CMF Definitions, Scope, and Evidence Quality

CMF Definitions, Scope, and Evidence Quality

Interpret a relative effect

A Crash Modification Factor (CMF) represents the ratio of expected crash frequency with a treatment or condition to expected frequency under a defined comparison condition. It concerns a specified outcome and population. A CMF is not a crash count, a probability that a road is safe, or a guarantee that the next year will contain a particular number of crashes.

For a suitable baseline $N_b$ and applicable CMF:

$$N_t=N_b\times CMF$$

If $CMF=0.72$, the treated expected frequency is 72% of the untreated expected frequency: an estimated 28% reduction. If $CMF=1.15$, the estimate is a 15% increase for the defined outcome. A value of 1 describes no estimated relative change; it does not establish that the true effect is exactly zero with certainty.

An SPF estimates the baseline frequency under modeled conditions. A CMF supplies a relative effect. These roles differ, even when a predictive procedure uses them together. Check whether a feature is already represented in the model before multiplying an additional effect for the same feature.

Convert CMFs and reduction percentages

The Crash Reduction Factor (CRF) expressed as a percentage is:

$$CRF=100(1-CMF)$$

Conversely, $CMF=1-CRF/100$. A CMF of 0.62 gives a CRF of 38%, not 62%. A negative CRF indicates an estimated increase. Be careful with the acronym CRF: economic analysis also uses “capital recovery factor,” which is a different concept.

For an illustrative untreated expectation of 10 targeted crashes per year, a CMF of 0.62 gives 6.2 treated expected crashes and a reduction of 3.8. Fractional values describe expected means; actual events remain integers. This example does not assert that a particular named treatment has that effect.

Match collision type and severity

A CMF for run-off-road injury crashes should not be applied automatically to all crashes at a corridor. A total-crash CMF and a fatal-and-injury CMF measure different outcomes. Likewise, fatal-and-injury categories may include several injury levels and are not automatically identical to fatal-plus-suspected-serious categories.

Disaggregate the baseline to the scope supported by the evidence. If a study concerns pedestrian crashes, determine whether the local baseline uses the same definitions. If the study concerns a particular intersection control or facility type, review that match too. Applying a credible estimate to an incompatible outcome can produce a misleading benefit even when the arithmetic is correct.

Some treatments affect several outcomes differently. A change that reduces angle crashes can increase rear-end events. A lower total count may conceal a serious-injury increase, and an increase in minor events may accompany a severe-harm reduction. Review the relevant types and severities rather than treating the CMF as one universal score.

Assess study quality and applicability

The FHWA CMF Clearinghouse provides estimates and study information. Its quality ratings concern evidence quality, not the size of the reduction or automatic federal funding eligibility. A five-star estimate is not necessarily the best match for the site, and a low numerical CMF is not automatically the most trustworthy.

Read the source study and documentation: design, sample, exposure adjustment, selection bias, standard error or interval, comparison conditions, facility, region, and implementation. No universal rule assigns a particular star rating solely because a standard error is below 0.10. Different study designs can provide strong evidence, and EB is not a mandatory prerequisite for every highest-quality rating.

Transferability needs judgment. Weather, reporting, users, fleet, maintenance, and operation can differ. Local SPF calibration does not create a universal requirement to recalibrate every CMF through a single factor. Use the method and evidence appropriate to the treatment and document the uncertainty.

Interpret precision and confidence intervals

A CMF is an estimate. Its uncertainty may be reported through a standard error or interval. Under an appropriate approximate normal assumption, a 95% interval can be illustrated as $CMF\pm1.96SE$. Not all estimators use symmetric normal intervals; use the study's actual method.

For an illustrative CMF 0.72 with SE 0.06, the interval is about [0.602, 0.838]. It lies below 1 under that approximation. For 0.88 with SE 0.08, it is about [0.723, 1.037], including 1. The second result does not demonstrate a reduction at the corresponding two-sided 5% test level, but it also does not prove no effect.

Statistical significance is not the same as practical importance or applicability. A precise small effect may matter over a large network. An imprecise large effect may be promising but uncertain. Consider mechanism, implementation, baseline harm, costs, and sensitivity alongside precision.

Read beyond the CMF point estimate

• Scope: crash type, severity, users, and comparison.
• Evidence: study design, sample, bias, and uncertainty.
• Application: local mechanism, implementation, and baseline compatibility.

Select evidence for a decision

For two candidate CMFs of 0.75, compare the studies' scope, quality, uncertainty, and local fit. Do not choose merely by star count or treatment popularity. If one concerns the relevant users and control while the other does not, that difference can be decisive. If neither fits, explain the limitation and investigate additional evidence.

Record the selected estimate, source, outcome, assumptions, and any alternative estimates considered. Carry the same scope into the economic appraisal and evaluation plan. A defensible decision can acknowledge uncertainty while choosing a reasonable, feasible treatment; precision should not be manufactured by copying an unsupported benchmark table.