CHAPTER 4

THE EFFECTIVENESS OF CHMSL FOR LIGHT TRUCKS


Before the CHMSL requirement was extended from passenger cars to light trucks (pickup trucks, vans and sport utility vehicles), there were questions if CHMSL would be as effective in trucks as in cars, or even if it would be effective in trucks at all. By now, CHMSL have been in light trucks for several years. Statistical analyses of calendar year 1994-96 crash records from six States indicate that light trucks equipped with CHMSL were approximately 5 percent less likely to be struck in the rear than light trucks without CHMSL. This point estimate of CHMSL effectiveness in light trucks is statistically significant, and it is very close to the long-term effectiveness in passenger cars (4.3 percent). Nevertheless, the uncertainty in the light truck estimate precludes the inference that CHMSL are as effective in trucks as in cars.


4.1 Analyses of six State files: method

The launch of CHMSL on light trucks differs from their introduction on passenger cars in several important features that will influence the analysis method.

When passenger cars were first equipped with CHMSL, model year 1986, the lamps were a new experience for the motoring public. The initial reaction to the lamps was perhaps quite different from what would happen when drivers became acclimatized to them. That made it critical to estimate effectiveness separately in each calendar year, to distinguish between the initial and the long-term effectiveness. By the time that the first CHMSL appeared on light trucks - in the MY 1991 Dodge Caravan, Plymouth Voyager, Chrysler Town & Country and Ford Explorer - the public was already acclimatized to the lamps (on cars). It is less important to estimate effectiveness separately in each calendar year, since it may be assumed that the long-term effectiveness was reached immediately.

CHMSL implementation in passenger cars was nearly simultaneous: MY 1986 in all models (except Cadillacs and very few others). In light trucks, CHMSL were phased in over a three-year period, MY 1991-94, since some manufacturers installed them on certain models up to three years before the MY 1994 effective date. We must know the specific make-model of each truck to determine if it had CHMSL (unlike cars, where it was basically good enough just to know the model year). State files that include the Vehicle Identification Number (VIN) on their crash records allow identification of truck make-models by decoding of the VIN. NHTSA has access to six State files that include VINs, document the impact location on the vehicle, and are available for some or all of the calendar years 1994-96:

Florida Maryland Missouri
North Carolina Pennsylvania Utah

The Indiana and Texas files employed in Chapter 2 include "make-model" codes, but they are not detailed enough to distinguish between all the specific truck models addressed in this chapter; the Virginia file does not contain make-model information.

The basic contingency table analyses of Chapters 2 and 3 compared the proportion of crashes that were rear impacts in all cars of the first year with CHMSL - MY 1986 - to the corresponding proportion in all cars of the last year without CHMSL - MY 1985. Here, too, we will compare all trucks of the first year with CHMSL to all trucks of the last year without CHMSL, but both of these groups will include a mix of model years, since different models got CHMSL in different years.

The "vehicle age effect" - the trend of decreasing proportions of rear impacts as cars get older, for reasons unrelated to CHMSL - was calibrated for passenger cars in Chapter 2 by looking at this trend for cars 0-15 years old in many calendar years of data. It cannot be assumed that light trucks will have the same "vehicle age effect" as cars. At the same time, it is not necessarily appropriate to use a similar calibration procedure for trucks as for cars. The method of Chapter 2 assumes that CHMSL were introduced in all vehicles at the same time (correct for cars, but not for trucks). In addition, major changes in the truck population since 1985 raise doubts whether the effect in recent years should be calibrated from older data. Instead, the method for handling the vehicle age effect will be essentially the same as in Farmer's evaluation of passenger car CHMSL [9]. This method, as will be described below, relies entirely on data for the last two MY without CHMSL and the first two MY with CHMSL.

The starting point for the analysis is a list of make-models that were equipped with CHMSL. Information on CHMSL installations was obtained from Chrysler, Ford, General Motors, Mazda, Nissan and Toyota. The make-models of pickup trucks, vans and sport utility vehicles (SUV) included in the analyses are listed in Table 4-1. For example, Dodge Caravan and Plymouth Voyager were equipped with CHMSL in MY 1991; thus, the last two MY without CHMSL are 1989-90 and the first two MY with CHMSL are 1991-92. Note that all GM and Mazda trucks, and a substantial proportion of the other trucks, got CHMSL in MY 1994, simultaneous with the Federal requirement.

Excluded from the analyses are the Jeep Grand Cherokee, Ford Explorer, Ford Windstar and Toyota T-100, which had CHMSL from their inception, and which did not have a similar predecessor vehicle without CHMSL. Also excluded is the Dodge Ram Van and Wagon, which got CHMSL in the middle of MY 1993 (thus making it impossible to define a first MY with CHMSL and an adjacent last MY without CHMSL). On the other hand, Table 4-1 includes models that got CHMSL as part of a major redesign, as long as the redesign did not fundamentally change the model's size, function or market class (e.g., Dodge Ram Pickup, Ford full-sized van/wagon). Also included are models that merely changed their names, without any substantial redesign (e.g., Chevrolet K-Blazer to Tahoe).

As stated above, the range of model years on the data file must include the first two MY with CHMSL. For the many make-models that got CHMSL in 1994, that is only possible in the CY 1995 and 1996 data files. Those two calendar years are the mainstay of the analysis. Since the sample drawn from those two CY alone is somewhat limited, CY 1994 data have been added to the analysis. However, the CY 1994 data is restricted to the make-models that had at least 2 MY with CHMSL by 1994 - i.e., the models that got CHMSL in 1993 or earlier. For example, none

TABLE 4-1

LIGHT TRUCK MAKE-MODELS THAT GOT CHMSL IN 1991-94

MODEL YEARS INCLUDED IN THE EFFECTIVENESS ANALYSES

Mfr. Model Last 2 MY

without CHMSL

First 2 MY

with CHMSL

Chrysler Caravan/Voyager 1989-90 1991-92
Grand Caravan / Town&Country / Grand Voyager 1989-90 1991-92
Dakota 1992-93 1994-95
Ram pickup 1992-93 1994-95
Cherokee (excl. Grand Cherokee) 1992-93 1994-95
Wrangler 1992-93 1994-95
Ford F pickup 1990-91 1992-93
Full-sized van/wagon 1990-91 1992-93
Bronco (full-sized) 1990-91 1992-93
Ranger 1991-92 1993-94
Aerostar 1992-93 1994-95
GM C/K pickup 1992-93 1994-95
S/T pickup 1992-93 1994-95
Full-sized van / wagon 1992-93 1994-95
Astro / Safari van 1992-93 1994-95
Lumina APV / Silhouette / TransSport 1992-93 1994-95
K Blazer / Yukon / Tahoe 1992-93 1994-95
Suburban 1992-93 1994-95
S/T Blazer/Jimmy 1992-93 1994-95
Geo Tracker 1992-93 1994-95
Mazda Pickup 1992-93 1994-95
MPV 1992-93 1994-95
Nissan Pathfinder 1991-92 1993-94
Pickup 1992-93 1994-95
Toyota Previa 1991-92 1993-94
Landcruiser 1991-92 1993-94
Pickup (compact) 1992-93 1994-95
4-Runner 1992-93 1994-95

of the GM trucks are included in the CY 1994 data. As of December 1997 data were available for the following calendar years:

Florida 94-96 Maryland 94-96 Missouri 94-96
North Carolina 94 Pennsylvania 94-96 Utah 94-95

Crash involvements are tabulated by impact type: rear vs. other. Trucks with unknown impact locations and parked trucks are excluded from the analysis. The definitions of "rear impact" are:

Florida impact = 7,8,9
Maryland impact & veh_dam1 both 7-12, or one is 7-12 and one is blank
Missouri veh_dam1 = 7,8,9
North Carolina veh_dam1 = 10-13
Pennsylvania impact = 4,5,6,7,8
Utah impact & impact2 both 7-9, or one is 7-9 and one is blank

The vehicles deleted because they were "parked" are:

Florida veh_man1 = 8,9
Maryland veh_man1 = 10
Missouri veh_man1 = 13 or veh_man2 = 13 or veh_man3 = 13
North Carolina contfact = 7,10,11
Pennsylvania veh_man1 = 19
Utah veh_man1 = 11

These definitions are the same as in Chapter 2 (except in the case of North Carolina; data from that State were not used in the Chapter 2 analyses). The basic tabulation of crash involvements of vehicles listed in Table 4-1, by impact type and YT, the truck's model year relative to the transition MY to CHMSL for that make-model, combines the data for all six States in all the CY that these data were available:

YT REAR IMPACT?
NO YES
-2 (Two MY before CHMSL) X(-2) R(-2)
-1 (Last MY before CHMSL) X(-1) R(-1)
0 (First MY with CHMSL) X(0) R(0)
1 (Second Mywith CHMSL) X(1) R(1)

For example, since the Dodge Caravan got CHMSL in 1991, YT = -2 when MY = 1989; the MY 1989 cases go into the X(-2) and R(-2) cells, the MY 1990 cases into the X(-1) and R(-1) cells, the MY 1991 cases into the X(0) and R(0) cells, and the MY 1992 cases into the X(1) and R(1) cells. Since the Dodge Dakota got CHMSL in 1994, the MY 1992 cases go into the X(-2) and R(-2) cells, the MY 1993 cases into the X(-1) and R(-1) cells, etc.

The unadjusted effect of CHMSL is the actual, observed reduction in the log odds of a rear impact from the last MY before CHMSL (YT = -1) to the first MY with CHMSL (YT = 0):

E U = [ log R(-1) - log X(-1) ] - [ log R(0) - log X(0) ]

Generally speaking, E U will understate the actual benefit of CHMSL, because there is a "vehicle age effect" working in the opposite direction. That effect can be estimated by comparing trucks of the second year to the first year before CHMSL (when neither had CHMSL); or by comparing trucks of the first year to the second year with CHMSL (when both had CHMSL). The age effect is the average of those two statistics:

DEL BEFORE = [ log R(-2) - log X(-2) ] - [ log R(-1) - log X(-1) ]

DEL AFTER = [ log R(0) - log X(0) ] - [ log R(1) - log X(1) ]

DEL AVG = [ DEL BEFORE + DEL AFTER ]

The adjusted effect of CHMSL is the difference of the unadjusted effect and the age effect:

E A = E U - DEL AVG

Since DEL AVG is usually negative, E A will generally be more positive than E U. Finally, the percentage reduction of rear impacts attributed to CHMSL is calculated by converting the log odds ratios back to odds ratios:

E % = 1 - exp ( -1 * E A )

The statistical significance of the adjusted effectiveness estimate for CHMSL is tested by performing a CATMOD analysis of the preceding 4 x 2 tabulation of YT by impact location [32]. The dichotomous dependent variable is REARIMP (1 for rear impacts, 0 for other locations). The independent variables are YT (4 groups, but treated as a linear or "direct" variable with values -2, -1, 0 and 1) and CHM (= 0 when YT = -2 or -1; = 1 when YT = 0 or 1). Essentially, YT models the "vehicle age effect" and CHM models the true effect of the lamps after controlling for vehicle age. The effect of the lamps is significant when the coefficient for CHM has a statistically significant 2 value in the CATMOD analysis.

The preceding analysis could be biased because it aggregates data from all the States. For example, pre-CHMSL trucks could be overrepresented in a State where vehicles of all types have lower-than-average proportions of rear impact crashes. To adjust for biases like that, a CATMOD analysis is performed on the 6 x 4 x 2 table of State by YT by impact location. The dependent variable is again REARIMP. The independent variables are State (a nominal variable with six categories), YT (a linear variable), and CHM (a dichotomous variable). The effect of the lamps is significant, after controlling for State and the vehicle age effect, when the coefficient for CHM has a statistically significant 2 value in this analysis.


4.2 Overall effectiveness

The basic table of crash involvements by impact type and YT, the truck's model year relative to the transition MY to CHMSL, combining the data for six States in all CY, looks like this:

REAR IMPACT?
YT Frequency

Row Percent

NO YES
-2 (Two MY before CHMSL) 27871

74.21

9685

25.79

-1 (Last MY before CHMSL) 29882

73.90

10554

26.10

0 (First MY with CHMSL) 35116

74.40

12080

25.60

1 (Second MY with CHMSL) 33195

73.78

11796

26.22

Although the table includes over 44,000 rear impact cases, the sample is nevertheless much smaller than what was available for the passenger car analyses of Chapter 2 (where a table spanning MY 84-87 and aggregating the eight States would have included 800,000 rear impact cases over a ten CY period, or an average of 80,000 per calendar year). Correspondingly, we can expect the overall effectiveness estimate based on this table to be less precise than the passenger car estimate in any single calendar year, let alone an aggregation of calendar years, such as 1989-95.

It is obvious from an inspection of the row percentages that the estimate of CHMSL effectiveness will be positive. The percent of crash involvements that were rear impacts declined from 26.10 in the last MY before CHMSL to 25.60 in the first MY with CHMSL - despite a vehicle age trend in the opposite direction, as evidenced by an increase in the percentage of rear impacts from YT -2 to YT -1 and another increase from YT 0 to YT 1.

The unadjusted effect of CHMSL is:

E U = [ log 10554 - log 29882 ] - [ log 12080 - log 35116 ] = .0264

The statistics used for estimating the "vehicle age effect" are:

DEL BEFORE = [ log 9685 - log 27871 ] - [ log 10554 - log 29882 ] = -.0163

DEL AFTER = [ log 12080 - log 35116 ] - [ log 11796 - log 33195 ] = -.0325

DEL AVG = [(-.0163) + (-.0325) ] = -.0244

The adjusted effect of CHMSL is the difference of the unadjusted effect and the age effect:

E A = .0264 - (-.0244) = .0508

The percentage reduction of rear impacts attributed to CHMSL is calculated by converting E A back to an odds ratio:

E % = 1 - exp ( -1 * .0508 ) = 4.95 percent

This point estimate is very close to the long-term, 1989-95 effectiveness estimate for passenger car CHMSL, 4.3 percent (see Table 2-6). The similarity of the point estimates, however, may be deceptive, because the light truck estimate is subject to considerable sampling error (whereas the passenger car estimate had 95 percent confidence bounds ranging from 2.9 to 5.7 percent). When the statistical significance of the light truck estimate is tested by a CATMOD analysis of the preceding 4 x 2 table with YT and CHM as the independent variables, the coefficient for CHM has 2 = 4.51, slightly more than the 2 needed for statistical significance at the two-sided .05 level (3.89 - the critical value customarily used in NHTSA evaluations). Similarly, in the CATMOD analysis of the 6 x 4 x 2 table of State by YT by impact location, the coefficient for CHM has  2 = 4.62, again somewhat more than the 2 needed for statistical significance.

Confidence bounds for effectiveness may be obtained by noting that the standard deviation of the CHM coefficient in the first CATMOD analysis is .0245. In other words, the confidence bounds on E A are .0508 (1.96 x .0245): a range from .0028 to .0988. When these changes in the log odds ratio are converted to percentage reductions in rear impacts, the confidence bounds for effectiveness, E % range from 0.3 percent to 9.4 percent.

These levels of uncertainty preclude definitive inferences about the effectiveness of CHMSL for light trucks. There may be some support for temporarily accepting, as a "null" hypothesis, that CHMSL are about equally effective in trucks than cars. But it is also conceivable that they are less effective in trucks than in cars, or, for that matter, more effective than in cars.


4.3 Effectiveness by truck type and size

Given that all the light truck data combined were not really sufficient for a precise effectiveness estimate, limited returns can be expected if the data are subdivided by truck type or size. Nevertheless, it is worth doing at least a preliminary analysis to see if there are any conspicuous outliers (in either a positive or a negative direction).

Table 4-2 performs the basic effectiveness analysis separately for pickup trucks, sport utility vehicles (SUV's) and vans. For example, an inspection of the row percents in the top section of

TABLE 4-2: CHMSL EFFECTIVENESS BY TRUCK TYPE (CY 1994-96 data from six States)

PICKUP TRUCKS

REAR IMPACT? DELBEFORE = -.0455

Eu = .0223

DELAFETR = -.0394

DELAVG = -.0425

EA = .0648

YT Frequency

Row Percent

NO
YES
-2 (Two MY before CHMSL) 13978

75.25

4597

24.75

-1 (Last MY before CHMSL) 14878

74.39

5121

25.61

0 (First MY wtih CHMSL) 20897

74.82

7034

25.18

1 (Second MY with CHMSL) 18370

74.07

6432

25.93

observed effectiveness = E % = 6.27 percent

SPORT UTILITY VEHICLES

REAR IMPACT? DELBEFORE = -.0196

Eu = -.0006

DELAFETR = -.0352

DELAVG = -.0274

EA = .0268

YT Frequency

Row Percent

NO
YES
-2 (Two MY before CHMSL) 4165

74.07

1458

25.93

-1 (Last MY before CHMSL) 4737

73.69

1691

26.31

0 (First MY wtih CHMSL) 5994

73.68

2141

26.32

1 (Second MY with CHMSL) 5665

72.99

2096

27.01

observed effectiveness = E % = 2.64 percent

VANS

REAR IMPACT? DELBEFORE = .0235

Eu = .0314

DELAFETR = -.0101

DELAVG = .0067

EA = .0247

YT Frequency

Row Percent

NO YES
-2 (Two MY before CHMSL) 9728

72.83

3630

27.17

-1 (Last MY before CHMSL) 10267

73.29

3742

26.71

0 (First MY wtih CHMSL) 8225

73.90

2905

26.10

1 (Second MY with CHMSL) 9160

73.70

3268

26.30

observed effectiveness = E % = 2.44 percent

Table 4-2 makes it clear that the CHMSL effectiveness estimate will be positive. The percentage of crash involvements that were rear impacts dropped from 25.61 to 25.18 in the model year that CHMSL were introduced, in spite of the fact that it was on a rising trend the year before, and rose again the year after. The unadjusted log odds reduction for CHMSL is .0223 and the average age effect is -.0425. The point estimate is that CHMSL reduced rear impacts in pickup trucks by 6.27 percent. This reduction is not statistically significant: in the CATMOD analysis of the 4 x 2 table for pickup trucks, with YT and CHM as the independent variables, the coefficient for CHM has 2 = 3.64.

Similarly, the middle section of Table 4-2 analyzes the crash involvements of SUV's. The point estimate is that CHMSL reduced rear impacts in SUV's by 2.64 percent. This reduction is not statistically significant: in the CATMOD analysis, the coefficient for CHM has 2 = 0.23. The lower section indicates a 2.44 percent reduction of rear impacts for CHMSL in vans. This point estimate is also nonsignificant ( 2 = 0.22).

The differences between the three point estimates, 6.27, 2.64 and 2.44 percent, are very much in the "noise" range. In a CATMOD analysis of the 3 x 4 x 2 table of truck type by YT by impact location, with TRKTYP, YT, CHM as the independent variables, the coefficient for the interaction term TRKTYP * CHM had 2 = 0.67. That falls well short of the 2 needed for statistical significance. In other words, no significant differences in effectiveness by truck type were detected.

Table 4-3 compares the effect of CHMSL in compact and full-sized light trucks. The observed effectiveness of CHMSL in compact pickup trucks, SUV's and vans is 3.34 percent. This point estimate is not statistically significant ( 2 = 1.41). The effectiveness in full-sized light trucks, 7.28 percent, is likewise nonsignificant ( 2 = 3.32).

The differences between these two point estimates, 3.34 and 7.28 percent, is in the "noise" range. In a CATMOD analysis of the 2 x 4 x 2 table of truck size by YT by impact location, the coefficient for the interaction term SIZE * CHM had a nonsignificant 2 = 0.66.

While the data are still insufficient to distinguish the CHMSL effects in different types or sizes of trucks, one trend in the results is encouraging. All five point estimates (pickup, SUV, van, compact, full-sized) were positive. In four of the five analyses, even the unadjusted CHMSL effects were positive, showing a reduction in the proportion of rear impacts in the model year that CHMSL were first installed.

TABLE 4-3

CHMSL EFFECTIVENESS BY TRUCK SIZE

(CY 1994-96 data from six States)

COMPACT TRUCKS

REAR IMPACT? DELBEFORE = -0084

Eu = .0129

DELAFETR = -.0339

DELAVG = -.0211

EA = .0340

YT Frequency

Row Percent

NO YES
-2 (Two MY before CHMSL) 17719

73.63

6347

26.37

-1 (Last MY before CHMSL) 20245

73.46

7313

26.54

0 (First MY wtih CHMSL) 22500

73.71

8023

26.29

1 (Second MY with CHMSL) 20192

73.05

7448

26.95

observed effectiveness = E % = 3.34 percent

FULL-SIZED TRUCKS

REAR IMPACT? DELBEFORE = -.0226

Eu = .0448

DELAFETR = -.0391

DELAVG = -.0308

EA = .0756

YT Frequency

Row Percent

NO YES
-2 (Two MY before CHMSL) 10152

75.26

3338

24.74

-1 (Last MY before CHMSL) 9637

74.83

3241

25.17

0 (First MY wtih CHMSL) 12616

75.67

4057

24.33

1 (Second MY with CHMSL) 13003

74.94

4348

25.06

observed effectiveness = E % = 7.28 percent