## Which Method to Use

Safety performance functions, development (calibration, crash modification factors [Part C], crash proportions), IHSDM and other software

Moderator: khardy

### Which Method to Use

I have a suburban STOP controlled intersection that I'm considering for signalization. It seems there are two possible approaches for predicting the crash frequency that will result from signalization:

1) Use the predictive method (and EB) to get an expected number of crashes for current conditions (STOP control) and then apply the CMF for conversion from a STOP to a signal. This method is described in section D.4.4 on page D-5. Using this approach yields a prediction of less crashes with the installation of a signal.

2) Use the predictive method (and EB) to get an expected number of crashes for current conditions and then use the predictive method for the intersection assuming signalization and compare the two. This method is described in Part C and, specifically for suburban roads in Chapter 12. This approach yields a prediction of more crashes with the installation of a signal.

I'm wondering which approach is correct or what I'm not understanding correctly?

Joe O

Posts: 11
Joined: Fri Jan 14, 2011 9:29 am

### Re: Which Method to Use

Based on the HSM if you change the traffic control you should not use EB. To compare the two traffic control options based on Part C models you better not to use EB. Try no-EB for both signalies and Stop-control and see if the Signalized intersection has less predicted number of crashes. My guess is that it would.

Posts: 17
Joined: Thu Dec 02, 2010 11:49 am

### Re: Which Method to Use

I looked at it again. Here's what I did this time:

1) Used a calibrated SPF (i.e. the predictive method) + EB for the existing STOP controlled condition then applied the CMF for an urban traffic signal (.95). Note that section D.4.4 says this approach provides "the most reliable results." The predictive method + EB gives an expected crash frequency of 2.5 per year. If I apply the CMF for a signal then the expected crash frequency would be 2.5(.95) = 2.4 crashes/year.

If I don't apply EB the predicted crash frequency is 3.1 per year. Again, applying the CMF the predicted crash frequency with a signal goes to 3.1(.95) = 2.9 crashes/year.

If I apply a calibrated SPF for signal control (keeping everything else the same) I get a prediction of 5.6 crashes per year.

I have checked my math and I believe I am applying the methods correctly. That said, I could be making a mistake. If someone would like to check here is the data I am using for the intersection:

Major Street Volume: 29350
Minor Street Volume: 2700
Left Turn Lanes: Present on all four approaches (only considered on the major street approaches for STOP controlled analysis)
Right Turn Lanes: Present on one major street approach and both minor street approaches (only considered major street approach for STOP analysis)
Left Turn Phasing: Permitted Only LT phasing on all four approaches when considering signalization option.
All other CMF's = 1.0
Crash History - 1.75 multi-vehicle crashes/year over past four years. 0 single vehicle crashes.
Local calibration factor = 1.37 for STOP controlled intersections and 1.45 for signal controlled intersections.
Joe O

Posts: 11
Joined: Fri Jan 14, 2011 9:29 am

### Re: Which Method to Use

Joe
Your numbers look fine. I did a little bit of sensitivity analysis with the Part C models and realized that for major AADT around 10,000, the predictions of total crashes for Stop Control and Signalized intersections are close. For major AADTs lower than 10,000, Signalized intersection prediction is lower than Stop Control prediction and for major AADTs more than 10,000 Signalized intersection prediction is higher than Stop Control one. However, the warrant for signalization replacing Stop Control is just the opposite. The warrant kicks in As the AADTs go higher, and that is mostly from the operational point of view. Anyway, because of the following reasons I recommend you to use Part C models and not Part D model that you mentioned:

1. Part C models are part of the whole prediction model and are related to a SPF for which you have calibration factors for your area. Based on the HSM using Part D CMFs combined with the SPFs of Part C is not recommended.
2. Part C models are affected by the AADTs and you have this data available.
3. Part C models are affected by the turn lanes and you have this data available.
4. Part C models have also the split of the predicted crashes by severity.

For your specific case it seems replacing the Stop sign with Signal is not an improvement from safety point of view. However, it might be required or preferred from operational point of view. If this is the case then may be besides the prediction of crashes, an estimation of delay time and delay cost might help you to make a final decision.

Posts: 17
Joined: Thu Dec 02, 2010 11:49 am

### Re: Which Method to Use

Thanks for the reply. I was thinking as well that the predictive method was the better approach for basically the same reasons as you stated.

However, this brings up some interesting questions:

1) Is a CMF for conversion from STOP control to signal control even needed? If there's an SPF and we agree that it is the better approach why would we even need a CMF for this?

2) Is the CMF valid? At nearly all volume levels the predictive method finds an increase in crashes with signalization yet the CMF always predicts a decrease.

The second question is the more interesting of the two to me. It has many more far reaching implications.

Chapters 4 - 6 of the HSM outline the traditional approach to safety analysis: Network screening to identify problem locations followed by a review of the crash history, geometry, traffic control etc. in order to identify possible countermeasures to address identified problems.

Unfortunately, this approach seems to get lost in later chapters. Part C calls for the uncritical application of CMF's in the models to predict accident frequency. This despite the fact that the CMF's may not always make sense. For example, you wouldn't expect an accident reduction because of the presence of a right turn lane on an approach with no rear end crash history associated with right turners or on an approach with no right turn volume. Yet the CMF for a right turn lane at a 4-legged signalized intersection is 0.96 (see Table 12-26, page 12-44) -- always leading to a predicted decrease. Part D calls for the application of CMF's to predicted crash frequencies without any sort of critical analysis or engineering judgment. It essentially endorses the option we agreed was not desirable in my example above.

Do we really have that much faith in the CMF's? I took a look at the study that determined the CMF for conversion of a rural STOP controlled intersection to a signalized intersection (link here: http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_rpt_617.pdf. The study involved intersections that State DOT's were planning to convert from STOP control to signals. The fact that they planned to signalize the intersections suggests that the intersections had been through some sort of screening process and signalization was deemed appropriate. Chances are if they were at the top of the funding list, they had a crash history that was part of the determination. That being the case, of course we would expect that there would be a reduction in crashes. However, that doesn't mean that we should expect a crash reduction at ANY rural STOP controlled intersection. It only means that we should expect a reduction under the conditions that lead to signalization at those locations.

The study authors note as much. They state: "The results also provide guidance on the intersection conditions under which signal installation would be most beneficial". They made this statement noting that their study signals tended to see an increase in rear end crashes along with a reduction in right angle crashes. Again, they were recognizing the need for network screening, diagnosis and proper selection of countermeasures.

Mohamad, our discussion above did not include any discussion of crash history at my location and only a cursory discussion of geometry. While we had the benefit of an SPF to help us recognize that the CMF was not valid in our application there are many CMF's without that benefit. How do we know under what conditions they are valid?

I see this as a big issue as we move forward with the HSM. Many of the CMF's simply cannot be applied using a "black box" approach. Practioners need to utilize the methods outlined in Chapters 4 - 6 to make sure that they are selecting the appropriate countermeasures for the conditions. I hope that is the intent in the manual but as it is currently written (and being presented) it is far from clear that that's the case.

Note that my long-winded comment here is not intended as criticism of the HSM or the efforts put forth to develop it. It is a welcome addition to our profession in my opinion. That said, I realize that the HSM is a work in progress. My intent here is to hopefully generate some discussion. I feel strongly about this and I'm concerned that there is potential for misuse of CMF's without the proper analysis and judgment applied to make sure that they are appropriately applied.

Thanks again for the opportunity to discuss.
Joe O

Posts: 11
Joined: Fri Jan 14, 2011 9:29 am

### Re: Which Method to Use

Joe
I agree with you 100%. I hope comments that HSM users put here would be used by the TRB Highway Safety Performance Committee in looking into the development of the 2nd edition of the HSM.

Posts: 17
Joined: Thu Dec 02, 2010 11:49 am

### Re: Which Method to Use

Installing a signal where there once was two-way stop control will increase crashes. Specifically, you can expect an increase in the number of rear-end collisions due to the (new) need to stop on the main street. That's why signal warrants require an examination of the types of crashes - the types of crashes you prevent (right-angle) through the installation of a signal are typically more severe than the ones that increase (rear-end).
prytyka

Posts: 2
Joined: Mon Mar 26, 2012 8:17 am

### Re: Which Method to Use

I am sure there is a set of circumstances in which replacing minor stop-controlled traffic control with signal in Urban areas will increase number of crashes. In Part D of HSM, page 14-13, we see that in another set of circumstances this replacement has a CMF of 0.95, meaning that it will decrease the number of crashes by 5%. What I believe the best approach is, if required data are available, to use the crash prediction models in Part C. By using these models, major affecting factors, including AADTs on major and minor approaches as well as some geometric characteristics such as skew angle, will have their saying in the estimation of the expected number of crashes. This is probably the best approach for comparing the two types of traffic control. However, since we are talking about changing the intersection type, from Part C point of view, we cannot use EB in these evaluations.