modifying negligent driving behavior through warning letters

28
Accid. Anal. & Prev. Vol. 2, PP. 147-174. Pergamon Press 1970. Printed in Great Britain. MODIFYING NEGLIGENT DRIVING BEHAVIOR THROUGH WARNING LETTERS* ROBIN S. MCBRIDE and RAYMOND C. PECK Department of Motor Vehicles, Sacramento, California (Received 10 September 1969) THE POST-LICENSING CONTROL of traffic violators is a major problem throughout the United States. In California alone, over 100,000 “negligent” drivers are subjected to some form of driver improvement action each year. These actions become progressively more stringent as negligent driving continues: a letter, warning the driver of the consequence of repeated negli- gent driving practices; a group meeting, designed to aid the driver in improving his record; and finally, an individual hearing, where some form of restrictive action is usually taken by the Department. All of these efforts aim to reduce the number of traffic accidents and citations, with an ultimate objective of reducing traffic fatalities. The extent to which group and individual meetings accomplish the goal of reducing accidents is questionable. Several studies conducted by the California Department of Motor Vehicles (DMV)? have shown that the meetings do reduce traffic convictions but are of little value in reducing accidents (Coppin, 1961; Coppin, Marsh and Peck, 1965; Coppin, Peck, Lew and Marsh, 1965). Numerous studies by other investigators have produced equivocal findings or negative results similar to the California findings (Kaestner, 1968; Toms, Kastelle and LeSuer, 1966; Chalfant, 1960). There have been few attempts to evaluate the effectiveness of warning letters in reducing accidents and violations. The Texas Department of Motor Vehicles compared a group of drivers who received an advisory letter with a control group (Temple and Ferguson, 1958). Little difference was found between the groups on follow-up driving record. In another study, Kaestner et al. (1967) compared a “standard” warning letter used by the Oregon Department of Motor Vehicles with a personalized letter and control group (no letter). There was no significant difference between the standard and control group, but the personal- ized approach resulted in drivers under 25 having fewer entries on their record in a follow- up period. With the exception of the Kaestner study, no attempt has been made to evaluate the effectiveness of the specific content of the letter in modifying driver behavior. Such letters are typically bureaucratic in nature and their utility pretty much taken for granted by licensing administrators (Campbell, 1958). The advantages of an effective warning letter program are obvious when it is recognized *The authors of this paper have been awarded the Metropolitan Life Award for Research in Accident Prevention for the year 1970 on the basis of the research presented here. tsubsequent to the preparation of this manuscript, a group educational approach was found to be effective in reducing accidents (Marsh, 1969). 147 A.A.P. 2/3--e

Upload: independent

Post on 15-May-2023

0 views

Category:

Documents


0 download

TRANSCRIPT

Accid. Anal. & Prev. Vol. 2, PP. 147-174. Pergamon Press 1970. Printed in Great Britain.

MODIFYING NEGLIGENT DRIVING BEHAVIOR

THROUGH WARNING LETTERS*

ROBIN S. MCBRIDE and RAYMOND C. PECK

Department of Motor Vehicles, Sacramento, California

(Received 10 September 1969)

THE POST-LICENSING CONTROL of traffic violators is a major problem throughout the United States. In California alone, over 100,000 “negligent” drivers are subjected to some form of driver improvement action each year. These actions become progressively more stringent as negligent driving continues: a letter, warning the driver of the consequence of repeated negli- gent driving practices; a group meeting, designed to aid the driver in improving his record; and finally, an individual hearing, where some form of restrictive action is usually taken by the Department. All of these efforts aim to reduce the number of traffic accidents and citations, with an ultimate objective of reducing traffic fatalities.

The extent to which group and individual meetings accomplish the goal of reducing accidents is questionable. Several studies conducted by the California Department of Motor Vehicles (DMV)? have shown that the meetings do reduce traffic convictions but are of little value in reducing accidents (Coppin, 1961; Coppin, Marsh and Peck, 1965; Coppin, Peck, Lew and Marsh, 1965). Numerous studies by other investigators have produced equivocal findings or negative results similar to the California findings (Kaestner, 1968; Toms, Kastelle and LeSuer, 1966; Chalfant, 1960).

There have been few attempts to evaluate the effectiveness of warning letters in reducing accidents and violations. The Texas Department of Motor Vehicles compared a group of drivers who received an advisory letter with a control group (Temple and Ferguson, 1958). Little difference was found between the groups on follow-up driving record. In another study, Kaestner et al. (1967) compared a “standard” warning letter used by the Oregon Department of Motor Vehicles with a personalized letter and control group (no letter). There was no significant difference between the standard and control group, but the personal- ized approach resulted in drivers under 25 having fewer entries on their record in a follow- up period.

With the exception of the Kaestner study, no attempt has been made to evaluate the effectiveness of the specific content of the letter in modifying driver behavior. Such letters are typically bureaucratic in nature and their utility pretty much taken for granted by licensing administrators (Campbell, 1958).

The advantages of an effective warning letter program are obvious when it is recognized

*The authors of this paper have been awarded the Metropolitan Life Award for Research in Accident Prevention for the year 1970 on the basis of the research presented here.

tsubsequent to the preparation of this manuscript, a group educational approach was found to be effective in reducing accidents (Marsh, 1969).

147

A.A.P. 2/3--e

14s ROBIX S. MCBRIDE AND RAYMOND C. PECK

that: (1) a letter contact is much less expensive than a meeting, (2) a large population of drivers can be contacted, (3) a letter may be as effective as a meeting, and (4) a letter can reach drivers before negligent driving patterns develop to the extent departmental meetings are necessary.

The present study was structured around several key questions: (1) Does the degree of threat and personalized content (Intimacy) in a warning letter have

any effect on subsequent driving behavior? (2) Are experimentally developed letters more effective than the current “standard”

letter used by the Department or a control condition (no letter)? (3) Is a warning letter accompanied by a questionnaire allowing the driver to “sound off”

a more effective contact than issuing only a letter? (4) Does a follow-up letter acknowledging improved driving record reinforce continued

improvement? (5) Are certain types of letters or letter combinations more effective for various age, sex

and marital status groupings? (6) Can a warning letter program be justitied on a cost/benefit basis? In designing the study, emphasis was placed on exploring Threat and Intimacy dimensions

of warning letters as well as evaluating the effectiveness of California’s present letter program. It was thought that the development and use of dimensional factors would favor increased generality and the theoretical significance of findings.

Subjects METHOD

From November 1966 through January 1967, 18,000 negligent drivers throughout California were selected from the central driver record files in Sacramento. These drivers became eligible for departmenta contact by meeting or approaching the lega negligent operator definition of four points in 12 months *. Thus, each subject must have had a point count of at least three and not more than four within the 12 months prior to warning letter issuance.

In addition, each subject had to meet the following selection conditions: I. Aged 16-60 years. 2. Possess a valid California drivers license. 3. Had no prior driver improvement contact with the Department of Motor Vehicles

or evidence of existing departmental action against his license. The subjects were randomly assigned to the letter treatment groups by the fourth and

fifth digits of the eight digit drivers license number. This strategy varied at times to equalize the treatment cells.

Development of the ex~erim~ntaI warning letters An extensive review of the literature was initially made in the areas of mass communica-

tion techniques, marketing, advertising and behavior modification principles (McBride, 1967). Based on the literature review and nature of the warning letter program, two letter dimensions were tentatively selected for manipulation: intensity of Threat and Intimacy of style.

*In California, the point count is based on the number of moving violations, convictions and responsible accidents on a driver’s record. Most moving traffic violations and responsible accidents count one point each, except for more serious violations (drunk d:iving etc.) which count two points.

Modifying negligent driving behavior through warning letters 149

Techniques known or thought to be effective in changing attitudes such as repetition of an

idea, rewarding acceptable behavior, reprimandin g unacceptable behavior, ordering of

material etc., were incorporated in the construction of the warning letters. The subject content (ideas) and order of presentation were held constant throughout the different inten- sities of a dimension, since the message to be conveyed was largely constrained by ad- ministrative policy. Also, the separate effects of ideas and mode of intensity could not be separated without resorting to a forbiddingly complex research design. Therefore, only the manner in stating an idea was manipulated to obtain different intensities of stimulus dimensions. Although operational definitions of the stimulus dimensions are somewhat limited, the foregoing provided a workin, 0 framework for construction of the two dimen-

sions :

Threat. Threat was manifested by the number of words and sentences which have been shown by independent studies to have varying degrees of threat stimulus (Semantic Differential Scale). When empirical evidence was lacking on action words or phrases, subjective judgements were made as to the threat intensity. Intimacy. Intimacy content was varied through the use of personal pronouns in references to the subject and department, and type of format used in each letter. The letters were varied in accordance with the Flesch Readability and Interest Scales (Flesch, 1948). These scales measure communication in terms of reading ease, personal interaction between the communicator and communicatee, and the interest value of the communica- tion. In structuring the letters on this dimension, emphasis was directed toward conveying the idea that a driver improvement analyst was personally interested in the individual’s improvement to the other extreme of bureaucratic formality.

A content analysis* was conducted by the author to assure that the basic ideas were present in all letters. The letters were developed to represent three levels of each dimension, so that there were nine possible combinations of the levels:

Low Threat/Low Intimacy Medium Threat/Low Intimacy

Low Threat/Medium Intimacy Medium Threat/Medium Intimacy

Low Threat/High Intimacy Medium Threat/High Intimacy High Threat/Low Intimacy High Threat/Medium Intimacy High Threat/High Intimacy

After the nine letters were structured on two stimulus dimensions (Threat and Intimacy) of three intensity levels, a further validation study was made to determine empirically if the letters would be perceived as varying on these postulated dimensions by negligent operators. A group of 108 negligent drivers who attended a driver improvement meeting (DIM) were requested to evaluate the letters on the Threat and Intimacy dimensions. The nine letters were randomly assigned to the subjects. Each subject rated one letter on the degree of Threat and Intimacy on a twenty item Likert-scaled questionnaire. The rating items were developed on an a priori basis by the authors and were discussed in an earlier study (McBride, 1967). The subject’s ratings of the letters were in accordance with the intended dimensionality and the relationships were significant on both factors, That is, the

*Due to the length and nature of this analysis, this phase of the study has been omitted from the study. The material is available at the California DMV. An outline is presented in Exhibit A.

150 ROBIN S. MCBRIDE AWD RAYMOND C. PECK

higher the intended intimacy or threat, the higher the ratings on these factors. There was no evidence of interaction between the dimensions in the ratings of the validation sample.

EXHIBIT A

Warning Letter Content Analysis

All letters contain the following basic subject content: 1. 2. 3. 4.

Statement of driving record. _

5. 6. 7.

Statement of negligent driver classification. Obligation of department or analyst to insure that traffic laws are obeyed. Consequence of negligent driver dehavior: (a) threat of action by DMV (b) inconvenience (time, money, job) (c) loss of life, or limb. Statistics or figures show greater likelihood of serious consequences if poor driving continues. Attitude of department toward negligent driver. Periodic review. (a) progress (b) future action

8. Need for immediate improvement.

Study design The basic design of the present study involved an evaluation of the effectiveness of four

orthogonal factors: (a) Threat content, (b) Intimacy content, (c) Questionnaire, (d) Rein- forcement. The design can be best characterized as a 2J factorial.

A decision was made to include only four of the nine experimental letters in the study due to the complexity of administering all letters. The final four experimental letters rep- resented the combination of the bi-polar extremes for the threat and intimacy dimensions as follows: (a) high threat/high intimacy (ffT/Nf); (b) high threat/low intimacy (HT/LZ); (c> low threat~high intimacy (LT/Hf); (d) low threat/low intimacy (LT/LZ). The “standard” (S7’D) currently used by the Department was also included in the study (Exhibits B-F).

Questionnaire. Throughout the assignment period, a questionnaire was included with each letter on an every-other-day basis. Thus, one-half of the subjects in each letter group received a questionnaire. The purpose of the questionnaire was to provide the subject with an opportunity to comment on enforcement policies and to state his reaction to receiving the letter. It was hypothesized that a letter accompanied by a questionnaire might result in a more effective treatment by forcing the subject to reflect on his driving behavior and become an active participant in the treatment process. In addition to this potential treatment effect, the questionnaire was structured to provide normative data on mileage, occupation, education and attitudes toward driving (Exhibit C).

Reinforcement. Seven months subsequent to the initial warning letter contacts, all sub- jects who had no accidents or violations were identified. One-half of the subjects in each treatment cell were randomly assigned to the reinforcement letter group, who were later issued a letter acknowledging their improvement (Exhibit I-I).

Control group. Subjects who were assigned to the control group* were sent a postage-paid return card requesting verification of address. The card stated that DMV was automating the vehicle registration files and wanted to update the address file. The purpose of this

*To measure impact on non-recipient rate, one-half of the control subjects received their address change in an envelope without DhW identified. There was a significantly lower non-recipien! rate for the group with DMV on the envelope (3.4 vs. 4.8 per cent); the groups did not differ on subsequent driver record and were combined for all analyses.

Modifying negligent driving behavior through warning lettsrr 151

r

l-

High Threat/High Personal

1

LICENSE 30.

-I

BECAUSE OF YOUR DRIVING DURING THE PAST YEAR, YOU ARE IN DANGER OF HAVING YOUR DRIVERS LICENSE TAKEN AWAY!

Your case has been referred to me as part of my special driver improvement case load.

Your record shows that on several occasions during the past year, you have been convicted of hazardous violations of the traffic laws. Your record places you dangerously close to being classified as a negligent driver and I must study it and think about the possibility of withdrawing your license to drive.

At this moment, I am looking at fatal accident reports which clearly show that irresponsible driving patterns like yours cannot only cause financial disaster, but can cause you to be maimed, disfigured or even killed!

I am sure you know that the Department and I have a responsibility to keen reckless drivers off our hinhwavs. You must realize that your dangerous driving habits canno: ---and will not -- be allowed to continue. I know from years of experience that reckless drivers can improve if they try. Unless improvement occurs in your case, you will leave us with no choice other than to restrict or even withdraw your driving privilege.

Although I am recommending that no action be taken at this time, your record will be checked periodically to determine if we will have to withdraw your license. Remember, the action that will be taken depends on you:

DRIVER RECORD:

Violations- Accidents- Sincerely yours,

3&-%=4&

Driver Inpr PG ement Analyst

EXHIBIT B

152 Roan S. MCBRIDE AAD RAYMOND C. PECK

High Threat/Low Personal

LICENSE NO. l- 1 VIOLATIONS

ACCIDENTS

I- -l

YOU ARE IN DANGER OF RAVING YOUR DRIVING PRIVILEGE UITHDRAWN!

During the past year you have on several occasions been convicted of hazardous violations of the traffic laws. This places you dangerously close to being categorized a8 a negligent operator in accordance with Section 12810 of the Vehicle Code. This section empowers -- and in fact obligates -- the Department of Motor Vehicles to utilize its discretionary authority in taking hazardous drivers off the streets and highways.

Statistics clearly indicate that irresponsible driving patterns such as yours often result in the maiming of innocent people and in destruction of human life. Thus we cannot -- and will not -- tolerate negligent and hazardous driving on the streets and highways of this state. Violation of traffic laws must cease or the Department will be forced to take harsh measures against your driving privilege. The revocation or even restriction of one's license can result in severe personal and economic disaster.

Your record will henceforth be placed in an action pending file and reviewed by the Department on a periodic basis to determine if restrictive measures will be necessary.

It is never too late to improve, but in your case, improvement must be immediate if restrictive action is to be avoided.

Department of Motor Vehicles Division of Drivers Licenses

BY R. V. Thunen

EXHIBI-C C

Modifying negligent driving behavior through warning letters 153

Low Threat/Low Personal

r -I LICENSE NO.

VIOLATIONS

L -I ACCIDENTS

The Department of Motor Vehicles is reviewing the records of all drivers in danger of being classified as negligent operators in accordance with Section 12810 of the California Vehicle Code. This code permits the Department to utilize its discretionary authority in the interest of public safety.

During the past year you were convicted of several violations of the traffic laws which place you in danger of being legally classi- fied as a negligent operator.

Statistics show that drivers who violate traffic laws frequently represent increased safety risks to themselves and to the public. In addition, continued traffic violations and/or accident involve- ment may result in economic inconvenience to yourself and to others.

The Department does not want to take restrictive measures and is con- fident that you will cease driving in an unsafe manner. Henceforth, your case will be reviewed on a periodic basis and any further action will depend upon your future driving performance.

The Department wishes to emphasize that the time for improvement is now -- not later.

Department of Motor Vehicles Division of Drivers Licenses

BY R. V. Thunen

EXHIEIIT D

154 ROBIN SMCBRILW AND RAYMOND C. PECK

I- 1

LICEKSE NO.

I- -J

Because of your driving record during the past 12 months, your case has beer, given to me as part of my special driver improvement case load. b>/ review of your record indicates that during the past year you have been convicted on several occasions of violations of the traffic laws. Since you are very close to the legal definition of a negligent driver, I muat make a recommendation to the Department as to what we should do about you.

I know from years of experience that some drivers with records like yours are a danger to themselves, their loved ones! and their cormnun- ity. In addition, there is always the danger of financial loss.

I am sure you must realize the important responsibility which the Department and I have to keep unsafe drivers off our highways. I am counting on you to show me that you can drive in a safe, responsible manner. I will personally check you record for improvement to determlne if further action is necessary.

It has been my experience that drivers like you can improve if they really try. Remember, now is the time for you to improve your driving -- not later.

DRIVZR ZXORD:

Violations - Accidents -

EXHIBIT E

&Modifying negligent driving behavior through warning letters

WARNING

LlCEsrE x.3 ,...__..___.. . ..-_..--..

DtTPARTXllA’T Or: .\lOTOK \‘I3lICLI3

155

156 ROBIN S. MCBRIDE AND RAYWOND C. PECK

DRIVER INFOWATION REQUZSi License No.

The Department of Uo(otor Vehicles is very interested in your attitude toward drivin3 and your reaction to the enclosed warning letter. Please return thin questionnsire vi:h e.ll the items cozpleced. The information vi11 be used to etudy and evaluate the Department’s Driver ~tove~nc program. All infonacion vi11 be held in strict confidence and vi11 not have any influecca 03 your driving pclvflege.

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

L’hac kind of vork do yau do? (occupetion) Industry

HFY many jobs have you had in the past 12 moorha?

Are you no”: m single 0 married u divorced 0 separated a vidoved

Number of dependents (other than self):

Circle the highest grade attended: 1 2 3 4 5 6 7 8 9 10 11 12 11 14 l> 16 17+ nrade school hinh school COlleRe

L

How many miles have you driven in the last 12 fmothr?

Check the block or blocks vhich you feel apply in ycur case:

,

n 0

E7

u 17

0

F!y driving has been highly negligent: however. I realize the need for safe driving habits.

I drive 8s carefully a, other drivers but I wee at the wrong place ac the wrong time.

NY driving has not been careless or negligent at all; hovever, I realize that more effort is needed on my part .to avoid further violations and/or accidenrs.

I got d =W deal from the police and did not deserve most of the cickecs I received.

Hy driving has been slightly negligent; hovever, I realize the need for safe driving habits.

I do nor feel that the traffic laos I violete increaee my chances of being involved in an accident.

Afrer each word, check the poslrion which best describes your feelings af:er you read the warning 1errer7-for example. in part %l’, you can show how disgueted You vere by cbesking one of the spaces to the right of the word “Disgusted.” Be sure to make a check afrer each word.

Very Noderacely Slighrly NOc at all a. 3isgusted - - - - b. llpser - - - - c. Scared - - - - d. Indifferent - - - - e. AWKY - - - - F. Embarrassed - - - - g. Grateful - - - - h. Amused - - - -

I feel the Warning letter is:

a Justified in my case. L%Y?

a Not justified in my c&se. uhy?

Additional c~mnents (if ‘ny)

EXHIBIT G

Modifying negligent driving behavior through warning letters 157

DRIVER RECORD PROGRESS RZPORT

(ReinforcenenL)

License No.

As of this date the Department's records indicate that you have

not been involved in traffic citations or accidents since receiv-

ing our warning letter several months ago. Therefore, it is my

pleasure to acknowledge this improvement in your driving record.

It is indeed unfortunate that more drivers do not show a similar

improvement. The Department will continue to review your record

over the next six months for further evidence of progress. We

hope that you will continue to drive safely and protect your

driving privilege.

Driver Improv

EXHIBIT H

“dummy contact” procedure was to identify all non-recipient controls so they could be removed from the study along with non-recipient experimentals (Exhibit I).

Criterion measures. The criteria for the evaluation of treatment effects were total accidents and countable traffic violations. Total accidents included all injury-fatal accidents and pro- perty damage accidents on file at the Department of Motor Vehicles. Countable traffic violations included convictions for violations involving the safe operation of a motor vehicle (moving) and the more serious equipment violations.

Statistical modei and procedures. Because of the unreliable and insensitive nature of the criterion measures, it was realized at the outset that large samples would be required to achieve adequate statistical power. A total sample of 18,000 was obtained for this study. Analysis of previous accident and violation distributions indicated that the above sample would enable one to detect a 10 per cent treatment effect with an a of 0.10. At this a level, the estimated /3 errors ranged from 0.50-0.75 for accidents and 0.01459 for violations. The a level was set at 0.10 in order to minimize the risk of not detecting a real effect.

After all subjects had been assigned to treatments, the adequacy of the random assign- ment procedure was assessed by comparing the treatment groups on several variables-

158 ROBIN S. M&RIDE AKD RAYMOS~ C. PECK

r

Your name and your car license number have been selected as part of an administrative survey to determine the accuracy of the Department of Motor Vehicles inforination about the addresses of licensed drivers and motor vehicle owners in California. In the near future the Department will use an electronic computer to maintain the drivers license files of all California drivers, and the information about the accuracy of drivers’ addresses is needed for this purpose.

Enclosed is a post card with your address. We would appreciate it if you would mark the appropriate box and do one of the following:

1. If your address is correct, please ouil the card, without changes, toXiG3Zpartment. No postage is necessary.

2. If your address is not correct, please correct it and then mail thzard to the Deoartment. NO

postage is necessary. If you wish ihe address changed on your registration card, please enter your car identification information.

Thank you for your cooperation in assisting us in this survey.

Very truly yours,

R. WARD, Manager Operations h Standards Section

Enc.

EXHIBIT 1

age, marital status, sex and prior drivin, 0 record. A one-way, multiple analysis of co-

variance (ANOCOV) technique was used to eliminate any possible bias and to reduce error variance. The adjusted treatment means obtained from the ANOCOV were then used as input for several factorial ANOV-Threat x Intimacy x Questionnaire, and Threat x

Intimacy x Questionnaire x Reinforcement. A posteriori and a priori contrasts were evaluated by Dunnetts and conventional t tests, respectively. In some cases, comparisons between treatments and controls were done after collapsing factors in accordance with a priori decision rules. First order (Subject x Treatment) interactions were evaluated by testing for differences in covariate slopes across treatment, and higher order interactions were evaluated through a special ANOV procedure for unequal cells (Federer and Zelen,

Modifying negligent driving behavior through warning letters 159

1966). The first 7 months subsequent drivin, 0 record was used to evaluate the effects of

initial contact and the last 5 months driving record was used to evaluate the effect of reinforcement. The majority of the analysis was based on parametric procedures although not ail assumptions were met. The effects of the violations were regarded as negligible because of the large samples, lack of strong evidence for heterogeneity of variance*, and the robustness of the F tests (Winer, 1962). In addition, non-parametric procedures were employed for further verification.

RESULTS

A comparison of the letter groups on the subject and prior driver record variables is presented in Table 1.

TABLE ~.SLJBJE~~VARIABLESAND PRIOR DRIVIKG RECORD BYTREATMENT

Treatments Sample size

Percent females

Variables Three year Three year

Percent Age prior prior single accidents violations

Total (all letter combinations)

HllLllQ HT/LIjNQ HTiffIlQ NT/HI/N& LVLIIQ LT/LI/NQ LTlHllQ LT/Hl/NQ ST@ Q STDINQ

Control

14620 15.35 48.92 30.02 0.68 4‘34 1462 16.00 46.78 30.44 0.69 4.35 1462 16.00 50.07 29.66 0.64 4.38 1462 15.46 50.55 30.77 0.67 4.35 1462 15.03 51.43 28.88 0.65 4.34 1462 14.57 45.96 30.86 0.68 4-34 1462 16.08 49.83 29.35 0.73 4.26 1462 15.59 50.96 29.70 0.69 4.36 1462 15.39 47-54 30.52 0.64 4.34 1462 16.55 48.29 29.99 0.69 4.32 1462 12.80 47.84 29.99 0.67 4.37 3019 15.67 50.X 30.15 0.65 4.29

Tests of significance xz = 11.56 ,$ = 20.57 F = 4.11 F = 1.91 F = 0.71 ns. P< 0.05 P<O.Ol PC 0.05 n.s.

Although the differences by treatment on the subject variables were slight, some are greater than could be expected by chance with these samplest. The age, marital status and prior driver record differences were all significant at the P < 0.05 level. Since these variables were correlated with subsequent driving record, a multiple ANOCOV method was used to control any bias and to increase the statistical power of the experiment. All the subject variables were entered as covariates, including a quadratic term for the non-linear component in the age-driver record trend.

Accident incolcement first secen months All of the covariates (subject and prior record variables) used in the ANOCOV proved

highly significant (P<O.OOj). In addition, the relationship between the covariates and

*Fm. = 1.16, (IO,io d.f.) for 1 year accidents-total sample. tInspection of the subject assignment log indicated that slight biases were introduced by attempts to equalize

the treatment cell numbers.

160 Roerx S. ?&BRIDE A.VD RAYMOND C. PECK

subsequent accidents was essentially equal among all treatments, as evidenced by the non- significant equality of slopes term F (60/a) = 1.16, P>O,X. This lack of evidence for a first order interaction (Treatment x Covariate) indicated that a covariance model assuming common regression slopes was appropriate.

The unadjusted and adjusted accident means are presented in Table 2. All adjustments were extremely minor due to the small correlation of subject variables with accidents and the small initial treatment biases.

TABLE 2. ACCIDENT ht.%xs 7 MONTHS SUBSEQUENT TO LETTER-QUESTIONNAIRE COSTACT'

LT HT Treatment LI HI L1 HI STD Control

Q 0.104 0.112 0.120 0.108 0.109

(0.103) (0.114) (0.118) (0.107) (0.108) 0.130

(0.130) 0.102 0.107 0.119 0.121 0.113

NQ (0.103) (0.105) (0.120) (0.123) (0.114)

*Values in parentheses represent unadjusted means.

The summary results* of the factorial ANOV on Threat x Intimacy x Questionnaire dimensions are presented in Table 3.

TABLE 3. ANOV SUMMARY OF 7 MONTH ACCIDESTS SUBSEQUENT TO TREATMENT

Source df MS F

Questionnaire (Q) 1 0.0039 < I.00 Intimacy (I) 1 OGIO6 < 1.00 Threat (T) 1 0.3701 3.05* QxI 1 0.0198 < 103 QxT 1 0.0674 < 1.00 IXT 1 0.0917 < 1.00 QxlxT 1 oxK30 < 1.00 Within cellst 17,622 0.1214

tError term based on total population selected for study including the control group.

*p<o.10.

The only significant effect was on the Threat dimension, where the subjects receiving the LT letters had significantly fewer accidents (P~0.10) than those subjects receiving the HT

letters. No significant main effects or interactions with the Idimacy and @estionnaire

treatment were found. In accordance with the CI priori decision rule, all factors not exhibiting suggestive main

*A one-way Chi-square analysis of the accident dichotomy (O-l) produced the following results: x2 (uncollapsed) = 17.3, 10 df, P<O,iO; ,yz (questionnaire collapsed) = 13.6, 5 df, P<O.O2.

Modifying negligent driving behavior through warning letters 161

effects or interaction (P < 0.25) were collapsed before making comparisons with the control. Thus, all Intimacy and Questionnaire treatment dimensions were combined under each Threat level for comparison with the control group, and the two STD letter groups were combined. This left the four groups-three experimental and one control-presented in Table 4. Dunnett’s test for multiple comparisons (Edwards, 1962) indicated that only the subjects who received the LT letters had significantly fewer accidents than the control

group (P < 0.05).

TABLE ~.TREATME&T COMBINATIONS vs. CONTROL GROUP ON 7 MOXHS WBSEQUENT ACCIDENTS

Treatments n Treatment

..F VS.

control

LT letters (collapsed on intimacy & ques.) HT letters (collapsed on intimacy & ques.) STD letter (collapsed on ques.) Control group

5848 0.1060 2.16* 5848 0.1172 1.14 2924 0.1112 1.46 3019 0.1298

*PiO.O5, Dunnett’s f (4, cc df) comparison between treatments and control group.

Countable violations seven months subsequent to treatment All covariates (subject variable) were significantly related to subsequent violations

(P<O.OOl) with the exception of prior accidents. The relationship between these variables and the criterion was again the same for all treatments as evidenced by the non-significant slopes across treatment F (6O/co) = 1.16, P >0.25.

A comparison between the unadjusted and adjusted means on violations is presented in Table 5. The adjustment on the violations means was small but somewhat larger than that for accidents.

LT HT Treatment LI HI LI HI STD Control

0.522 0.511 0.501 0.515 0.491 Q (0.512) (0.521) (0.490) (0.507) (0.486)

0.555 (0.554)

0.521 NQ

0.538 0,539 0.506 0.54 I (0.522) (0.533) (0.548) (0.52 1) (0.548)

*Values in parentheses represent unadjusted means.

Analysis of the threat, intimacy and questionnaire dimensions showed no significant differences as illustrated in Table 6. However, a three-way interaction was significant at P < 0.25, which precluded combining dimensions for comparison with the control group.

The STD letters could also not be collapsed on Questionnaire because of an excessive difference between the two groups on accidents (P < 0.25). Consequently, none of the letter

162 ROBIX S. MCBRIDE AND RAY&,IOND C. PECK

TABLE 6. ANOV SUM.MARY OF 7 .MO>~ VIOLATIOSS SUBSEQUEYI TO TRE ATME4T

Source df MS F

Questionnaire (Q) 1 0.56089 < 1.00 Intimacy (I) 1 0.0328 1 i 103 Threat (T) I 0.18249 c: 1~00 QxI 1 0.06875 < 1.00 QxT 1 0.00143 < 1.00 IXT 1 0.12354 c: 1.00 QxlxT 1 1.05556 1.58 Within cells* 17,622 0.6687

*Error term based on total population selected for study including the control group.

dimensions were combined under the a priori decision rule when comparing treatments with the control group.

Dunnett’s t,.oj statistics for 10 co df multiplied by the pooled error term. gives the difference between the treatment and control mean which must be exceeded for significance (2.42 x

0.0368 = 0.0895). An inspection of Table 5 indicates that none of the differences (T= - J!?c) exceeded the critical product.

The analysis thus far has shown a definite treatment effect on accidents, but nothing on violations. The t test matrices (Tables 7 and 8) provide further substantiation for an ac- cident effect and also suggest an effect on violations. All letters were directionally better than the control on accidents as illustrated in the t test matrix for adjusted means, Table 7, and six letters were significantly better (PC 0.05) and one at the (P~0.10). The remaining three letters were significantly better at the (P < 0.25). Since a large number of tests are made in this matrix, one would expect some significant t ratios by chance. However, the fact that

TABLE I. t TEST MATRIX FOR ADJUSTED 7 MONTH ACCIDENT MEANS

Treatment Treatment

HT/Ll/ HT/HI/ LTILII LT/HI STD/ HTILII NT/HI/ LTjLI/ LT/HI/ STD/

Q Q Q Q Q 1VQ 1VQ 1’JQ IVQ NQ

HT/LI/Q - HTIHIIQ 0.91 - LULIl Q 1.25 0.33 - LTI HIi Q O-67 -0.25 -0.58 =DiQ 0.88 -0.04 -0.37 G?l - HTjLIjNQ 0.06 -0.86 -1.19 -0.61 -0.82 - HTlHljNQ -0.06 -0.98 -1.31 -0.73 -0.94 -0.12 - LT/LI/NQ 1.42 0.50 0.17 0.75 0.54 1.36 1.48 - LTIHIINQ 1.05 0.15 -0.19 0.04 0.19 1.00 1.12 -0.37 - STDI NQ 0.52 -0.39 -0.73 -0.15 -0.36 0.46 0.58 -0.90 -0.54 - Control -0.86 - 1.92 -2.31 -1.63 -1.88 -0.93 -0.79 -2.50 -2.09 -1.46

P<O.Ol = 2.33. P<O.O5 = 1.64. P<O.25 = 0.67.

Modifying negligent driving behavior through warning letters 163

all letters were directionally better and a relatively large number were significantly better

than the control gives strong support for an overall letter effect. Comparisons between

letters can also be made from the matrix in Table 8.

TABLE 8. t TEST MATRIX FOR ADJUSTED 7 MOXTH VIOLATION MEASS

Treatment

Treatment

HTILII HTIHII LT/LI/ LTIHI/ STDj HT/LI/ HT/HI/ LT/LI/ LTIHII STDJ Q Q Q Q (2 NQ NQ NQ NQ NQ

HT1.W Q HTI HII Q LTILII Q LTIHIIQ STDI Q HTILIINQ HT/HI/NQ LTILIINQ LT/HI/NQ STDINQ Control

- -0.46 -0.70 -0.33

0.33 -1.27 -0.15 -0.66 -1.20 -1.32 -2.08

- -0.24

0.13 0.79

-0.81 0.31

-0.20 -0.74 -0.86 -1.54

- 0.36 - 1.03 0.67 -

-0.57 -0-94 -1.60 - o-54 0.18 -0.49 1.12 - 0.03 -0.33 -1.00 0.60 -0.51 -

-0.51 -0.86 -1.57 0.03 -1.08 -0.55 - -0.62 -0-99 -1.65 -0.05 -1.17 -0.65 -0.08 - -1.27 -1.67 -2.47 -0.61 -1.90 -1.31 -0.64 -0.55

P<O.Ol = 2.33. P-CO.05 = 1.64. P<O.25 = 0.67.

The 1 matrix, for violations (Table 8) is at least suggestive of a treatment effect. All letter treatments are directional/y better than the control and six of the ten letter treatments are signij?cuntZy better at the P~0.05. At first glance, this tiding may seem inconsistent with the Dunnett’s procedure. The difference is due solely to the greater alpha protection (and beta error) associated with the more conservative Dunnett test.

Accident involvement following reinforcement The covariates, sex, prior accidents and prior violations, were significantly related to

accidents 5 months subsequent to reinforcement (P~0.05) and the combined effect of all covariates was highly significant (P~0.01). As in the first seven month evaluation, the slopes were not significantly different across the treatments (P>O.25), indicating that an additive covariate model (no covariate x treatment interaction) was justified. The adjusted and un- adjusted accident means are presented in Table 9.

The summary results of the four-way ANOV on the adjusted means is presented in Table 10.

No significant effects were found on any of the four factors. However, Reinforcement had a marginal main effect and was contained in two significant interactions. It was a significant factor in combination with Threat in a two-way interaction and both Threat and Intimacy in a three-way interaction.

Subjects who initially received a LT letter had fewer accidents than those subjects who received a HTletter under the Reinforcement letter conditions. The converse is the case under the No Reinforcement condition, where subjects who received the HT letter had fewer accidents than those who received the LT letter in the reinforcement period.

A.&P. 2/3-c

164 ROBINS. MCBRIDE AND RAYMOSD C. PECK

TABLE 9.iiCCIDENT MEAXS j MONTHSSUBSEQUENTTO REISFORCEMEhT*

L‘T i-/T Treatment LI HI LI HI STD

a 1va Q IVQ Q XQ Q NQ Q NQ

0,060 R 0.072 0.046 0.03 1 0~018 0.065 0.072 0,056 0.070 0.071 (0.059) (0.072) (0.047) (0.030) (0.045) (0.065) (0,073) (0.057) (0.070) (0.073)

NR 0.07 1 0.059 0.075 0.052 o-045 0.065 0.073 oG4s 0.063 0.042 (0.071) (0.060) (0.075) (0,082) (0.04s) (O-064) (0.073) (0.046) (O-063) (0.042)

*Values in parentheses represent unadjusted means.

TABLE 10. ANOV SUMMARY OF 5 MONTH ACC~DENTSSUBSEQUENII

TO RE~~~ORCE~tE~~

source df MS F

Reinforcement CR) 1 0.1536 2.31 . _ Questionnaire (Q) Intimacy (1) Threat (r) RxQ RXI RXT

Qxl QxT IXT RXQXI RxQxT RxlxT QxlxT RxQxlxT Within cellS

1 0.0062 1 OGiIO8 1 O-0150 I 0.0037 1 0.1640 1 0.2128 1 0.2058 I 0~0001 1 0.0715 i 0.0397 1 OGO21 1 0.2396 1 o- I347

0.0868 0.0664

< IGil -=c 130 < 1al <iGO

2.47 3.20* 3.10*

-Cl.00 1.08

< lal < 1.00

3.61* 2.03 1.31

tError term includes standard letter treatments. *pto.to.

The significant Questionnaire x Intimacy interaction is surprising, in that it is unrelated to the reinforcement factor. Subjects who were sent a Questionnaire with a LI letter had fewer accidents than those not sent the ~ues~j~n~~ire in the reinforcement period. Subjects issued a fil letter without a questionnaire had fewer accidents than those issued a question- naire. Why such an interaction would emerge after seven months is difficult to explain, other than chance significance.

Under the Reinforcement condition, subjects who initially received the LT/HI letter had fewer accidents than those who received the LT/LI letter. Subjects who received the Hr/LI letter had fewer accidents than those subjects who received the HT/HI letter. Most of the interaction effects were contributed by an initial treatment combination under the rein- forcement condition. Under No-Reinjbrcement, the slope differences between ieveis of intimacy over levels of threat were slight.

Modifying negligent driving behavior through warning letters 165

Analysis of violations for the reinforcement period

Table 11 depicts the adjusted and unadjusted means by treatment for the Reinforcement period. All covariates, with the exception of prior accidents, were significantly related to violations in the last five months (P<O,Ol). The combined contributions of the covariates was highly significant (P-C 0,001). The equality of slopes value was not significant, P >0.25), supporting the use of a common regression slope variance model. Analysis of violation means revealed no significant main effects due to Reinforcement, Questionnaire, Intimacy or Threat as shown in Table 12. A highly significant interaction was found on Threat x Intimacy (P<O.OOl). This finding presents a paradox similar to the significant Questionnaire x Intimacy interaction on accidents, in that an interaction not evident in the first 7 months

emerged in the last 5 months and is unrelated to Reinforcement. This suggests some form of time interaction in which letters of contrasting levels (LT/HI, HT/LI) have more enduring effects on violations. Subjects who were issued the LT/LZ letters had fewer violations than

TABLE 11. VIOLATION MEANS 5 MOSTHS SUBSEQUENT TO REDTORCEMENT*

Treatment LT HT STD

LI HI LI HI

Q NQ Q NQ Q NQ Q 1VQ Q NQ

0.323 R

0.313 0.218 0.274 0.278 ,027s 0.307 0,294 0.276 0.230 (0.320) (0.312) (0.222) (0.272) (0.265) (0.275) (0.301) (0.307) (0.273) (0.237)

NR 0.295 0.295 0.244 0,272 0.282 0.229 0.302 0.285 0.273 0.263

(0.291) (0.300) (0.249) (0.271) (0.285) (0,223) (0.304) (0.287) (0.275) (0.263)

*Values in parentheses represent unadjusted means.

TABLE 12. ANOV SUMMARY OF 5 MONTH VIOLATIONS SUBSEQUENT TO REINFORCEMENT

Source df MS F

Reinforcement (R) Questionnaire (Q) Intimacy (I) Threat (T) RxQ RxI RxT

QxI QxT IxT RxQxI RxQxT RxIxT QxIxT RxQxIxT Within csllt

1 0.1742 1 OX)035 1 O-2552 1 O%lO78 1 0.1662 1 0.2786 1 0.0349 1 0.4285 1 0.7420 1 3.4225 1 00325 1 0.0404 1 0.0495 I 0.1344 1 0.2103

9339 0.3267

<lW Cl40 <I03 <l%Xl <I40 <14il <140

1.31 2.27

10.48* cl.00 <lGl <lG.J c I-00 <I~00

*p<o-001. iError term includes standard letter treatments.

I66 ROBN S. MCBRIDE AND RAYMO~ C. PECK

those issued the LT/LI letters. Subjects issued HT/LI letters had fewer violations than those issued HT]HI letters.

t Test comparisons of Reinforcement effects on accidents and violations indicates that Reinforcement would be effective in reducing accidents with a LT/HI letter. There is no evidence of a Reinforcement effect on violations, however.

TABLE 13. COMPARIMN OF REINFORCEME~ EFFECTS ON ACCIDENTS BY TREATMEFI (f TEST MATRIX)*

Treatment LT HT

LI HI LI XI STD

Q -0.655 -1.738 -0-036 -0.063 0.418 NQ 0.767 - 3.035 -0GO4 0.451 1.719

*Negative sign indicates lower mean for reinforced subjects.

TABLE 14. COMPARISON OF REINFORCEMENT EFFECTS ON VIOLA- TIONS BY TREATMENT (r TEST MATRIX)*

LT HT LI HI LI HI

Q 0.745 -0.700 -0.130 0.121 NQ 0.048 0.060 1.228 0.245

*Negative sign indicates lower mean for reinforcement subjects.

It should be noted that the analysis for the Reinforcement period was confined to only those subjects who were incident free for the first seven months. It is theoretically possible that Reinforcement or Reinforcement interactions with initial treatments may be significant for subjects ~~c~de~f~ee, but these subjects may not have fewer accidents or violations than subjects not contacted at all (control group). Evaluation of the effectiveness of Reinforce- ment as a total operational program should therefore entail a discussion of the performance of the control group and subjects not eligible for reinforcement.

Table 15 gives the appropriate accident means for making total program comparisons in the last five months. The overal Reinforcement program means were obtained by pooling Reinforced and non-eligible subjects; the overall No Reinforcement program means were derived from the pooling of non-reinforced and non-eligible subjects.*

The data in Table 15 strongly suggest that initial letter effects wear-off in the last five months and that Reinforcement is not generally effective in reducing accidents. When all treatments are combined, neither Reinforcement nor initial treatment is even directionally superior to the control group in the last five months.

*The eligible but not reinforced Ss represent one-half of the random sample of incident free drivers for the first seven months. When pooling the means for the total sample, the non-reinforced sample means were weighted by twice their sample size,

Modifying negligent driving behavior through warning ietters 167

TABLE 15. COMPARIKX-4 OF REINFORCEMENT ASD SO-REINFORCEMENT LEl7ER PRCGRAMS

WlTH THE COSTROL GRGLT

Groups No-reinforcement program

Treatment programs Reinforcement

program Control

Means 0.0685 0.0664 0.0639 Variance 0.0738 0.0693 0.0665

Treatment by subject interactions The following analysis was directed toward the development of an optimum procedure

for dispersing warning letters to sub-populations of negligent drivers based on accident reduction. At this point, accident frequency was selected as the criterion instead of viola- tions because of its greater social relevance. The analysis was concerned with the differential effect of the warning letters by age and marital status on accidents (i.e. treatment x subject interaction). It will be recalled that the presence of first order interactions was evaluated earlier by comparing covariate slopes across treatments and that no evidence of first order interaction (subject x treatment) was evident. The present section employs a different method of testing interaction, and the analysis is extended to interactions of higher order. Since males represent approximately 85 per cent of the negfigent driver population, decisions regarding the issuing of various types of warning letters should be based primarily on the male population. Only when dramatic treatment by sex interaction occurs, can separate warning letter programs for males and females be seriously considered. A treatment x sex analysis showed no significant interactions by sex (P > 0.25). Although a combined analysis could be justified at this point, the sexes were analyzed separately to prevent any possible obscurement.

To simplify computations, the collapsed treatments resulting from the initial seven month analysis were used. Thus, the four groups were analyzed by age and marital status

for males and females separately: LT, NT, STD letters and the Control. Tables 16 and 17 show the accident mean for males and females respectively.

TABLE 16. SUBSEQUENT 7 MONTHS ACCIDENT MEANS BY TREATNEXT, AGE AND MARITAL STATUS (MALES}

Treatment Age

Total* 16-30 3140 41-60 Married Single Married Single Married Single

LT 0.098 O-082 0.140 o.oss 0.068 0.101 0.110 HT 0.118 0,116 0.143 0.075 0155 @I04 0.114 STD 0.092 0.124 0.126 0.089 0.022 0.105 0.083 Control 0.126 0.126 0.162 O-097 0.108 0,125 0.134

Total 0.108 0127 0.088 0.110 Married 0.103 Single 0.114

* Marginal total entries represent the unweighted means across cells and will not necessarily conform to the treatment means of prior analysis.

168 ROBIS S. MCBRIDE AND RAL’MOSD C. PECK

TABLE 17. SGBSEQUEST 7 UOSTH ACCIDENT MEA% BY TREZIT~IENT, AGE .A\D MARITAL STATUS (FENALES)

Treatment TOtal*

Age 16-30 31-60

Married Single Married Single

LT 0.080 0.063 0.092 0.056 0.107 HT 0.091 0.100 0.143 0.046 0.076 STD 0.104 0.09s 0.133 0.119 0.074 Control O.OSS 0.062 0.157 0.0% 0.074

Total 0,091 0.105 0.076 Married 0.074 Single 0,107

*Marginal total entries represent the unweighted means across cells and will not necessarily conform to the treatment means of prior analysis.

The unadjusted means were used since the data were not available by the age and marital status levels for adjusted means. However, the factorial ANOV operates essentially the same as the ANOCOV, in that the effects of subject variables are controlled by the strati- fication made in the factorial ANOV. A technique developed by Federer and Zelen (1966) was used to compute the sum of squares in the factorial AKOV for unequal cell numbers. The differences in cell number size were due mainly to base driver population distributions on age and marital status, rather than differences on these variables by treatments.

Tables 18 and 19 give the ANOV summary for males and females respectively. The ANOV (Table 18) for males shows a significant treatment effect (PcO.01) which is

consistent with prior analysis for males and females combined. Age was highly significant at the PxO.01, forming a U-shaped curve on accidents. That is, drivers under 30 and over 41 had higher accident means than drivers between the ages of 31-40. The two-way and three-way interactions, Treatment x Maritul Status and Treatment x Age x Marital Status, were significant at the P < 0.10 and P< 0.01, respectively.

TABLE 18. ANOV SUMMARY OF 7 MONTH ACCIDENTS BY TREATMEST, AGE AND hlARITAL STATUS (hlALES)

Source ss df MS F

Treatment (T)

Age (A) Marital status (IMS) TxA T x ic1.S A x MS T x A x MS Within cells

1.1736 3 0.3912 3.066t 2.1656 2 1.0828 8.486:

0.2064 1 @2064 1.618

0.6808 6 0.1135 0.889

0.9617 3 0.3206 2.512*

0.5154 2 0.2577 2.020

2.4156 6 0.4026 3,155:

1901.2842 14906 0.1276

*P< 0.10. TP < 0.05. p-c 0.01.

Source

Treatment (T) Ase (4 Marital St3_tUS (.Cf.s) T :< ‘4 T Y :WS A x .tts T x A x .WS Within cells

*PC 0.05.

Xlodifymg negiipznt driving behavior through warning letters 169

TABLE 19. .-iNOV SLXMARY OF 7 MOSTH ACCIDESTS BY TREAThtEh-r, AGE Ab’D

MARITAL sr.*Tcs (FE.MALES)

SS di MS F

0,1560 3 0.0510 0.593 04376 1 04576 5.2lS* 0.5976 1 0.5976 6.811* 04764 3 0.15SS 1,811 0475s 3 0.0596 0,680 O-2256 1 0.2256 1.571 0.3473 3 0.11% 1-370

I!36.8686 2,lO i O-0577

Because of the large number of statistical tests required and reduction in sample size, tests of significance were not conducted on the individual components of interaction. For single, male, negligent drivers under the age OF 30 and 31-40, the standard letter ap- peared most effective. This same tendency was also apparent for the oldest age group, suggesting that continued issuance of the STD letter to the single, negligent driver is justified from an optimum treatment standpoint.

The trends for married males were not as clear as those for single males. However, the LT Ietter was directionally the more effective treatment for married negligent drivers, particularly those under 30.

Summary Table 19 for females shows highly significant main effects due to age and marital status. No significant treatment effects or interactions kvere found. Comparison of the female accident means across treatments indicate that the low threat letter had the lowest or second lowest mean except for single females over the age of 30. This would suggest that the best single strategy for females would be use of a LT contact.

Non-parametric-cost benefit analysis Presented in Table 20 is a cost comparison of the four experimental warning letters,

standard letter and the control group on two criteria of program success: (1) negligent operator points, and (2) mean number of departmental actions during the subsequent one year period. Only the negligent operator point variable was tested for significance, since it underlies the action variables and is less subject to artifacts such as departmental work load and variations in discretionary selection procedures. The chi-square was significant (x2 = 12.8, 5 df, P<O.O5), which conforms to the parametric t test findings presented earlier.

Looking at actions, it is clear that the control group did poorly compared to the letter treatments, particularly the STD letter. The data in Table 20 suggests a maximum savings of 3.1 discretionary actions per 100 drivers (14.9-11-8). The cost of departmental actions such as driver improvement meetings (S2.8 1) and individual hearings (S13.84) for each treatment are also included in this table. The cost rates were based on follow-up discretionary action costs per initial warning Ietter issued.

Since the control group is not issued a warning letter, a saving of 61 cents would have to be obtained to offset the cost of issuing a warning letter. The best letter in Table 20 resulted in a total driver improvement program cost of 52.08 (61 cents cost of warning letter + $1.47 actions). This cost exceeds the no letter cost (control) of $1.87, indicating that the ne

170 ROBIX S. MCBRIDE A?XJ RAYMOND C.PECK

TABLE 20. COST BENE~ OF WARMSG LEITER PROGRAM-DEPARTMENT

ACrlOSS(ONE YEAR)

Treatment Per cent of iMean number cost of

drivers with of actions per actions per 1 or more XOP 100 drivers W/L issued (5)

HTILI 56.2 13.5 1.7’ HT,‘HI 57.4 13.3 1.66 LT/Ll 55.8 13.3 1.69 LT/HI 54.9 12.9 1.63 STD 55.7 11.8 147 Control 58.9 14.9 1.87

cost of issuing the letter is 24 cents. It should be emphasized that this figure is highly dependent on the stringency of DMV follow up. If the Department were to take action after the first subsequent negligent operator point, then the warning letter would be com- pletely self supporting since action savings would be distributed over more people. The present data indicates that action was often not taken until approximately three or more subsequent NOP’s were accrued, even though DMV’s policy would permit action at a lower count in most cases.

Much more dramatic savings were evident in terms of accident cost reduction.* Table 21

TABLE 21. COST BENEFIT OF WARXISG LElTER PROGRAM-ACCIDENTS (ONE

YEAR)

Treatment Assumed reinf. Assumed no reinf.

program program Accident Savings/ Accident Savings/

mean letter(S)* mean letters(L)*

HT/LI 0.182 25.20 0.182 25.20 HT/ HI 0.185 18-90 0.182 25.20 LTILI 0.176 37.80 0.176 37-80 LTI HI 0.162 67.20 0.187 14-70 STD 0.185 18.90 0.173 44.10 Control 0.194 - 0.194 -

*Cost of identifying reinforcement subject not included in cost or reinforcement program.

depicts the average savings per letter issued as compared to the control group for the follow ing two categories: (1) accident reduction for the total one year period assuming a Re- inforcement program, and (2) accident reduction for the total one year period assuming a No Reinforcement program.

Under an assumed Reinforcement program, the LT/HI letter resulted in a very favorable accident cost reduction. However, the LT/HI-Reinforced letter is not markedly superior to the LT/LI/No Reinforced, or STDINo Reinforced letters. Without further research sub- stantiating the reinforcement effect on a LT/HI letter, the authors hesitate to recommend a

reinforcement program at this time.

*The direct cost of reportable accidents used in this study was S2100 per accident (California DMV, 1968).

Modifying negligent driving behavior through warning letters 171

DISCUSSION

In general, the data in the preceding sections strongly support a warning letter program as compared to no program at all (See Table 22 for summarization). Low threat letters and the STD letters appeared to be the most effective letters for accident reduction up to a seven month period. Thereafter, the effects appear to diminish. The intimacy dimension and questionnaire did not have any significant effect on accidents or violations, which suggest that the threat appeal is the most relevant dimension for motivating drivers to improve.

TABLE 22. SUMMARY OF FIRST 7 MOSTHS ANALYSES (ACCIDEZITS)

Sample Experimental letters Standard letters Control vs. all treatments

Total sample

Threat dimension (low) significant ANOV P < 0.10

No significant in- timacy or question- naire elfect

No significant difference on questionnaire nor significantly different than experimental letters

All letters better than control. Low threat letters significant P < 0.05

Low threat letters significantly better than control P<O.Ol

Standard letters significantly better than control P< 0.01

All letters better except high threat for single 3140 age

Males (all dimensions collapsed except for threat)

Age significant PiO.10 Treatment by M.S.; significant P<O.lO Treatment by M.S. by age; significant P-co.01

Females (all dimensions collapsed except for threat)

No significant treatment effect or interactions

Age; significant PC 0.05 M.S.; significant PcO.05

The fact that the low and moderate threat (standard) appeal were more effective than the high threat appeals in reducing accidents (first seven months) seems to conform to the hypothesis inferred from prior studies on the effectiveness of mass communications (Janis and Terwillinger, 1962) :

“When a relatively high level of fear is induced by the warnings presented in a persuasive communication, the recipients will become motivated to develop psychological resistances to the communicators’ arguments, conclusions and recommendations.” Inferences about the effectiveness of persuasive communications in modifying behavior

probably involve several levels of explanation. There are several potential intervening variables which could influence the effectiveness of communications; the personality and

172 ROBIN S. LMCBRIDE ASD RAYMOND C. PECK

attitudes of the individual, the content and style of the message, and the delivery of the communicator, to name a few. A change in behavior may occur on a peripheral and cog- nitive level or on a more underlyin g, affectual level. An individual may improve his driving

behavior to avoid fines, restriction of drivers license and other forms of punitive actions without modifying basic attitudes. On the other hand, an individual, through introspection and other psychological processes, may modify existing attitudes and potentially change his basic driving habits.

Evidence for the interaction between the personality characteristics of the recipient of a persuasive message and the content of the message was shown in a study concerned with changing dental hygiene practices (Janis and Feshback, 1954). The authors concluded that the “communication” which contained a high degree of threat tended to stimulate defensive reactions and that acceptance of the message was dependent upon the basic “anxiety” level of the subject at the time of the message. Subjects classified as low on anxiety tended to be more accepting of a high threat communication, whereas those subjects classified as high on anxiety were more rejecting of a high threat communication.

For many drivers, a license revocation would have rather drastic socio-economic implica- tions. Although the anxiety level of subjects in this study was not known, it does not seem unreasonable to assume that the warning letters elicited different emotional responses. Indirect evidence for this was found in a previous analysis of the returned questionnaire data (Peck, 1968). The recipient was asked to give his reaction to receiving the warning letter by indicating his degree of anger, disgust, fear [scared) and upset on a five point scale. These reactive responses were then correlated with the threat-intimacy intensity of the letter. Significant correlations (P < 0.01) were obtained between the reactive variables, “upset” and “scared” and threat level of the letter. That is, higher upset and scared responses were given by subjects who received high threat letters.

That the intimacy dimensions did not even approach significance or interact with threat in reducing accidents came as somewhat of a surprise to these authors. Previous research (Kaestner, Warmoth and Syring, 1967) showed that a personalized approach was effective for changing driving behavior for drivers under 25. The personalized letter from Kaestner’s

study appeared to be quite similar to the HI letters in this study. There were some differences in the content, however, which could account for the discrepancy in findings. Also ad- ministrative procedures, geographical differences and population differences make direct comparisons difficult. In this study, the LT and STD letters, both very formal, were the best overall treatments. It may be that the credibility of the HZ letter was questioned by recipi- ents and/or a LI, rather bureaucratic letter conformed with the expectation of the recipient. There is also evidence from the questionnaire responses, that HI increased the threat intensity of the LT letter (Peck, 1965). However, this lack of orthogonality exhibited by the questionnaire reactivity items was not accompanied by a similar interaction on the sub- sequent driving record variables in this study.

Although the questionnaire provided potentially important attitudinal and normative data, the questionnaire as a treatment does not seem to be warranted for an operational warning letter program. This conclusion is based on the lack of statistical evidence for a questionnaire effect. If further analysis indicates that the questionnaire responses have significant prognostic utility regardin, 0 future driving record, their use could, of course, be

justified on that basis. There could be several reasons why the follow-up letter did not generally prove to be an

effective strategy. Although based on a reinforcement paradi,sm-that of increasing the

blodifying negligent driving behavior through warning letters 173

probability of a desired response-it may not have had functional relevance for driving.

In conditioning theory and experimentation, the reinforced response is usually some well defined and directly measured activity, with reinforcement closely following the desired response. In the present study the reinforced response (improved driving) was not measured directly and occurred over a seven month period of time. While it is still pos- sible that such a reward could be an effective motivator, its operation may not stem from reinforcement principles. It is also possible that the letter simply had no reward value for this population. The content, timing and frequency of the reward (letter) may not have been adequate. Perhaps a reward such as removing negligent operator points from the record would have been a more effective motivator. Further research should be directed toward developing new reinforcement strategies, such as increased frequency and greater functional significance of the reward. In addition, a negative reinforcement (punishing letter) for those subjects who are involved in driving incidents may have some merit. Although many theorists have not advocated negative reinforcement as a means of deterring undesirable behavior, there is evidence to indicate that negative reinforcement can be effective in

conditioning avoidance behavior (Solomon, 1964). The intricacy of the relationship between the initial stimulus (letter) and reinforcement

was illustrated by the significant Intimacy x Threat x Reinforcement interaction. Ap- parently, when the initial contact is highly intimate and of low threat, the follow-up letter did function as a positive reinforcer. This su,, =uests that the Reinforcement letter functioned to verify for the subject the intended implications of the LT/HZ strategy-namely, of a benevolent Driver Improvement Analyst concerned with the subject’s well-being and success.

The fact that the warning letters had greater impact on accidents than on convictions V,XS unexpected. Prior research has shown traffic violations to be a much more stable and sensitive index of driving behavior than accidents. Compared to accidents, violations occur much more often, exhibit higher intercorrelations over time period, and are more highly correlated with most psycho-social variables (Goldstein, 1964). In addition, most other studies on driver improvement and driver training have demonstrated a greater capacity for reducing violations than accidents. The only explanation these authors can offer for this deviation is that the effective warning letters actually altered the underlying driving at- titude or habits of certain recipients. Instead of merely obeying laws, some drivers were possibly motivated to drive more defensively.

The development of an effective warning letter is particularly significant in view of the nature of accident involvement. Prior research has consistently demonstrated that accident populations are largely transitory, with most accidents involving previously accident-free drivers. Because of this transitory aspect, it is obvious that the overall accident rate of the total driving population cannot be markedly reduced by focusing on a select, aberrant, population. Attempts should therefore be made to reach a larger proportion of the popula- tion by an effective but inexpensive program. One way of accomplishing such an objective would be to send warning letters to drivers as soon as they begin to accumulate violations- say, after the second violation. Not only would this provide a much larger base for the reduction of accidents, but it is possible that the letter might have greater impact on low count drivers and on drivers contacted early in their violation careers.

Another method of increasing the impact of the warning letter program would be to take advantage of known subject x treatment interactions. While the present study in- dicated some interaction by age and marital status, the practical benefits of sending different warning letters to various se,ments of the population must be weighed against its increased

174 ROBIN S. MCBRIDE AND RAYMOXJ C. PECK

administrative complexity. From the standpoint of accident reduction, the present findings suggest that a LT letter should be used for female drivers and married males under 30, whereas a STD letter should be used For all others. Such a strategy seems operationally feasible and would take advantage of most of the interactions indicated by this study.

Acknowledgemenr-IMajor support for this study was provided by the Bureau of Public Roads, Contract NO. HPR-PR-I (6) BC014, Interagency A.greement No. 13513. The study was conducted by the Research and Statistics Section under the generai direction of Ronald S. Coppin, Chief of Research, wiih the coopera- tion of Keith Ball. Chief. Division of Drivers Licenses. and Ronald Thunen. Chief. Division of Field Offices.

Acknowledgem&t anh appreciation is extended tb the many individuhls Lvhb supplied invaluable as- sistance throughout the project. Particular appreciation is extended to the following personnel in the Division of Drivers Licenses; Paul Thorgrimson, Assistant Chief; Lloyd Brad!ey, Driver Improvement Manager; Carl Morgan, Driver Improvement Analyst; Pat Stefani, Supervising Clerk; and Nancy Graham, Super- vising Clerk (retired).

William Marsh, Associate Statistician, is acknowledged for his role in the formative stages of the project. The opinions, findings, and conclusions expressed in this publication are those of the authors and not

necessarily those of the Bureau of Public Roads.

REFERENCES

CALIFORNIA DEPARTMENT OF MOTOR VEHICLES (1968). Research and Statistics. 1966 accident costs. Un- published.

CAMPBELL, B. J. (1958). Dicer improvement: The point s,vstem. University of North Carolina, Chapel Hill, North Carolina.

CHALFANT, M. and KING, G. (1960). Reports of studies on the effectiveness of driver improvement pro- cedures. Michigan State University, Michigan.

COPPIN, R. S. (1965). A controlled evaluation of group driver improvement meetings. Department of Motor Vehicles, Sacramento, California.

COPPIN, R. S., MARSH, W. C. and PECK, R. C. (1965). Are-evaluation of group driver improvement meetings. Department of Motor Vehicles, Sacramento, California.

COPPIN, R. S., PECI;, R. C., LEW, A. and MARSH, W. C. (1965). The effectiveness of short individual driver improvement sessions. Department of Motor Vehicles, Sacramento, California.

EDWARDS, A. L. (1962). Experimenfal design in psychological research. Holt, Rinehart, & Winston, New York. FEDERER, W. T. and ZELEN, M. (1966). Analysis of multifactor classification with unequal numbers of ob-

servations. Biomefrics, 22 (3), 525-552. FLESCH, R. (1948). A new readability yardstick. J. appl. Psychol. 32, 221-233. GOLDSTEIN, L. G. (1964). Human variables in traffic accidents: a digest of research. Traffic Safety Research

Review 8 (l), 26-30. JANIS, I. L. and FESHBACH, S. (1954). Personality differences associated with responsiveness to fear-arousing

communication. /. Pers. 23, 153-166. JANIS, I. L. and TERWILLIGER, R. F. (1962). An experimental study of psychological resistance to fear-

arousing communication. /. abnorm. Sot. Psychol. 65, 403-410. KAES~ER, N. (1968). Research in driver improvement: The state of the art. Trafic Quarterly 23 (4), 497-520. KAESTNER, N., WARMOTH, E. J. and SYRING, E. M. (1967). Oregon study of advisory letters: The effective-

ness of warning letters in driver improvement. Trajic Safety Research Review 11 (3), 67-72. MARSH, W. C. (1969). Modifying negligent driving behavior: a preliminary evaluation of selected driver

improvement techniques. Department of Motor Vehicles, Sacramento, California. MCBRIDE. R. S. (1967). The measurement of warning letters on two dimensions: Threat and intimacy. Un-

published Masiers Thesis, Sacramento State Colleie, Sacramento, California. PECK. R. C. (1968). The orediction of driver behavior subsequent to receipt of official letters of reprimand.

Unpublishid Misters Thesis, Sacramento State College, Sacramento, California. SOLOMON, R. L. (1964). Punishment. Am. Psychol. 19, 239-353. TEMPLE, A. F. and FERGUSON, T. G. (1958). Safety letter survey. Department of Public Safety, Austin, Texas. TOMS, D., KASTELLE, C. and LE SUER, C. (1965). An evaluation of the Pierce County pilot program in driver

improvement utilizing the group discussion. Department of Motor Vehicles, Olympia, Washington. WINER, B. J. (1962). Statisticalprinciples in experimental design. XicCraw-Hill, New York.