• W hich in fluence does the efficiency o f contro ls have on the estim ation o f p robab ility o f

detection?

M E T H O D

P ro c e d u re

T he experim ental parad igm is a gam e in w hich a sm uggler and a custom s officer play against

each other. T he sm uggler has a red and a yellow chip, each chip is p layed 50 tim es. H e has to

choose betw een the chips and play them one at a tim e by taking one ch ip in one hand the

o ther hand staying em pty. T hen the custom s officer has to decide i f he w ants «to see» or if he

ju st says «go on» (frequency o f controls). I f he decides to contro l, he has to choose betw een

the tw o hands o f the sm uggler (efficiency o f contro ls). T he sub ject has the ro le o f the

sm uggler, the experim enter the ro le o f the custom s officer. U nnoticeable fo r the subject the

experim enter is in fo rm ed by a technical device, in w hich hand w hat k ind o f chip is hidden.

T herefore, he can vary frequency and efficiency o f contro ls independently from one another.

F o r a cover-story story subjects are to ld that their m im ical expression w hile b lu ffing w ould be

studied and their task is to hide the red chip from the experim enter. T o m ake subjects b e lie f

in th is story they are v ideotaped. F or getting a red chip through (but not fo r a yellow chip)

- 9 5 0 -

subjects are rew arded w ith a sw eetie. A fter the gam e subjects ra te probab ility o f detection ,

their ow n com petence and the com petence o f the experim enter.

F requency and effic iency o f con tro ls w ere varied w ith levels o f 20% and 80% probab ility

(sam e probab ilites fo r red and yellow chips). T able 1 show s that detection p robab ilities o f

4% , 16% and 64% em erge. T w o cells have the sam e probability o f detection o f 16% , tha t is

the sam e num ber o f chips is found in both groups. But in the first g roup a con tro l takes place

in only 20% o f the cases, bu t o f these 80% are detected (high effic iency) w hereas in the o ther

g roup m any con tro ls are experienced (80% ) bu t only 20% o f these are successfu l (low

efficiency). I f effic iency o f contro ls has an effect ratings in these tw o groups should differ.

T able 1: F requency and effic iency o f controls in exp erim ental groups. C ells depict

p robab ilities o f detection .

frequency o f controls effic iency o f co

20%

ntrols

80%

20% 4% 16%

80% 16% 64%

Subjects

5 subjects w ere random ly assigned to each cell (n=20). They w ere students from in troduc tory

courses o f psychology w ho earned credits fo r their participation . M edian age w as 21 years,

55% o f the subjects w ere w om en.

R E SU L T S

P recision o f estim ation

T o com pare sub jective ratings w ith actual probab ilities o f detection ob jective and subjective

p robab ilities are p lo tted against each o ther (F igure 1). T he d iagonal show s w here the es ti­

m ations shou ld be if subjective ratings correspond exactly to the ob jective probabilities. Low

probab ilities o f detection are overestim ated w hereas the estim ation o f h igher p robab ilities is

quite exact. T -tests y ield significant d ifferences at ob jective p robab ilities o f 4% and 16%

(p< .005; p< .010).

-951 -

Figure 1: O verestim ation o f objective p robab ility o f detection . V alu es o f groups w ith the

sam e ob jective prob ab ility o f d etection but d ifferent frequency and effic ien cy o f control

are averaged . R esu lts o f earlier experim ents are included (objective p rob ab ilities o f

25% and 40% ).

S ubjective p robab ilities (mean ratings in percent)

objective probab ility in percent

E fficien cy o f controls

T o evaluate the ro le o f effic iency o f contro ls ratings o f tw o groups w ith the sam e ob jective

probab ility o f detec tion (16% ) but d iffering effic iency o f con tro ls are com pared (F igure 2).

F igure 2: S u b jective probab ilities dep en d ing on efficiency o f controls. O b jective

probab ilities 16% .

"How likely is it to be detected when p laying the red ch ip ?" (mean ratings in percent)

4 5 .0

4 0 .0

3 5 .0

3 0 .0

con tro l e ffic ien t contro l no t e ffic ien t

A lthough both groups have the sam e ob jective probab ility o f 16% ratings o f sub jective

probab ility differ. In the g roup w ith rare bu t effic ien t con tro ls (left) probab ility o f detec tion is

- 9 5 2 -

estim ated h igher than in the g roup w ith frequen t bu t not effic ien t con tro ls (right). A t-test fo r

the d ifference betw een the tw o groups is sign ifican t on the 10% -level ( t= -2 .10; p= .097).

T o illustrate the resu lts o f sub jec ts’ ra ting o f com petence the tw o g roups’ m ean values o f the

fo llow ing item s are disp layed: agreem ent w ith the statem ent «I show a good pokerface.»

(F igure 3) and «H ow good is the experim en ter in th is gam e?» (F igure 4).

T here are d ifferences betw een the groups depending on efficiency o f contro l desp ite o f an

equal ob jective probab ility o f detection . W hen con tro ls are rare bu t effic ien t subjects th ink

they have a w orse pokerface and consider the experim enter to be better in the gam e. A nsw ers

to o ther questions reveal th a t they are also m ore excited and th ink they are w orse p layers if

e ffic iency is high.

F igu re 3: R atings o f ow n com petence

Agreement with "I show a good pokerface." (mean ratings on scale 1-5)

4.0

3.5

3.0

2.5

2.0control efficient control not efficient

F igure 4: R atings o f exp er im en ter’s

com petence dep en d ing on effic ien cy o f

con tro ls.d ep en d in g on effic ien cy o f

controls.

"How good is the experimenter in this game?"(mean ratings on scale 1-5)

control efficient control not efficient

D ISC U SSIO N

R esults lead to the conclusion tha t peop le take ob jective probab ility o f de tec tion in to account

but w eigh the ir estim ation by effic iency o f contro ls. F urther research m ust therefore a im at

find ing the psychophysical function betw een subjective probab ility o f detection on the one

hand and ob jec tive p robab ility and effic iency on the o ther hand. T h is algorithm w ill probab ly

be a m u ltip lica tion o f the ob jec tive facto rs w eigh ted by their respec tive costs.

- 95 3 -

Probability o f detection is best elevated by m ore effic ien t contro lling . In the described

experim ent the d ifference in subjective probability betw een groups o f h igh and low effic iency

w as 7.6% (figure 2). T he non-effic ien t group had a con tro lling frequency o f 80% w hich

m eant 40 chips out o f 50 are checked. To achieve the sam e subjective probab ility o f detection

as in the h igh efficiency group 59 contro ls w ould have to be conducted . T hat w ould be an

increase o f 50% in con tro lling frequency.

C O N C L U SIO N

In rating p robab ilities o f detection subjects take ob jective p robab ilities in to account but

overestim ate low probabilities. A high efficiency o f con tro ls con tribu tes substantially to the

risk o f detection . O bviously , the relation o f detections to con tro ls (effic iency) is im portan t in

the con tex t o f police enforcem ent.

R E F E R E N C E S

K azenw adel, J. & V ollrath , M. (1995). D as D unkelfeld der T runkenheitsfahrten . In H .-P .

K ruger (H rsg .). Das Unfallrisiko unter Alkohol. S tu ttgart: G ustav F ischer V erlag.

- 9 5 4 -