The latest Consumer Price Index report showed inflation slowing in October, due in part to a supposed 34% drop in the cost of health insurance from the same time last year. That left many people scratching their heads, since you’d be hard pressed to find anyone whose premiums dropped at all, let alone by that much. Is someone cooking the books at the Bureau of Labor Statistics?

The short answer is no, but the really crazy thing—which few people understand—is the BLS isn’t reporting what you pay in health insurance premiums. That’s why these figures don’t accurately reflects what American families have experienced over the last year. Meanwhile, a plethora of other data point to premiums increasing over the last 12 months, not falling.

It turns out that health insurance is a very difficult service to measure, in part because it is inconsistent over time. Imagine a new drug becomes available, and your health insurance now covers it. This new expansion of coverage means your health insurance is not directly comparable to what it was before, and there needs to be an adjustment for this change in quality.

Additionally, health insurance is a service distinct from medical care (like a doctor’s services) and medical commodities (like prescriptions), and the price changes for medical care and medical commodities are already accounted for in the CPI.

>>> How Bidenomics Led to the Most Expensive Thanksgiving Ever

This poses an additional challenge for the BLS in measuring the cost of health insurance, because if they simply measure gross health insurance premiums, they’d mostly be counting various costs of health care, not the service of providing the insurance.

What BLS is essentially trying to measure is the difference between what the insured pays in premiums and the value of the benefits he or she receives from the insurance. That difference is called net premiums, and BLS measures it by looking at the retained earnings of health insurance companies. This methodology is highly imperfect, and the last several years have exposed the flaws in spades.

Higher costs, partly from inflation, have decreased health insurers’ retained earnings over the last year, even as insurance premiums have been rising. Additionally, temporary changes in the health insurance market that followed in the wake of the pandemic caused large and unusual changes to profit margins in the industry. As those effects subside, retained earnings have fallen precipitously.

In the October CPI report, this manifested as a 34% decline in health insurance costs over the prior 12 months, even though countless other indicators show health insurance has become more expensive during the last year—not just in terms of higher gross premiums but in terms of less value to consumers.

The BLS recently made changes to its methodology of calculating health insurance costs in the CPI, but that dealt largely with how frequently the retained earnings data are updated and smoothing out sudden changes in the statistics. They still use the retained earnings model and the problems with indirect measurement of health insurance costs remain.

>>> Time to Turbo-Charge Health Care Price Transparency

The sub-index for the price of health insurance is now down to its August 2018 level. So, despite considerable empirical evidence showing that the cost of health insurance is significantly higher today than five years ago, the BLS is counting these costs as being equal.

To be fair, the alternative methodologies for measuring cost changes to health insurance are also highly imperfect, so the BLS is left to choose the least bad alternative. For example, counting gross premiums paid by the insured would capture the cost of the insurance, but then the medical care and medical commodities paid for by health insurance would effectively be double counted in the CPI.

Regardless of which flawed methodology is used, people need to be aware of how to interpret the data published by BLS. Economic statistics can be highly valuable for research and evaluating the effects of public policy, but we’d do well to remember the three kinds of lies as described by British prime minister Benjamin Disraeli: lies, damned lies, and statistics. Take everything with a grain of salt.