PUBH 8033 Week 6 Confounding Analysis Example

What Else Could Produce the Gap? A Confounding Analysis of a Composite County's Older Driver Class, Naming the Rival That Early Seats Went to Lower-Mileage, Crash-Free Drivers and Showing How Far Adjustment Moves the Estimate

Student Name

Walden University

PUBH 8033: Interpretation and Application of Public Health Data

Instructor Name

Month Day, Year

What Else Could Produce the Gap? A Confounding Analysis of a Composite County's Older Driver Class, Naming the Rival That Early Seats Went to Lower-Mileage, Crash-Free Drivers and Showing How Far Adjustment Moves the Estimate

The Gap to Be Explained

People who finished Larkin County's Safe Driving After 70 class appeared in the state crash file less often over the following year than those left on the waitlist, at roughly 46 versus 57 crashes per thousand drivers (illustrative ratio 0.81; interval 0.43 to 1.55). Before treating that gap as an effect of the class, this paper asks what else could produce it. The figures are composite.

The Rival Explanation

Seats were offered in order of application, and the class filled within weeks at the senior centers and libraries. Applicants from the state licensing office, many referred after a renewal visit or a recent crash, tended to apply later and were more often waitlisted. The intake data showed that licensing-office applicants were older and nearly twice as likely to report a crash or near miss over the preceding twelve months. The rival explanation is therefore specific: people who secured early seats were younger, drove fewer risky miles and had fewer prior crashes, so they would have had fewer crashes in the following year even without the class.

This is not a hypothetical concern. An evaluation of a mature driver course found a self-selection bias among attendees, whose driving records differed from those of comparable drivers before they ever attended (Nasvadi & Vavrik, 2007).

Checking the Rival in the Intake Data

The table supports the rival. Waitlisted applicants were older, more often reported prior crashes or near misses, came more often through the licensing office and had nearly twice the rate of prior police-reported crashes. They also reported driving somewhat fewer miles, which pulls in the other direction, since fewer miles usually means fewer crashes.

Table 1

Intake Characteristics of Completers and Waitlisted Applicants

Note. Composite figures for illustration.

Adjusting the Estimate

A Poisson regression of crash involvement on completion status, adjusting for age group, log annual miles, prior police-reported crash and enrollment site, moves the rate ratio from 0.81 to 0.92, with a 95% confidence interval from 0.48 to 1.76. Most of the change comes from adjusting for prior crashes and enrollment site. In other words, much of the apparent benefit reflects who got seats rather than what the class did.

What Adjustment Cannot Fix

Adjustment controls only for measured differences. Several plausible confounders were not measured: vision and cognitive changes, recent medical events, family pressure to stop driving and the reasons a licensing clerk handed someone a flyer. If those factors are linked to both late application and later crashes, the adjusted estimate is still biased, probably still toward showing a benefit. Self-reported miles are also imprecise; older drivers may misjudge their own mileage, so adjustment for exposure is incomplete. The estimate after adjustment should be read as "little or no measurable effect, with considerable uncertainty," not as "the class has no effect."

What the Evidence Elsewhere Suggests

The adjusted estimate is closer to what randomized evidence shows. In an Alabama trial, older drivers with impaired vision who were randomized to safety education crashed at about the same rate per person-year as controls, 1.08 times as often with an interval reaching from 0.71 to 1.64 (Owsley et al., 2004). A 2009 review likewise placed lecture-based retraining among interventions not shown to lower crash counts (Korner-Bitensky et al., 2009). When a nonrandomized comparison moves toward the randomized evidence after adjustment for an obvious rival, that is a sign the rival was real.

Checking the Direction of Bias

It helps to reason about which way each unmeasured factor would push the estimate. Applicants who had recently been told by a doctor or family member to stop driving were likely to apply late, after the senior center seats were gone, and likely to crash more if they kept driving. Leaving that factor out makes the class look better than it is. Applicants who were still working and busy may also have applied late and driven more miles, but they were likely healthier drivers, which would make the class look worse. On balance, the intake data suggest the first group is larger, since late applicants were older and more often referred through the licensing office. That reasoning implies the adjusted estimate of 0.92 is, if anything, still too favorable to the class.

A Sensitivity Check

One simple check is to repeat the analysis among senior center and library applicants only, excluding the licensing-office group whose referral path differs most. In that subgroup, completers and waitlisted applicants were more alike at intake, and the rate ratio was 0.95 with a wide interval. The estimate moves toward no effect as the comparison becomes fairer, which is what we would expect if selection, not teaching, explained most of the original gap.

A Better Comparison

The cleanest way to remove this rival would be to allocate seats by lottery among applicants, offering waitlisted applicants seats the following year. That would make completers and waitlisted applicants alike on average in both measured and unmeasured characteristics. The county could do this at no extra cost, since demand already exceeds supply.

What the County Should Report

The annual report should present the unadjusted and adjusted estimates side by side, with intervals, and a sentence explaining why they differ. It should describe the intake differences between completers and waitlisted applicants in a table like the one above. And it should state plainly that unmeasured differences may remain. Reporting only the unadjusted figure, as the program's first draft did, would invite readers to credit the class with a benefit that mostly reflects how seats were filled.

Conclusion

The unadjusted gap between completers and waitlisted applicants is largely explained by who obtained early seats. After adjustment, the estimate suggests little or no effect, consistent with randomized evidence on classroom-only programs. The finding argues against expanding the current class on the basis of the unadjusted figure and for a lottery-based evaluation of any new version.

References

Korner-Bitensky, N., Kua, A., von Zweck, C., & Van Benthem, K. (2009). Older driver retraining: An updated systematic review of evidence of effectiveness. Journal of Safety Research, 40(2), 105-111. https://doi.org/10.1016/j.jsr.2009.02.002

Nasvadi, G. E., & Vavrik, J. (2007). Crash risk of older drivers after attending a mature driver education program. Accident Analysis & Prevention, 39(6), 1073-1079. https://doi.org/10.1016/j.aap.2007.02.005

Owsley, C., McGwin, G., Phillips, J. M., McNeal, S. F., & Stalvey, B. T. (2004). Impact of an educational program on the safety of high-risk, visually impaired, older drivers. American Journal of Preventive Medicine, 26(3), 222-229. https://doi.org/10.1016/j.amepre.2003.12.005