The American Enterprise Institute’s Andrew Biggs reacted to my critique of his Yankee Institute study on public-sector pay in Connecticut with a lengthy response (see also my related blog post). I take this as a good sign: Maybe facts do matter, even in the Trump era.

Before delving into methodological issues, however, I’ll answer one pointed question Biggs had: No, my critique of his research wasn’t commissioned by labor unions, though EPI does receive about a quarter of its funding (27 percent, to be precise) from unions, who helped found EPI and are represented on our board. Most of the rest of our funding comes from foundations.

As it happens, Biggs’s study was brought to my attention by an advocacy group, Connecticut Voices for Children, that receives most of its funding from community foundations and almost none from unions. Connecticut Voices asked for my help in understanding Biggs’s methodology. Since this ended up taking longer than expected, I decided to write a report for which EPI didn’t receive any dedicated funding. I did share a late draft with an expert in public-sector compensation at the National Education Association in Washington, who had no influence on my analysis. I actually think my job would have been easier and the report better if I’d consulted with union representatives in Connecticut, though I understand why Biggs might not agree.

Now that I’ve come clean, I’d be curious to know who commissioned Biggs’s report and who he received feedback from—AEI board member and Education Secretary nominee Betsy DeVos, by any chance?

Moving on to methodology, I seem to have hit a nerve by pointing out that a paper purporting to compare public- and private-sector workers in Connecticut excluded 82 percent of public-sector workers from the sample used to compare wages and salaries. Biggs’s response is that (1) it’s appropriate to limit the sample to state employees if the focus is on state policymaking; (2) it would be labor intensive to include local government employees because they may receive different benefit packages; (3) teachers’ work schedules complicate pay comparisons; (4) the compensation of public-safety personnel reflects the risks of their jobs, and including them would make public-sector workers appear more “overpaid,” not less; and (5) excluding part-time workers is standard in such studies.

None of these arguments holds water except the last, which I never took issue with. As will be detailed below, Biggs’s cherry-picked sample can’t be justified by practical considerations, since dropping the various groups complicates his analysis and leaves him with a different sample of workers than he uses for benefit comparisons. And it’s disingenuous for Biggs to suggest that his goal is to focus on state policymaking when he is clearly pushing for cuts to pension and retiree health benefits that would affect workers employed in both levels of government.

Many people are understandably confused about which public-sector workers are state or local government employees. Though K-12 teachers are hired by local school boards, school funding comes from a mix of federal, state and local government sources and most teachers participate in state-wide pension plans—in this case, the Connecticut Teachers Retirement System (CTRS). Many other local government employees, including police officers and firefighters, also participate in state-wide plans, such as the Connecticut Municipal Employees Retirement System (CMERS). Like most states, Connecticut also has pension plans for state employees, the largest of which is the State Employees Retirement System (SERS), though there are two other plans for judicial employees. There are also smaller local plans, such as the City of New Haven Employees Retirement Fund. While Biggs points out that it would be time consuming to gather information on these smaller plans, the two largest state-wide plans alone—SERS and CTRS—cover 80 percent of public-sector workers in the state, so including teachers and other local government employees in the analysis, as I did, isn’t such a heavy lift.

Local government employees outnumber state employees, many of whom are doctors and other highly paid public-sector workers employed by the state university system, a fact documented by the Yankee Institute but never acknowledged by Biggs. While an argument can be made for taking a hard look at coaches’ and doctors’ salaries, Biggs, whose area of focus is retirement, is clearly gunning for cuts in pension and retiree health benefits that would affect teachers, police officers, and firefighters—not just the state employees included in his pay sample. Biggs specifically mentions the teachers’ retirement system in his critique of public-sector pensions.

Furthermore, though Biggs bases public-sector pension costs on the State Employee Retirement System, SERS participants include state police officers and other unspecified “hazardous duty” workers who were presumably dropped from his pay sample. It’s misleading to suggest, as Biggs does, that by dropping public-safety workers he excludes a group of highly-compensated workers from his public-sector sample when he excludes them only from his pay sample and it’s their benefits, not their salaries, which are generous. It’s also strange to single out public-safety personnel for the risks they incur on the job while otherwise ignoring working conditions experienced by other workers. Ignoring working conditions is necessary to avoid a host of complications, but then it’s inappropriate to exclude specific occupations on that pretext.

Some teachers also participate in the state retiree health plan, though most participate in another, less generous, teacher plan (this was brought to my attention after my report was published). Other retired local government employees don’t participate in either plan, though in my analysis I assume they all receive retiree health benefits that are equivalent in value to the more generous state plan, which exaggerates their cost. As I note in my report, putting a value on future retiree health benefits isn’t easy because these benefits aren’t funded in advance and employers can discontinue them at any time. In actuarial valuations, the normal cost of retiree health benefits in Connecticut appears implausibly high. Their normal cost is nearly double that of pensions in the state and is among the highest in the country even though the benefits themselves don’t appear unusually generous (the eligibility requirement is longer than average, for example).

The high cost of retiree health benefits is in part due to very high healthcare costs in Connecticut, while their cost relative to pensions partly reflects the fact that retiree health benefits are funded on a pay-as-you-go basis and discounted using a low rate of return rather than the expected return on fund assets. However, the high projected cost of retiree health benefits is also due to the actuaries’ assumption that health cost inflation will be high for the foreseeable future and will continue to outpace employers’ ability to shift these costs onto workers and retirees. This is probably unrealistic. Even rather pessimistic projections from the Congressional Budget Office recognize that healthcare costs can’t outpace economic growth forever, though the timing of the slowdown depends on public policy and other hard-to-predict factors.

In other words, Biggs and I both rely on actuarial estimates that likely exaggerate the cost of retiree health care. For other public-sector benefits, Biggs and I both use Bureau of Labor Statistics measures that include state and local government workers, another reason not to arbitrarily exclude any of these workers from pay comparisons. Factoring in the inflated cost of retiree healthcare suggests that more generous benefits do not make up for public-sector workers’ lower salaries as I conservatively estimate in my analysis.

Biggs is right to point out that there’s no perfect solution to the complication that teachers have long seasonal breaks that are only partly offset by less discretionary leave. (It’s also likely that teachers, who are excluded from overtime compensation, take home more work than the average worker, though this doesn’t concern Biggs.) This challenge shouldn’t be exaggerated, however, since survey measures attempt to adjust for differences in time spent working, albeit imperfectly, and there are ways researchers can address this problem—by focusing on weekly pay, for example. In short, this isn’t sufficient reason to drop the largest and possibly most underpaid group of public employees from the sample.

I go on at length in my report about Biggs’s choice of control variables in his pay regression, not because they matter more than his choice of discount rate to measure pension benefits—they don’t—but because the discount rate issue has been discussed ad nauseum whereas sins of commission in econometrics are less well understood than sins of omission (“omitted variable bias”). Biggs correctly points out that many of the control variables I criticize in his model only affect his measure of the public-sector pay penalty by one or two percentage points, though these do add up. It’s true I spent too much time discussing details of his pay analysis, though Biggs spent a lot of time on these details as well. Still, I regret belaboring the pay analysis because people get lost in the weeds.

Worse, some readers may have come away thinking: “If a conservative think tank says public-sector workers are compensated a lot more than private-sector workers, and a progressive think tank says they’re compensated the same, the truth must lie somewhere in the middle and it can’t hurt to trim public-sector pay and benefits.” This is faulty logic. Though average compensation is roughly the same across sectors, public sector workers with college degrees are compensated less and those without college degrees are compensated more. Keeping in mind that many private-sector employers don’t pay a living wage, cutting public-sector pay or benefits would make it harder for public-sector employers to recruit and retain college-educated workers while shifting the subsistence costs of non-college-educated workers and their families onto taxpayer-funded safety-net programs.

Another reason public employers shouldn’t cut workers’ pay and benefits is that these have already been cut significantly in ways that aren’t yet apparent in the data examined. Among non-hazardous-duty Connecticut state employees, for example, the normal employer cost of pensions for Tier I and II workers hired before June 1997 is 8.0 and 8.9 percent of pay, respectively. Many of the workers in the 2009-2013 sample that we analyzed are in these tiers. In contrast, the normal employer cost of pensions for Tier IIA and Tier III workers hired since June 1997 is much lower—3.1 and 4.2 percent of pay, respectively. Thus, even if public-sector workers in Biggs’s sample are currently compensated the same as private-sector workers, there will be a growing gap in pay between sectors as older workers retire unless private-sector workers also experience significant pay or benefit cuts.

Biggs notes that “EPI’s own research on public sector pay…controls for race, gender, ethnicity and citizenship, among other factors,” suggesting that I’m hypocritical to note that these will produce biased estimates of the pay gap between equally qualified workers in the two sectors. I do include these control variables in my analysis, while noting that “the results should be interpreted with caution and treated as conservative estimates of pay gaps” (emphasis added).

In other words, my colleagues and I bend over backwards to use standard control variables in pay regressions even though this serves to minimize the public-sector pay penalty. In contrast, Biggs’s overall methodology serves to bias the results in his favor, even though he prefers to highlight the fact that he includes one debatable variable—employer size—that has the opposite effect. I stand by my critique of his control variables even though Biggs says with respect to one group that “they can have important effects on the results.” My point exactly. Just because it’s possible to control for the fact that lower-paid workers tend to live in lower-cost neighborhoods, for example, doesn’t mean it’s appropriate to do so.

As for benefits, both Biggs and I probably exaggerate the higher cost of retiree health insurance, which is more prevalent in the public sector. This means public-sector workers are almost certainly compensated less as a group than private-sector workers, not the same amount as I conservatively estimate. Biggs notes that in the past EPI hasn’t explicitly included retiree health benefits in pay comparisons because these aren’t included in Bureau of Labor Statistics “Employer Costs for Employee Compensation” (ECEC) measures of benefit costs for active workers. Implicitly, however, some retiree health costs are included in these measures to the extent that they’re reflected in higher “blended” premiums paid on behalf of active workers (since Biggs doesn’t take this into account, he double counts these costs).

Finally, Biggs points out that past EPI research (though not my Connecticut paper) relies on BLS ECEC measures for the cost of pensions that as I noted can under- or overstate the cost of pensions depending on whether employers are neglecting or catching up on pension payments. Using ECEC pension data is a useful shortcut when it would be burdensome to compile information from hundreds of actuarial reports and when the data cover multiple years, but it’s not ideal and should be avoided when practical, such as when focusing on a specific state. At least ECEC data reflect actual outlays, which is more than can be said of Biggs’s preferred measure of pension costs, which I’ll discuss in a future blog post.