During a public health crisis, getting the research right is paramount
Beware the confundus charm
It started with a tweet.
A new peer-reviewed Utilities Policy article on water utility ownership, low-income households, and shutoffs? From a pair of professors at major research universities? This was right up my alley!
The paper’s title—Does public ownership of utilities matter for local government water policies?—is intriguing. Water system ownership, regulatory policy, and especially shutoffs are enormously important and notoriously difficult to analyze due to data difficulties. The tweet and top highlight finding were provocative: “Cities and towns with government-owned utilities shutoff customer drinking water less.” These are headline-grabbing claims, sure to draw the attention of water sector leaders and policymakers looking for ways to tackle an ongoing public health crisis. I dove into the paper immediately, excited about what I might find!
Unfortunately, the study is deeply flawed. Few readers—even within the academy—have the appetite to get into the methodological details necessary to understand what the data really show. My deep dive revealed that the article’s authors came to profoundly incorrect conclusions. Last week I emailed them with my concerns. This post is my attempt to clarify what the article obfuscates and to set the record straight.
The claims
From tweet to title to text, Homsy & Warner’s article promises to explore the role that “local government utility ownership play[s] in meeting equity and environmental goals.” The study’s literature review frames its goals as an inquiry on “Publicly owned versus privately owned water supply” (section 2.3), with interest in public/private differences in “equity” (2.1) and “environment: water resource management” (2.2). Specifically, they’re interested in whether publicly owned water systems provide more protections against shutoffs for low-income households and greater water conservation. The implied comparison is with private water utilities.
The empirical analysis uses a 2015 ICMA survey sent to municipal chief administrative officers. The sample included all municipalities, towns, and counties with populations over 25,000 and a sample of smaller communities of populations between 2,500 and 25,000. The survey yielded 1,897 responses for a 22% response rate.* Respondents were asked whether their governments own water utilities. If the respondent said yes, the community was coded as having a “publicly owned utility” or “Government-owned utility”; if the respondent said no, then it was coded as having a “not publicly owned utility.”
When "Government-owned" does not necessarily mean owned by a government.
The survey also asked questions about policy, including whether the city or county had taken action to "protect low-income households from water service shutoff.” The answer to this question is Homsy & Warner’s only measure of shutoff protection. They don't actually count shutoffs. About 8% responded that their governments provided shutoff protections.
These survey data were analyzed with a series of logistic regressions to predict the likelihood that a community has water shutoff protections as a function of public ownership and a series of other variables. Ownership emerges as a very strong predictor, leading Homsy & Warner to declare that: “if a utility is government-owned, the municipality is about two times more likely to have water shutoff protection policies.”
They conclude that “ownership matters, as communities with publicly owned utilities appear more inclined to protect residents from water service shutoffs.” That would be a big, important finding—if the data supported it. The trouble is that the data don’t show anything of the sort.
Where it all goes wrong: misleading measurement
The first hint that something was amiss was that just 55% of Homsey & Warner’s sample were publicly owned utilities, with the authors implying that the other 45% were served by private utilities. That struck me as extremely low, since about 85% of Americans get their drinking water from a local government, with about 15% served by private, investor-owned firms. How could the ICMA sample be so grossly skewed in favor of non-government utilities?**
The answer is in the way that Homsy & Warner code “public ownership.” Recall that the ICMA survey was sent only to counties and municipalities and that respondents who reported that their governments own water systems were coded as “publicly owned.” What’s missing is special districts.Special districts are local governments with narrow functions and limited powers; there are tens of thousands of such districts across the U.S.. The city and county officials responding to this survey who get their water from special districts would answer "no," and then Homsy & Warner would code the community's water system as "Not Publicly Owned."
Water districts, Potter. Remember the water districts.
Consider, for example, Central Arkansas Water. Headquartered in Little Rock, Central Arkansas Water serves a population of more than 450,000, including the cities of Little Rock, North Little Rock, Alexander, Sherwood, and Wrightsville, along with parts of unincorporated Pulaski County. Survey respondents from Pulaski County and all five of these cities would report that their governments did not own a water system. Homsy & Warner would code their water systems as "not publicly owned," even though a government--Central Arkansas Water--supplies their water.
Around 30% of the local government water systems in the United States are owned and operated by special districts. These special districts are definitely local governments. Without accounting for special district utilities, we cannot infer anything about public/private water system ownership from these data.
Sins of inference
But it gets worse. The study’s first highlighted claim is that “cities and towns with government-owned utilities shutoff customer drinking water less.”† Thing is, this study doesn’t measure shutoffs. It measures whether a municipal or county government has a policy to protect low-income households from shutoffs. Reliable shutoff data are notoriously hard to find, which is why serious research on the subject is rare. A reader who doesn’t get into the empirical weeds wouldn’t notice that data don’t support this claim.
Apart from problems with coding ownership and counting shutoffs is the question of what this all means for water governance. What should we infer from the fact that a municipality that owns a water system is more likely to have water shutoff protections than a municipality that does not own a water system? Homsy & Warner conclude that “communities with publicly owned utilities appear more inclined to protect residents from water service shutoffs.” That’s a bit like finding that people who own cars are more likely to have jumper cables than people who don’t own cars. Would we then infer that car owners care more about their families' transportation than transit riders?
Why it matters
Policymakers, advocates, regulators, and utility managers are looking for answers to a historically tough challenge in the COVID-19 pandemic. Researchers everywhere are working hard to find ways to help protect public health.
We certainly need to understand water shutoffs and how to prevent them. But the stakes are too high for policy researchers to play fast and loose with data and inferences. This isn’t an abstract, theoretical squabble over the literary interpretation of Hamlet. Real policies to protect real people in a moment of real crisis are on the line, and our communities need valid findings. The urgency of pandemic only heightens the need for rigorous, responsible policy research.
*It does not appear that Homsy & Warner adjusted their estimates to account for sample stratification or non-response bias.
**With just a 22% response rate, some of the issue is probably nonresponse bias. But it’s unlikely that nonresponse accounts for a 30% gap in share of publicly owned utilities.
†In this sentence and throughout the article, Homsy & Warner are vague about what publicly owned utilities are being compared with.