Office of Operations Freight Management and Operations

2. Methodology for Measuring Regional Reorganization Effects

2.1 Development of Regional Subgroups

The development of a regional analysis focused on applying the analytical framework developed for the national analysis to geographically grouped sets of corridors. A primary task of the regional analysis has been the development of regional groups to analyze the reorganization effects related to highway freight improvements. The project team developed three proposed regional groupings: A five-region model, a three-region model based on FHWA's Federal Lands Program (FLP) regions, and a distributed three-region model.

These regional models were then applied to the corridors used in the Phase II national analysis and a determination was made of the number of corridors available within each model for each proposed region. Table 3 lists the corridors used in the national analysis conducted under Phase II.

Table 3. Corridors Used in National Analysis
Corridor Code
Atlanta-Jacksonville ATL-JAX
Atlanta-Knoxville ATL-KNX
Atlanta-Mobile ATL-MOB
Birmingham-Chattanooga BIR-CHA
Birmingham-Nashville BGH-NSH
Cleveland-Columbus CLE-COL
Columbus-Pittsburgh PIT-COL
Dallas-Houston DAL-HOU
Dayton-Detroit DAY-DET
Harrisburg-Philadelphia HAR-PHI
Indianapolis-Chicago IND-CHI
Indianapolis-Columbus OH IND-COL
Kansas City-St Louis KNC-STL
Knoxville-Dayton KNX-DAY
Louisville-Columbus COL-LOU
Louisville-Indianapolis IND-LOU
Mobile-New Orleans MOB-NOR
Nashville-Louisville NSH-LOU
Nashville-St Louis NSH-STL
New Orleans-Birmingham NOR-BIR
Richmond-Philadelphia RIC-PHI
San Antonio-Houston SAN-HOU
St Louis-Indianapolis STL-IND
Denver-Kansas City DEN-KAN
Denver-Salt Lake City DEN-SAL
San Francisco-Salt Lake City SFO-SAL
San Francisco-Los Angeles SFO-LAX
Portland-Seattle POR-SEA
Boston-New York City NYC-BOS
Harrisburg-New York City NYC-HAR

This original set of corridors represented a selection of freight-significant corridors of varying lengths, traffic volumes, and performance characteristics. However, the corridors are concentrated in the Midwest and lack good representation for some parts of the country. This problem was addressed by adding new corridors to the analysis, as discussed later.

Table 4 describes the division of the corridors originally used in the national analysis into the proposed five regions.

Table 4. Original Corridors Incorporated into Five Regions
Region Corridor Code
East Coast Harrisburg-Philadelphia HAR-PHI
Richmond-Philadelphia RIC-PHI
Boston-New York City NYC-BOS
Harrisburg-New York City NYC-HAR
Southeast Atlanta-Jacksonville ATL-JAX
Atlanta-Knoxville ATL-KNX
Atlanta-Mobile ATL-MOB
Birmingham-Chattanooga BIR-CHA
Birmingham-Nashville BGH-NSH
Mobile-New Orleans MOB-NOR
New Orleans-Birmingham NOR-BIR
Midwest Cleveland-Columbus CLE-COL
Columbus-Pittsburgh PIT-COL
Dayton-Detroit DAY-DET
Indianapolis-Chicago IND-CHI
Indianapolis-Columbus OH IND-COL
Kansas City-St Louis KNC-STL
Knoxville-Dayton KNX-DAY
Louisville-Columbus COL-LOU
Louisville-Indianapolis IND-LOU
Nashville-Louisville NSH-LOU
Nashville-St Louis NSH-STL
St Louis-Indianapolis STL-IND
Southwest Dallas-Houston DAL-HOU
San Antonio-Houston SAN-HOU
West Coast Denver-Kansas City DEN-KAN
Denver-Salt Lake City DEN-SAL
San Francisco-Salt Lake City SFO-SAL
San Francisco-Los Angeles SFO-LAX
Portland-Seattle POR-SEA

The five-region model allows for an easily understandable allocation of corridors to regions with which most practitioners would recognize and identify. However, it creates problems of under-representation, particularly in the Southwest and East. Using a five-region model would require significant investment in collecting data for a sufficient number of corridors to cover each region.

Table 5 describes the division of the corridors originally used in the national analysis into the proposed three FHWA FLP regions.

Table 5. Original Corridors Incorporated into FHWA's Federal Lands Program Regions
Region Corridor Code
Eastern Region Atlanta-Jacksonville ATL-JAX
Atlanta-Knoxville ATL-KNX
Atlanta-Mobile ATL-MOB
Birmingham-Chattanooga BIR-CHA
Birmingham-Nashville BGH-NSH
Cleveland-Columbus CLE-COL
Columbus-Pittsburgh PIT-COL
Dayton-Detroit DAY-DET
Harrisburg-Philadelphia HAR-PHI
Indianapolis-Chicago IND-CHI
Indianapolis-Columbus OH IND-COL
Kansas City-St Louis KNC-STL
Knoxville-Dayton KNX-DAY
Louisville-Columbus COL-LOU
Louisville-Indianapolis IND-LOU
Mobile-New Orleans MOB-NOR
Nashville-Louisville NSH-LOU
Nashville-St Louis NSH-STL
New Orleans-Birmingham NOR-BIR
Richmond-Philadelphia RIC-PHI
Boston-New York City NYC-BOS
Harrisburg-New York City NYC-HAR
St Louis-Indianapolis STL-IND
Central Region Dallas-Houston DAL-HOU
Denver-Kansas City DEN-KAN
Denver-Salt Lake City DEN-SAL
San Francisco-Salt Lake City SFO-SAL
San Francisco-Los Angeles SFO-LAX
San Antonio-Houston SAN-HOU

Using FHWA's FLP regions as the model for sub-national division creates significant concentration in the East. In addition, the West region becomes extremely small, with only one corridor allocated.

Table 6 describes the division of the corridors originally used in the national analysis into the proposed three distributed regions.

Table 6. Original Corridors Incorporated into Three Regions
Region Corridor Code
East Coast Atlanta-Jacksonville ATL-JAX
Atlanta-Knoxville ATL-KNX
Atlanta-Mobile ATL-MOB
Birmingham-Chattanooga BIR-CHA
Birmingham-Nashville BGH-NSH
Harrisburg-Philadelphia HAR-PHI
Mobile-New Orleans MOB-NOR
New Orleans-Birmingham NOR-BIR
Richmond-Philadelphia RIC-PHI
Boston-New York City NYC-BOS
Harrisburg-New York City NYC-HAR
Midwest Cleveland-Columbus CLE-COL
Columbus-Pittsburgh PIT-COL
Dayton-Detroit DAY-DET
Indianapolis-Chicago IND-CHI
Indianapolis-Columbus OH IND-COL
Kansas City-St Louis KNC-STL
Knoxville-Dayton KNX-DAY
Louisville-Columbus COL-LOU
Louisville-Indianapolis IND-LOU
Nashville-Louisville NSH-LOU
Nashville-St Louis NSH-STL
St Louis-Indianapolis STL-IND
West Coast Dallas-Houston DAL-HOU
San Antonio-Houston SAN-HOU
Denver-Kansas City DEN-KAN
Denver-Salt Lake City DEN-SAL
San Francisco-Salt Lake City SFO-SAL
San Francisco-Los Angeles SFO-LAX

The distributed three-region model reduces the potential variability between regions that would be in a five-region model. It creates regional allocations with which practitioners can identify, but the three-region model reduces the burden of additional data collection to achievable proportions.

The three proposed regional division models are:

Five-Region Model
Regions Corridors
Midwest 12
East Coast 4
Southeast 7
Southwest 2
West Coast 5

Three-Region FLP Model
Regions Corridors
Central 6
Eastern 23
Western 1

Three-Region Model
Regions Corridors
Midwest 12
East Coast 11
West Coast 7

The three-region FLP model was determined as untenable due to the heavy weighting of states included in the Eastern region and the lack of good mapping to freight-significant corridors with the southern half of the West Coast included in the Central region and most of the Midwest included in the Eastern region. As can be seen from the table above, the FLP approach allowed only one of the original corridors to be included in the third (Western) region.

2.2 Testing of Regional Subgroups

The project team tested a selection of the subgroups for performance using the three final equations developed under Phase II: The pooled regression equation, the regression with fixed effects, and the general least squares regression with fixed effects. The objectives of the testing were:

  • To estimate the likely minimum number of corridors or observations required per region in order to derive significant results;
  • To determine the data collection required in order to develop a robust regional dataset; and
  • To use this information to determine which regional definition to use in the final analysis.

In order to estimate the likely minimum number of corridors or observations required, each of the three equations developed for the Phase II national analysis were re-run using the midwestern corridors from the five-region model and the East Coast corridors from the three-region model. In addition, the 23-corridor FLP Eastern region was included in order to improve the estimate of the required minimum number of corridors or observations required. However, as discussed earlier, the FLP model was determined unsuitable for use as a final regional grouping.

Equation 1: Pooled Regression

Equation 1 uses a semi-log functional form with data on freight demand, congestion related delays, and economic variables. In the following model, demand for daily truck traffic is specified as a function of per capita income, GDP growth rate, LTL rates, and delay. Table 7 provides the results from the original national analysis.

Estimating Equation:the following expression should read: LOG of parenthesis, Trucks/day, subscript r, subscript t, close parenthesis, equals beta subscript zero, plus beta, subscript one, multiplied by delay, subscript r, subscript t, plus beta subscript two, multiplied by GDP Growth, subscript r, subscript t, plus beta subscript three, multiplied by Real per Capita Income, subscript r, subscript t, plus LTL Rate subscript r, subscript tLOG(Trucks/dayr,t) = β0 + β1 * Delayr,t + β2 * GDP Growthr,t + β3 Real per Capita Incomer,t + β4 * LTL Rater,t

Table 7. Regression Results for Freight Demand (Pooled Regression)
Variable Coefficient Std. Error t-Statistic Probability of Non-Significance
Constant 8.162777 0.061204 133.3699 0.0000
Delay -0.001834 0.000418 -4.388016 0.0000
GDP Growth 0.067249 0.010604 6.342005 0.0000
Real per Capita Income 6.16E-05 4.67E-06 13.19443 0.0000
Real LTL Rates -0.004517 0.000296 -15.26863 0.0000
R-squared 0.323195 Mean dependent variable 8.805702
Adjusted R-squared 0.308871 S.D. dependent variable 0.389423
S.E. of regression 0.323743 Sum squared residual 19.80905
Durbin-Watson stat 0.142865    

Method: GLS (Cross-Section Weights)
Total panel (unbalanced) observations: 194

Where:

  • Trucks/day = Average number of trucks/day for the corridor
  • Delay = Average delay per mile
  • GDP Growth = Growth rate for the Gross Domestic Product
  • Real Per Capita Income = Average Per Capita income for all counties along the corridors
  • LTL Rates/trip = Less-than-truckload rates for 1,000 lb. shipment

For the national analysis, this specification produced coefficients that are of the expected sign. Further estimation also suggested that the coefficient of the delay variable may vary across corridors of various lengths and be higher in absolute terms for long routes (over 400 miles) than for short routes (up to 200 miles) and medium routes (200 to 400 miles). However, the Durbin-Watson (DW) statistic in the above specification was low, at around 0.14. A low DW statistic indicates a potential problem of autocorrelation within and across the individual cross-sections (or other forms of misspecification). This arises from complications with panel data, in particular:

  • Errors are not independent between time periods and
  • Errors are correlated between corridors.

As such, re-estimating Equation 1 with fewer observations was not expected to produce strong results. Table 8 describes the results of Equation 1 for each of the three regional sub-groups tested.

Table 8. Results for Equation 1 for Three Regional Sub-groups (FHWA Eastern Region)
Three-Region FLP Model
FHWA Eastern Region
23 corridors with
2 no-data corridors
21 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 21
Total pool (unbalanced) observations: 151
Variable Coef Std. Error t-Stat Prob.
Constant 8.245 0.127 64.774 0.000
Delay 0.001 0.002 0.243 0.809
GDP Growth 0.040 0.026 1.524 0.130
Real Per Capita Income 0.000 0.000 5.109 0.000
Real LTL Rates -0.001 0.000 -2.455 0.015
R-squared 0.999 Mean dependent var 13.772
Adjusted R-squared 0.999 S.D. dependent var 7.762
S.E. of regression 0.234 Sum squared resid 7.966
Durbin-Watson stat 0.318    

Table 8. Results for Equation 1 for Three Regional Sub-groups (East combined)
Three-Region Model
East (combined)
11 corridors with
2 no-data corridors
9 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 9
Total pool (unbalanced) observations: 65
Variable Coef Std. Error t-Stat Prob.
Constant 7.873 0.166 47.487 0.000
Delay -0.006 0.001 -3.874 0.000
GDP Growth 0.039 0.042 0.919 0.362
Real Per Capita Income 0.000 0.000 6.268 0.000
Real LTL Rates -0.002 0.001 -2.147 0.036
R-squared 0.994 Mean dependent var 10.448
Adjusted R-squared 0.994 S.D. dependent var 2.972
S.E. of regression 0.232 Sum squared resid 3.241
Durbin-Watson stat 0.265    

Table 8. Results for Equation 1 for Three Regional Sub-groups (Midwest)
Three-Region Model
Midwest
12 corridors with
0 no-data corridors
12 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 12
Total pool (unbalanced) observations: 86
Variable Coef Std. Error t-Stat Prob.
Constant 8.604 0.149 57.930 0.000
Delay 0.004 0.003 1.568 0.121
GDP Growth 0.037 0.030 1.258 0.212
Real Per Capita Income 0.000 0.000 1.920 0.058
Real LTL Rates -0.001 0.001 -1.028 0.037
R-squared 0.999 Mean dependent var 13.183
Adjusted R-squared 0.999 S.D. dependent var 5.737
S.E. of regression 0.192 Sum squared resid 2.990
Durbin-Watson stat 0.390    

As expected, each of the regional subgroups performed poorly in terms of the DW statistic, ranging from.0.29 to 0.36. None of the equations achieved significant results for all included variables. The 65-observation Eastern region achieved significant and correctly signed results for the Delay variable, which is the key measure of highway performance included in the equation.

Equation 2: Regression with Fixed Effects

To mitigate the problems associated with Equation 1, the model was re-estimated using the fixed effects method during the Phase II national analysis. The estimating equation was reformulated as

Estimating Equation:the following expression should read: Log parenthesis, Trucks/day, subscript r, subscript t, closed parenthesis, equals beta subscript r, plus beta subscript one, multiplied by Delay, subscript r, subscript t, plus beta subscript 2, multiplied by GDP Growth, subscript r, subscript tLOG(Trucks/dayr,t) = βr + β1 * Delayr,t + β2 * GDP Growthr,t

Wherethe following expression should read: beta subscript r, parenthesis r equals one to twenty-eight, closed parenthesisβr ( r = 1…28) are corridor-specific constant, or fixed effects.

The results of the estimation developed for the national analysis are provided in Table 9.

Table 9. Regression Results for Freight Demand (Fixed Effects)
Variable Coefficient Std. Error t-Statistic Probability of Non-Significance
Delay -0.002575 0.000496 -5.186827 0.0000
GDP Growth 0.073356 0.007632 9.611514 0.0000
Fixed Effects        
Atlanta-Jacksonville 8.716456      
Atlanta-Knoxville 8.918446      
Atlanta-Mobile 8.362351      
Birmingham-Nashville 8.353916      
Cleveland-Columbus 8.605421      
Columbus-Pittsburgh 8.535460      
Dallas-Houston 8.623438      
Dayton-Detroit 8.592172      
Harrisburg-Philadelphia 8.526888      
Indianapolis-Chicago 8.655297      
Indianapolis-Columbus 8.821301      
Kansas City-St. Louis 8.610851      
Knoxville-Dayton 8.807753      
Louisville-Columbus 8.708366      
Louisville-Indianapolis 8.679238      
Mobile-New Orleans 8.325837      
Nashville-Louisville 9.049711      
Nashville-St. Louis 8.086613      
Richmond-Philadelphia 9.035896      
San Antonio-Houston 8.423007      
St. Louis-Indianapolis 8.664829      
Denver-Kansas City 7.441348      
Denver-Salt Lake City 8.156259      
San Francisco-Salt Lake City 7.459972      
San Francisco-Los Angeles 8.343404      
Portland-Seattle 8.794208      
Boston-New York City 8.639654      
Harrisburg-New York City 8.727643      
R-squared 0.924207 Mean dependent variable   8.805702
Adjusted R-squared 0.910804 S.D. dependent variable   0.389423
S.E. of regression 0.116304 Sum squared residual   2.218348
Durbin-Watson stat 1.117309      

Method: GLS (Cross-Section Weights)
Total panel (unbalanced) observations: 194

For the national analysis, the high R-squared value, 0.92 indicated that the variables considered explain changes in freight demand very closely. Delay and GDP growth variables have the correct signs, and they are statistically significant. The DW statistic, 1.11, is significantly higher than that reported in the first model.

Equation 2 was also re-run using the three regional subgroups. Table 10 describes the results for each.

Table 10. Results for Equation 2 for Three Regional Sub-groups (FHWA Eastern Region)
Three-Region FLP Model
FHWA Eastern Region
23 corridors with
2 no-data corridors
21 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 21
Total pool (unbalanced) observations: 151
Variable Coef Std. Error t-Stat Prob.
Constant 8.682 0.045 192.020 0.000
Delay -0.005 0.001 -6.088 0.000
GDP Growth 0.065 0.011 5.685 0.000
Fixed Effects:
ATL-JAX 0.071      
ATL-KNX 0.270      
ATL-MOB -0.289      
BGH-NSH -0.297      
CLE-COL -0.037      
PIT-COL -0.115      
DAY-DET -0.055      
HAR-PHI -0.104      
IND-CHI 0.007      
IND-COL 0.175      
KNC-STL -0.013      
KNX-DAY 0.195      
COL-LOU 0.063      
IND-LOU 0.030      
MOB-NOR -0.320      
NSH-LOU 0.425      
NSH-STL -0.564      
RIC-PHI 0.397      
STL-IND 0.015      
NYC-BOS 0.026      
NYC-HAR 0.080      
R-squared 0.999 Mean dependent var 12.179
Adjusted R-squared 0.999 S.D. dependent var 4.820
S.E. of regression 0.110 Sum squared resid 1.539
Durbin-Watson stat 1.179    

Table 10. Results for Equation 2 for Three Regional Sub-groups (East combined)
Three-Region Model
East (combined)
11 corridors with
2 no-data corridors
9 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 9
Total pool (unbalanced) observations: 65
Variable Coef Std. Error t-Stat Prob.
Constant 8.626 0.058 148.596 0.000
Delay -0.005 0.001 -5.887 0.000
GDP Growth 0.076 0.015 4.958 0.000
Fixed Effects:
ATL-JAX 0.086      
ATL-KNX 0.285      
ATL-MOB -0.274      
BGH-NSH -0.282      
HAR-PHI -0.093      
MOB-NOR -0.305      
RIC-PHI 0.412      
NYC-BOS 0.041      
NYC-HAR 0.091      
R-squared 0.999 Mean dependent var 13.535
Adjusted R-squared 0.999 S.D. dependent var 6.478
S.E. of regression 0.137 Sum squared resid 1.009
Durbin-Watson stat 1.176    

Table 10. Results for Equation 2 for Three Regional Sub-groups (Midwest)
Three-Region FLP Model
Midwest
12 corridors with
0 no-data corridors
12 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 12
Total pool (unbalanced) observations: 86
Variable Coef Std. Error t-Stat Prob.
Constant 8.727 0.063 137.923 0.000
Delay -0.005 0.002 -2.357 0.021
GDP Growth 0.056 0.016 3.534 0.001
Fixed Effects:
CLE-COL -0.047      
PIT-COL -0.125      
DAY-DET -0.064      
IND-CHI -0.002      
IND-COL 0.166      
KNC-STL -0.023      
KNX-DAY 0.177      
COL-LOU 0.052      
IND-LOU 0.022      
NSH-LOU 0.410      
NSH-STL -0.574      
STL-IND 0.007      
R-squared 0.999 Mean dependent var 10.104
Adjusted R-squared 0.999 S.D. dependent var 3.201
S.E. of regression 0.085 Sum squared resid 0.518
Durbin-Watson stat 1.149    

As with the national analysis, Equation 2 performed well for each of the regional groupings, all with R-squared statistics above 0.99.

Equation 3: GLS Regression with Fixed Effects

The model was also estimated with V/C ratio and "fixed effects." The estimating equation was reformulated as:

Estimating Equation:the following expression should read: Log, parenthesis, Trucks/day, subscript r, subscript t, closed parenthesis, equals, beta subscript r, plus beta subscript one , multiplied by VC Ratio subscript r, subscript t, plus beta subscript two, multiplied by GDP Growth, subscript r, subscript t, plus beta subscript three, multiplied by real per capita income, subscript r, subscript t, plus beta subscript four, multiplied by Real LTL Rates, subscript r, subscript t.  Where beta subscript r, parenthesis, r equals one to twenty eight, closed parenthesis, are corridor specific constant, or fixed effectsLOG(Trucks/dayr,t) = βr + β1 * VC Ratior,t + β2 * GDP Growthr,t β3 * Real per Capita Incomer,t + β4 * Real LTL Rater,t

Wherethe following expression should read: beta subscript r, parenthesis r equals one to twenty-eight, closed parenthesisβr ( r = 1…28) are corridor-specific constant, or fixed effects.

Table 11 describes the results of the original national analysis using Equation 3.

In the original national analysis, this model had a high R-squared value (0.954), indicating that the selected variables are highly correlated with changes in demand for transportation. The VC Ratio, GDP Growth, and Real Per Capita Income variables have the correct signs. The VC Ratio and Real LTL Rate variables are statistically significant. The DW statistic is higher than earlier estimations.

Table 11. GLS Regression Results for Freight Demand with Cross Section Weights (Fixed Effects)
Variable Coefficient Std. Error t-Statistic Probability of Non-Significance
VC Ratio -0.145737 0.024236 -6.013214 0.0000
GDP Growth 0.005716 0.004550 1.256128 0.2109
Real Per Capita Income 8.11E-06 4.04E-06 2.006642 0.0465
Real LTL Rates 0.003148 0.000230 13.65898 0.0000
Fixed Effects
Atlanta-Jacksonville -0.145737      
Atlanta-Knoxville 0.005716      
Atlanta-Mobile 8.11E-06      
Birmingham-Nashville 0.003148      
Cleveland-Columbus        
Columbus-Pittsburgh 8.343270      
Dallas-Houston 8.671927      
Dayton-Detroit 7.985723      
Harrisburg-Philadelphia 7.827820      
Indianapolis-Chicago 8.331450      
Indianapolis-Columbus 8.187617      
Kansas City-St. Louis 8.232353      
Knoxville-Dayton 8.249458      
Louisville-Columbus 8.055459      
Louisville-Indianapolis 8.262573      
Mobile-New Orleans 8.492228      
Nashville-Louisville 8.111981      
Nashville-St. Louis 8.373134      
Richmond-Philadelphia 8.328012      
San Antonio-Houston 8.378806      
St. Louis-Indianapolis 8.150193      
Denver-Kansas City 8.808472      
Denver-Salt Lake City 7.632464      
San Francisco-Salt Lake City 8.470470      
San Francisco-Los Angeles 8.100525      
Portland-Seattle 8.268212      
Boston-New York City 6.563344      
Harrisburg-New York City 7.438120      
R-squared 0.954116 Mean dependent var 8.805702
Adjusted R-squared 0.945336 S.D. dependent variable 0.389423
S.E. of regression 0.091048 Sum squared residual 1.342952
Durbin-Watson stat 1.315943    

This model performed well using the entire set of corridors, but sub-division into smaller groupings with fewer observations indicated that the nine- and 12-corridor groupings did not have a sufficient number of observations to generate statistically significant results.

The results of the three regional analyses are described in Table 12.

Table 12. Results for Equation 3 for Three Regional Sub-groups (FHWA Eastern Region)
Three-Region FLP Model
FHWA Eastern Region
23 corridors with
2 no-data corridors
21 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 21
Total pool (unbalanced) observations: 151
Variable Coef Std. Error t-Stat Prob.
Constant 8.223 0.104 78.752 0.000
VC_Ratio -0.162 0.037 -4.361 0.000
GDP_Growth 0.003 0.009 0.383 0.703
Real Per Capita Income 0.000 0.000 0.867 0.388
Real LTL Rates 0.003 0.001 4.392 0.000
Fixed Effects:
ATL-JAX 0.095      
ATL-KNX 0.426      
ATL-MOB -0.264      
BGH-NSH -0.424      
CLE-COL 0.088      
PIT-COL -0.060      
DAY-DET 0.001      
HAR-PHI -0.193      
IND-CHI 0.008      
IND-COL 0.246      
KNC-STL -0.137      
KNX-DAY 0.129      
COL-LOU 0.081      
IND-LOU 0.131      
MOB-NOR -0.088      
NSH-LOU 0.567      
NSH-STL -0.620      
RIC-PHI 0.217      
STL-IND 0.018      
NYC-BOS -0.184      
NYC-HAR -0.201      
R-squared 0.999 Mean dependent var 15.836
Adjusted R-squared 0.999 S.D. dependent var 18.020
S.E. of regression 0.090 Sum squared resid 1.020
Durbin-Watson stat 1.1896    

Table 12. Results for Equation 3 for Three Regional Sub-groups (East combined)
Three-Region Model
East (combined)
11 corridors with
2 no-data corridors
9 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 9
Total pool (unbalanced) observations: 65
Variable Coef Std. Error t-Stat Prob.
Constant 7.796 0.210 37.165 0.000
VC_Ratio -0.088 0.072 -1.223 0.227
GDP_Growth 0.000 0.016 -0.027 0.979
Real Per Capita Income 0.000 0.000 1.391 0.170
Real LTL Rates 0.002 0.002 1.219 0.228
Fixed Effects:
ATL-JAX 0.179      
ATL-KNX 0.464      
ATL-MOB -0.184      
BGH-NSH -0.251      
HAR-PHI -0.180      
MOB-NOR -0.001      
RIC-PHI 0.248      
NYC-BOS -0.214      
NYC-HAR -0.206      
R-squared 0.999 Mean dependent var 15.440
Adjusted R-squared 0.999 S.D. dependent var 8.714
S.E. of regression 0.112 Sum squared resid 0.650
Durbin-Watson stat 1.221    

Table 12. Results for Equation 3 for Three Regional Sub-groups (Midwest)
Five-Region Model
Midwest
12 corridors with
0 no-data corridors
12 corridors for analysis
Dependent Variable: LOG(AADT-Trucks)
Method: Pooled EGLS (Cross-section weights)
Included observations: 8
Cross-sections included: 12
Total pool (unbalanced) observations: 86
Variable Coef Std. Error t-Stat Prob.
Constant 8.487 0.132 64.309 0.000
VC_Ratio 0.071 0.122 0.581 0.563
GDP_Growth 0.014 0.016 0.841 0.403
Real Per Capita Income 0.000 0.000 -1.430 0.157
Real LTL Rates 0.005 0.001 4.389 0.000
Fixed Effects:
CLE-COL 0.039      
PIT-COL -0.093      
DAY-DET -0.003      
IND-CHI 0.118      
IND-COL 0.203      
KNC-STL -0.289      
KNX-DAY -0.062      
COL-LOU -0.005      
IND-LOU 0.156      
NSH-LOU 0.522      
NSH-STL -0.660      
STL-IND -0.006      
R-squared 0.999 Mean dependent var 14.241
Adjusted R-squared 0.999 S.D. dependent var 19.442
S.E. of regression 0.070 Sum squared resid 0.341
Durbin-Watson stat 1.311    

2.3 Selection of the Regional Groupings

The results of the analysis of the proposed regional groupings indicate that results generating reliability similar to that in the Phase II national analysis require groupings with an estimated minimum of 15–20 corridors. It can not be determined in advance of data collection and re-analysis whether any particular grouping will generate usable results. However, the performance of the test groupings against the three Phase II equations did not demonstrate any particular regional weakness.

Given the need for a regional grouping where every defined region needs at least 15-20 corridors, the project group and FHWA selected the distributed three-region grouping and began the task of collecting data for additional corridors to generate sufficient observations for each region. Having experienced issues related to incomplete data during the Phase II analysis, the project team selected sufficient additional corridors for each region in case a few corridors could not be used. In addition, an attempt was made to include corridors that covered a range of route types, from heavily urban to more rural and from a variety of areas within each region.

The following corridors were selected for addition to the dataset:

  • Omaha-Chicago
  • Barstow-Bakersfield
  • San Francisco-Portland
  • Billings-Sioux Falls
  • Chicago-Cleveland
  • Amarillo-Oklahoma City
  • Detroit-Pittsburgh
  • Galveston-Dallas
  • San Diego-Los Angeles
  • Miami-Atlanta
  • Seattle-Billings
  • Miami-Richmond
  • Seattle-Sioux Falls
  • New York City-Cleveland
  • Portland-Salt Lake City
  • Philadelphia-New York City
  • Los Angeles-Tucson
  • Dallas-El Paso
  • Tucson-San Antonio
  • Memphis-Dallas
  • Barstow-Salt Lake City
  • Memphis-Oklahoma City
  • Barstow-Amarillo
  • Stouts-Oklahoma City
  • Nogales-Tucson
  • Seattle-Blaine
  • San Antonio-Dallas
  • Knoxville-Harrisburg
  • Laredo-San Antonio

Having added these corridors, a final regional division was made. Table 13 describes the corridors selected for inclusion in each region.

Table 13. Proposed Final Corridors for Three-Region Analysis
East Region – 18 corridors Central Region – 18 corridors West Region – 23 corridors
Atlanta-Jacksonville ATL-JAX Amarillo-Oklahoma City AMA-OKL Barstow-Amarillo BAR-AMA
Atlanta-Knoxville ATL-KNX Billings-Sioux Falls BIL-SIO Barstow-Bakersfield BAR-BAK
Atlanta-Mobile ATL-MOB Chicago-Cleveland CHI-CLE Barstow-Salt Lake City BAR-SAL
Birmingham-Nashville BGH-NSH Cleveland-Columbus CLE-COL Dallas-El Paso DAL-ELP
Birmingham-Chattanooga BIR-CHA Dayton-Detroit DAY-DET Dallas-Houston DAL-HOU
Detroit-Pittsburgh DET-PIT Indianapolis-Chicago IND-CHI Denver-Kansas City DEN-KAN
Harrisburg-Philadelphia HAR-PHI Indianapolis-Columbus OH IND-COL Denver-Salt Lake City DEN-SAL
Knoxville-Harrisburg KNX-HAR Kansas City-St Louis KNC-STL Galveston-Dallas GAL-DAL
Miami-Atlanta MIA-ATL Knoxville-Dayton KNX-DAY Laredo-San Antonio LAR-SAN
Miami-Richmond MIA-RIC Louisville-Columbus COL-LOU Los Angeles-Tucson LAX-TUC
Mobile-New Orleans MOB-NOR Louisville-Indianapolis IND-LOU Nogales-Tucson NOG-TUC
New Orleans-Birmingham NOR-BIR Memphis-Dallas MEM-DAL Portland-Salt Lake City POR-SAL
Boston-New York City NYC-BOS Memphis-Oklahoma City MEM-OKL Portland-Seattle POR-SEA
New York City-Cleveland NYC-CLE Nashville-Louisville NSH-LOU San Antonio-Dallas SAN-DAL
Harrisburg-New York City NYC-HAR Nashville-St Louis NSH-STL San Diego-Los Angeles SDG-LAX
Philadelphia-New York City PHI-NYC Omaha-Chicago OMA-CHI San Francisco-Los Angeles SFO-LAX
Columbus-Pittsburgh PIT-COL St Louis-Oklahoma City STL-OKL San Francisco-Portland SFO-POR
Richmond-Philadelphia RIC-PHI St Louis-Indianapolis STL-IND San Francisco-Salt Lake City SFO-SAL
        San Antonio-Houston SAN-HOU
        Seattle-Billings SEA-BIL
        Seattle-Blaine SEA-BLA
        Seattle-Sioux Falls SEA-SIO
        Tucson-San Antonio TUC-SAN

The final regional categorization utilizes a three-region approach. Data availability and quality are the main reasons that a more detailed regional disaggregation has not been pursued.

However, the additive benefit estimation calculator will achieve significantly more specificity to local conditions then the example illustrated herein. This will be achieved though the use of corridor-specific AADTT in the additive benefit calculation and corridor-specific delay characteristics in the transformation of the elasticity estimate.

Additive benefit estimates are provided in this report as an illustration of what the additional reorganization benefit would be for the average corridor in each region. This is not the additional benefit proposed as the addition to every corridor's total benefit estimation. The tool to be developed in a subsequent task will utilize the regional elasticities calculated for this report and corridor-specific data, such as daily truck traffic, current delay, delay reduction, localized cost data, and other locally specific data to develop a corridor-specific additive freight reorganization benefit factor to be applied to calculated freight related benefits.

2.4 Rate, Flow, Commodity, and Performance Data

In addition to adding corridors, the existing dataset was also expanded by adding years of observations. Additional HPMS data collection has occurred since the development of the Phase II dataset. The project team was able to add three years of observations for most corridors, significantly expanding the total number of observations.

Data on heavy-duty vehicle traffic volumes, freight rates, and commodity flows were collected from several different sources. The data used are described below. Note that data on 30 corridors were available from Phase II of this study. Additional years of data were collected for these corridors. Data for 29 new corridors were also collected to enhance the size and regional coverage of our database. Of those 29 corridors, 25 were added to the final sample.

2.4.1 Rate Data

Freight rates for each corridor were obtained from SMC3's Czarlite[5] database. This database serves as the benchmark for thousands of LTL contracts. Rates were obtained for each corridor using an origin and destination zip code. The rates were defined by using a 1,000 pound, class 70 shipment type to estimate an average rate for each corridor for the years 1993-2004.

2.4.2 Commodity Flow Data

The commodity flows in each corridor were characterized using FAF data from FHWA. Both the original FAF and a newly released version (FAF2) were used since the geographic detail and years available were somewhat different between these databases. The FAF2 database has information on commodity flows between major geographic regions, including some metropolitan area city pairs for the year 2002. Data on the tonnage, value, and commodity types being moved in both directions in the study corridors were developed. Since the geographic regions available in FAF2 did not map exactly to the corridor origin and destination cities used in the analysis, the regions most closely matching this study's corridors were utilized to obtain an approximate picture of existing commodity flows. Commodity flows were developed for 58 distinct corridors and captured information on freight moving in both directions along the corridor.

The original FAF database contains county-to-county movements of freight by commodity type for 1998 and forecast years. Commodity flows for each corridor were developed from this database as well. The purpose of the commodity flow analysis was to understand the differences that exist between the corridors and to develop an understanding of how these differences might affect the results of the modeling.

2.4.3 HPMS

The HPMS Sample database was used to develop information on the average V/C ratios for the corridors being studied. The HPMS Sample database was obtained for the years 1993-2003. Each year of sample data contains approximately 110,000 records. Each record represents one segment and includes data on segment ID, state, county, route number, average annual daily truck traffic, peak and off-peak commercial vehicle percentages, V/C ratio, as well as many other items.

FHWA's list of "freight significant corridors" was used to identify many of the corridors used in this study. Additional corridors were added based on expert judgment or the need to increase regional coverage or include corridor types that were not represented. For instance, a number of major international trade lanes were added to increase the coverage of international commodity flows. A number of rural low-volume freight corridors were added to enhance coverage of this corridor type.

For each corridor identified, the relevant highway routes that would be used between a given city pair were identified. The set of HPMS segments representing these highway routes were then determined. One problem encountered was that, in some cases, routes designated in HPMS would change across years or between states. For instance, in one year a route would be identified as 40, and in another year a route would be identified as I40. There were also variations in how routes were designated between states.

In order to work around this, the data for each state were manually inspected to determine the route designations used for each corridor. Based on this analysis, a list of segments that characterized each corridor was developed.

HPMS data for each study corridor defined were aggregated to develop summary information describing average truck volumes and V/C ratios for each corridor. Truck volumes were obtained by multiplying the off-peak truck percentage by the AADTT for each segment. Corridor averages for truck volumes and V/C ratios were obtained averaging all the segments for each corridor, weighted by the segment length.

A number of problems were encountered. Many segments did not have data for all the years in the analysis. In addition, some segments had zero values in the off-peak truck percentage field. In order to compare data across years for each corridor, it was necessary to eliminate segments from the analysis that did not have data across all years or were missing data. For some corridors, data were unavailable for particular years. In order to address this, the years of data used in each corridor was adjusted to include the most segments, while at the same time making available the largest number of years of data available for analysis.

For example, if most segments were missing for a particular year of data, then that year would be omitted. In some cases, there were duplicate records for some segment IDs. These were dropped from the analysis. A number of corridors that were initially examined did not have enough data available for them. An additional 29 corridors to those used in the national analysis had enough information to develop corridor averages. Table 14 shows the total number of segments available and the number of segments used to develop the data for each of the 29 corridors. Also shown are the years of data that were available for each corridor.

Post data cleaning resulted in 55 corridors, representing both the corridors in the Phase II national analysis and newly added corridors, being usable for our sample.

Table 14. Number of Segments and Years Used to Characterize Each Additional Corridor
Corridor Name Total Segments Segments Available for Analysis Years Covered
Dallas-El Paso 147 141 1997-2000
Memphis-Dallas 148 77 1993-1994 & 1996-2004
Memphis-Oklahoma City 230 70 1993-1994 & 1996-2004
Amarillo-Oklahoma City 155 121 1999-2001
Barstow-Amarillo 371 324 1997-2000
Barstow-Bakersfield 55 48 1994-2000
Barstow-Salt Lake City 234 180 1997-2004
Billings-Sioux Falls 186 140 1993-2004
Denver-Kansas City 111 54 1993-2003
Galveston-Dallas 53 50 1997-2000
Knoxville-Harrisburg 272 224 1996-2004
Laredo-San Antonio 36 34 1997-2000
Los Angeles-Tucson 259 244 1997-2000
Miami-Atlanta 205 163 1998-2002
Miami-Richmond 183 63 1994-2004
Nogales-Tucson 117 92 1996-2000
Philadelphia-NYC 55 16 1997-2002
Pittsburg-Detroit 51 30 1998-2003
Portland-Salt Lake City 365 270 1993-2004
San Antonio-Dallas 111 106 1997-2000
San Diego-Los Angeles 49 39 1994 -2000
Seattle-Billings 212 202 1993-2004
Seattle-Blaine 38 38 1993-2004
Seattle-Sioux Falls 375 321 1993-2004
St. Louis-Oklahoma City 205 194 1999-2002
Tucson-San Antonio 316 248 1996-2000
Chicago-Cleveland 44 16 1996-2004
New York City-Cleveland 683 651 1996-2004
Omaha-Chicago 188 108 1996-2004

2.5 Variables for Regional Discrimination

Having settled on a three-region approach, the project team developed improved variables with greater power for regional discrimination for inclusion in the original equations. These included replacement of GDP as the variable used to describe economic growth with Gross State Product (GSP), which was developed by selecting from the states in which the corridors begin and end. Use of GSP allowed for improved discrimination between regions by better tying performance improvements to local economic growth. National Producer Price Index (PPI) inputs, used for discounting income and truck rate variables, were replaced with localized Consumer Price Index (CPI) numbers developed by the Bureau of Economic Analysis.

Rate-data specific to each corridor, traffic flows, and capacity information provide the key corridor-level discrimination. The addition of regional economic variables enhances the ability of the model to distinguish between regions.

Table 15 describes some of the key characteristics of each region given the improved dataset.

Table 15. Descriptive Statistics of Corridors by Region
  Region Average Minimum Maximum
Daily Truck Flows All Corridors 6,133 642 12,731
East Coast Corridors 7,113 642 12,433
Midwest Corridors 6,835 789 12,731
West Coast Corridors 4,666 1,726 8,338
Population All Corridors 4,445,071 392,572 16,971,055
East Coast Corridors 4,554,063 448,948 10,035,145
Midwest Corridors 2,836,174 392,572 9,154,470
West Coast Corridors 5,869,570 2,184,897 16,971,055
Per capita
income '000
(real)
All Corridors 18.122 12.403 26.079
East Coast Corridors 19.284 14.571 26.079
Midwest Corridors 17.617 14.228 20.842
West Coast Corridors 17.642 12.403 21.056
LTL Rates per mile
(real)
All Corridors 0.615 0.165 1.902
East Coast Corridors 0.749 0.217 1.902
Midwest Corridors 0.654 0.256 1.368
West Coast Corridors 0.472 0.165 1.684

2.6 Measuring Highway Performance

A primary requirement of the analysis methodology used to estimate the reorganization effect is to relate the demand for trucking and the rates that carriers charge shippers to a measure of highway performance. Although there are numerous measures that could be employed for this purpose (such as level-of-service indices), prior study suggests that the V/C ratio can serve as a reliable proxy to facility performance.[6]

Consequently, in this study, highway performance variables include V/C ratios and measures of delay along the corridors that are included in this analysis. In HPMS, V/C ratios represent the 30th busiest hour during a given year for a particular segment. V/C ratios for the corridors were estimated by taking the weighted average of the V/C ratios measured along individual segments (the length of the segments was used as the weight). Per-mile delay for a given corridor is the weighted average of all the delays on segments along the corridor.

Delay data were estimated by using the V/C ratios for selected segments and by estimating free flow and congested flow travel times. First, travel time on each segment was estimated assuming free flow of the traffic. Next, travel time with congestion was estimated as follows.

The model applies a standard equation to derive actual (congested) speed from information on the V/C ratio and free flow speed. Two equations were considered and tested:

  1. The Bureau of Public Roads (BPR) equation:

    the following expression should read: Congested Speed equals Free-Flow Speed divided by the term open parenthesis 1 plus 0.15 multiplied by open bracket volume divided by capacity close bracket raised to the power of 4 close parenthesisCongested Speed = (Free-Flow Speed) / (1 + 0.15 * [volume/capacity] ^ 4)
  2. The Metropolitan Transportation Commission (MTC) equation:

    the following expression should read: Congested Speed equals Free-Flow Speed divided by the term open parenthesis 1 plus 0.20 multiplied by open bracket volume divided by capacity close bracket raised to the power of 10 close parenthesisCongested Speed = (Free-Flow Speed) / (1 + 0.20 * [volume/capacity] ^ 10)

Finally, the difference between free-flow travel time and congested travel time is assumed to be the delay on this segment. The total delay figures were divided by the total length of the segments to estimate delay per mile as a highway performance measure for these corridors.

One approach considered, and then rejected, was not using a weighted average for delay per mile. A non-weighted approach would have the advantage of emphasizing the degree of delay in the highly congested segments of the corridor.

The weighted average approach is used in an attempt to describe the characteristics of an entire corridor. AADTT is the aggregate of the corridor segments. Therefore a delay factor that is also descriptive of the entire corridor is needed. By not weighting the delay by segment length, one would be vulnerable to accusations of over-valuing the portion of the corridor that is highly delayed and thereby over-valuing the relationship between performance improvements or reductions and demand.

Figure 3
Figure 3. Assumed Speed-Flow Relationships

Note: Assumes a free-flow speed of 60 mph.

  1. CzarLite is a nationwide (48 contiguous states) database of baseline class rates established on a territorial basis.
  2. Cohen, Harry. 1999. "On the Measurement and Valuation of Travel Time Variability Due to Incidents on Freeways." Journal of Transportation Statistics. December. http://www.gcu.pdx.edu/download/2cohen.pdf.

Download the free Adobe Reader to view PDFsYou will need the Adobe Reader to view the PDFs on this page.

previous | next
Office of Operations