The main goal of this analysis, for both the Heartland and the Crescent Corridors, is to characterize the average behavior of the mode choice decision (in terms of probabilities) for all the shipment records included in PUMS that interact with any of the two corridors. The final output of this section can be interpreted as a proxy for the actual dynamics of the freight mode choice activity in the area surrounding the corridors as a function of the commodity type and the driving time by truck, by the time the PUMS data records were gathered.
The analysis presented employs the unweighted mode choice models which uses generalized cost (with 5% as opportunity cost) as its input variable, as discussed in Chapter 8. Commodity specific models were used where available; otherwise, the pooled models were used. To obtain the shipment-level data required by these models, the PUMS dataset was post-processed with the intent of only including the relevant records, i.e., the ones that may be impacted by the corresponding network improvements. The way in which these subsamples were selected varies for each corridor and it will be a matter of study later in this document. For each corridor, the three origin zip codes with the highest sample size were selected to apply the models and to evaluate the impact of the commodity type and the transit time by truck in the aggregated probabilities along each corridor for all the records starting at the selected origin.
For presentation purposes, the results were aggregated by taking the mean value of all the probabilities corresponding to records that lie within a specific interval of driving time by truck. The intervals were constructed starting at 0 and with a length of 4.5 hours. Driving times exceeding 13.5 hours were grouped in the same bin based on the current hours of service regulations for truck drivers, previously discussed (Federal Motor Carrier Safety Administration 2011). The results were aggregated based on the nine super-groups of the SCTG codes that the CFS presents in the data user guide for the PUMS. A general overview of the composition of the groups is presented in Table 59 (United States Census Bureau 2018).
Since the aggregation process is based on several sampling stages and does not account for all the possible records under each one of the conditions analyzed, there is a component of error in the results presented. To overcome the fact that some aggregate estimates might potentially exhibit counterintuitive results because the sampling error is high, a sample size threshold of 50 observations was defined. For each one of the bins corresponding to a driving time interval and a SCTG super-group, it is considered that the average probability estimated is reliable if the number of observations in that bin is greater than or equal to 50.
The motivation behind the chosen threshold comes from the fact that the data showed relatively small 95% confidence intervals (in terms of length) when the subsample size is above 50 observations. It should also be noted that the sample size within each one of the bins in the analysis and the reliability of the estimate obtained for each bin is out of the team’s control because the sample sizes are determined by the PUMS data.
Heartland Corridor’s Inspired Scenarios
Scenario #3: Estimation of Truck Market Shares by Commodity Groups
For the purpose of this analysis, observations from the states interacting with the Heartland Corridor were the only ones considered. These states are: Delaware, District of Columbia, Illinois, Indiana, Kentucky, Maryland, Michigan, Ohio, Pennsylvania, and Virginia. Other observations that were eliminated to arrive at an effective sample size of 249,279 (93.123% of the original sample of interest) include:
- intrastate shipments;
- observations lacking specific information on commodity type, and
- observations with Coal (SCTG 15) as the commodity type, as there was no shipment size model available for this commodity.
Essentially, the results presented in this section represent the estimates of the mode shares from the ZIP codes shown in Table 60 to destinations in the hinterland of the Heartland Corridor.
In interpreting the results, the reader ought to keep in mind that, since the market shares are estimated as the average of the shipment-level probabilities and the number of observations is relatively low, the market shares are bound to exhibit variability on account of the sampling error. That is the most likely reason to explain apparently counterintuitive results in addition to the fact that this numerical experiment does not control for the variability of all the factors involved in the mode choice process. However, the reader must be aware that some aggregate estimates were cataloged as “highly unreliable” because of having a low sample size (less than 50 observations) and are not displayed in the graphs. Besides, even if all the observations for coal were removed, the SCTG super group # 4 that contains coal, petroleum, gasoline, fuel oils, other coal and petroleum products (SCTG codes from 15 to 19) is displayed in the graphs because there were other observations associated with the remaining commodities in this group that were included in the analysis.
Origin Zip Code: 16874, Philadelphia
Origin Zip code: 45856, Ohio
Origin Zip code: 60629, Illinois
For the scenario of Philadelphia, the results indicate that most of the commodities exhibit a strong preference for trucks. Only the groups of coal, petroleum, etc. and chemicals/pharmaceuticals have different patterns. The former shows high sensitivity to driving time while the latter shows the lowest propensity to use trucks. When the shipment size models are used, the most sensitive groups are grains, oils, etc.; and coal, petroleum, etc.; with the former reaching probabilities close to zero.
In the case of Ohio, the constant patterns are exhibited by all commodity groups with chemicals showing the lowest probabilities, using PUMS information. For the analysis using the shipment size models, the coal, petroleum, etc. and grains, oils, etc. groups are the ones that appear to be more sensitive. The building stone group also shows a drop in the probabilities under this scenario.
The results for Illinois are very similar to the ones for Philadelphia and Ohio. Building stone, chemicals and coal groups are the most sensitive to variations in the driving time when the information about shipment size is taken from PUMS. In the context of the shipment size models, animals and agriculture, and grains, oils, etcetera; show very sensitive patterns by decreasing very rapidly as long as the driving time increases.
Crescent Corridor’s Inspired Scenarios
Scenario #4: Estimation of Truck Market Shares by Commodity Type
For the Crescent Corridor, the states under consideration were: Alabama, Delaware, District of Columbia, Georgia, Kentucky, Louisiana, Maryland, Massachusetts, New Jersey, New York, North Carolina, Ohio, Pennsylvania, South Carolina, Tennessee, Virginia, and West Virginia (17 in total). Similar to the analysis of the Heartland Corridor the following were eliminated to determine the effective sample of 366,649 observations (90.256% of the original sample of interest): (1) intrastate shipments; (2) observations lacking specific information on commodity type, and (3) observations with Coal (SCTG 15) as the commodity type, as there was no shipment size model available for this commodity. Table 61 shows the three origins chosen for the analysis in this corridor:
Origin Zip code: 07104, New Jersey
Origin Zip code: 30318, Georgia
Origin Zip code: 16874, Philadelphia
For the case of New Jersey, the group of wood, paper and textiles is the most sensitive group no matter what the source for the shipment data are. Additionally, the probabilities exhibited by this group are the lowest among the nine groups. The groups for building stone and coal are the ones that present the biggest changes with respect to the source of the shipment information.
In the case of Georgia, the results do not differ significantly when the shipment information from PUMS is used in comparison with the results when the shipment size models are used. Once again, wood, paper, textiles group shows the lowest probabilities and the highest sensitivity to variation in the driving time by truck.
For Philadelphia, the same conclusions already mentioned about the wood, paper, textiles group. The commodity groups that are more sensitive to the source of the information about the shipment are the coal group and the building group, with the latter showing high sensitivity in both cases.