Hey, local newspapers: Want to try to predict which of your subscribers are going to stick with you — and keep paying — no matter what? New research out of the Medill Local News Initiative at Northwestern suggests that creating a habit is the most important thing to focus on: The frequency of reading local news is “the single biggest predictor of retaining subscribers — more than the number of stories read or the time spent reading them.”

In some cases, in fact, “high rates of story reading and time spent per story” were actually associated with people dropping their subscriptions. Yes, the researchers say, this is indeed a “puzzling surprise” (more on it below).

Folks from Medill’s Spiegel Research Center, led by research director Edward Malthouse, analyzed 13 terabytes of anonymous reader and subscriber data from the Chicago Tribune, Indianapolis Star, and San Francisco Chronicle. In doing so, they were able to “trace anonymized, individual behaviors of people who kept and cancelled digital subscriptions.” The team had about a year of data from the Chronicle, nearly three years of data from the Tribune, and a little over two years of data from the Star.

For all three of the papers, regular reading habits were strongly correlated with subscriber retention. “Regular” here means daily readers, not people who come a couple times a month. So how can newspapers build these habits? One recommendation is email newsletters that readers can skim. And customization — the kind that big newspapers are already doing — is going to be key.

News outlets now have the opportunity to track when a subscriber has read five of the last seven stories on the City Council. When the eighth story is published, why not email the story or a link to the subscriber? The answer is that hardly any local news operations have the technological wherewithal to do this type of personalization. But this is just the kind of tactic that would serve customers most effectively.

At both the Chronicle and Star, the number of pages viewed per day and the amount of time spent per pageview were correlated with higher rates of subscription cancellation (there was no such correlation for the Tribune). This is weird, and Malthouse admitted he’s “wrestling” with it and hopes it will be addressed in future research. He offered a few possible explanations:

Maybe your coverage of certain things just isn’t very good. And if I’m just reading a headline, I get updated on it and it doesn’t bother me. But when I actually read the article, I realize that whatever this thing I’m interested in, I get way better coverage elsewhere.

Number 2 is…there’s some sort of generational thing. We’re just being trained now to skim, with Twitter and Facebook and all these things showing me lots of headlines without much depth…So there’s this cultural shift happening.

Number 3…either I’m overloaded or I’m reminded of all the negativity or the cynicism when I read the stories. So, if I’m just reading headlines I don’t get that. If I’m actually reading all these things, then I get ticked off because it just depresses me.

Malthouse also thinks that The Washington Post and New York Times “may already know about this. They have a lot of touch points — they update me. You don’t have to read this whole story if you’re tracking this issue of Trump and whatever — here’s what’s new about it in two sentences.”

In other words, skimming isn’t something that local newspapers should fear, the researchers found. “Becoming valuable to subscribers doesn’t necessarily mean taking up a lot of their time. In fact, readers who consumed a lot of stories and read them deeply were not more likely to remain subscribers.” Instead, local newspapers should focus on non-commodity coverage and find areas they can really own.

Research will continue (one future area to look at: Are subscribers who sign on at a deep discount — think $0.99 for the first month — much more likely to drop off later?) And the above research will also be published “in an academic journal at a later date,” according to Medill.