by **mdimaiuta** » Fri Jan 28, 2011 4:36 pm

Joe - my responses are below, in red:

"I've been working to understand the new methods outlined in the HSM. I've built a spreadsheet to evaluate and screen 4-legged signalized intersections using the approach shown in example 4.4.2.13 - Excess Expected Average Crash Frequency with EB Adjustments (starts on page 4.75) along with the information in Chapter 12 for urban/suburban arterial intersections.

I have a couple of questions:

1) The example data on page 4-76 shows three years of accident data at each location. However, the calculations shown in steps 6, 7 and 8 appear to only be for year three. In a previous example (example 4.4.2.8) the average observed and predicted accident frequency over three years was used. Presumably, that same approach would be valid in this example as well?

You are correct that the calculations shown in steps 6, 7 and 8 are only for year three. Taking an average over three years would be valid, keeping in mind that the Severity Weighted Excess calculated in Step 7 would then reflect the average (per year) expected crash value, rather than a total for the entire three year period.

2) Appendix C.A.1 contains information on calibration of the Part C predictive models. Section A.1.1.3 notes that "Observed crashes for all severity levels should be included in calibration." The example in that section assumes a rural two-lane four-legged signalized intersection for which there is only one SPF. In that case, the approach makes sense.

The methods in chapter 12 for urban/suburban 4-legged arterial intersections are more complex. They require the prediction of multi-vehicle and single vehicle crashes separately by severity and also pedestrian crashes and bike crashes separately. In that case I wonder if it wouldn't be more accurate to calculate separate calibration factors for each crash type and severity level? Since each is being calculated separately it is straightforward to calculate separate calibration factors as well. When I did this, I got calibration factors that varied between 1.25 and 2.41 depending on accident type and severity level. Is there any reason to not use this approach?

I recently spoke with the model developer about this issue. He said that there is no problem with calculating separate calibration factors for each crash type and severity level, as long as you have enough observed crashes in your calibration data set to support that type of breakdown. As noted in section A.1.1.2 of the Appendix to Part C, the desirable minimum sample size for the calibration data set is 30 to 50 sites, and the entire group of calibration sites should represent a total of at least 100 crashes per year. So, if you wish to derive separate calibration factors by crash type and severity level, then each of those factors would require a minimum of 30 to 50 sites and 100 observed crashes per year.

Any feedback would be appreciated. I'm happy to see these methods made available to practitioners.

Also, FWIW, the example calibration factor calculation references equation 10-18 which doesn't appear to be correct. I think it should be equation 10-10. Also one of the values in the equation does not match the value in equation 10-10. 10-10 says -5.13 and the example says -5.73."

I agree that these are errors on p. A-8 and A-9. The reference to equation 10-18 at the top of p. A-8 should be to equation 10-10, and the equation for N spf int should read e(-5.13 + ...) rather than e(- 5.73 + ...) The error propagates throughout the example problem, including Table Ex-1.

It appears that there are additional errors on pages A-8 and A-9:

- p. A-8, 3rd paragraph after the bullets: “The intersection has a left-turn lane on the major road, for which CMF1i is 0.67, and a right-turn lane on one approach, a feature for which CMF2i is 0.98.” This should be “The intersection has a left-turn lane on the major road, for which CMF1i is 0.82, and a right-turn lane on one approach, a feature for which CMF2i is 0.96.” Note that the correct values are shown in the bulleted items preceding the incorrect text.

- P. A-9, Table Ex-1: CMF1i (column 5) for “Intersection Approaches with Left-Turn Lanes” (column 4) equal to 1 should be changed from 0.67 to 0.82 (rows 1, 5, 7 and 8). The CMF2i values (column 7) for “Intersection Approaches with Right-Turn Lanes” should be 0.96 for one approach with a right-turn lane (rather than 0.98) and 0.92 for two approaches with a right-turn lane (rather than 0.95).

Given the above, all of the SPF Prediction values (column 3), Predicted Average Crash Frequency values (column 9), the resulting Calibration factor, and references to these values on p. A-8 are also wrong.