Early Assessment of Anxiety and Behavioral Response to Novel Swine-Origin Influenza A(H1N1)

Picture the spring of 2009. Headlines about a novel H1N1, or swine flu, are everywhere. The World Health Organization bumps up the pandemic alert level, and no one quite knows how big this will get. In that unsettled week, James Holland Jones and Marcel Salathé did something simple and gutsy: they put a survey on the internet and tried to capture our emotions and our actions in real time, not months later in a tidy retrospective. That week, while people were deciding whether to wash their hands more, skip the bus, or just keep calm and carry on. They kept it fast and broad. Adults with a networked computer could respond, and six thousand two hundred forty-nine did between April twenty-eighth and May fifth. That speed comes with a cost — an online convenience sample is never a perfect mirror of the population — but it offers a rare moving picture of the first days of public response. The average respondent was thirty-seven point six years old, about forty-seven percent were women, and roughly sixty-nine percent lived in the United States. Think educated, Western United States skewed, and very online. But also think thousands of time-stamped windows into how people felt and what they did as the story evolved day by day. What did they ask? First, feelings and beliefs. People rated their personal risk on a nine-point scale and their anxiety on another nine-point scale, from very calm to very anxious. They also judged how much control they felt they had to avoid infection. Then came behavior. The survey listed nine possible protective actions, from wash hands more often to avoid work or school to wear a mask. Each yes counted toward a protection index from zero to nine. The team also asked about information sources — internet, television, radio, print, health officials, and social networking tools — and how often people used them. And, crucially, they asked about contacts in the previous twenty-four hours, defined as face-to-face conversations or skin contact, sorted into six bins from fewer than five up to more than a hundred. That last piece matters for transmission: what you do and how many people you see are the knobs you can turn before there’s a vaccine or a drug. In the first days, concern was high. People rated swine flu as a substantial threat — second only to unintentional injury among the threats they were asked to compare — and the mood was jittery. But there was also nuance. When you look at personal risk, the distribution wasn’t just low or high. It was bimodal. Most people felt their own risk was low, but a second cluster landed right in the middle of the scale, a solid pocket of folks who felt neither safe nor doomed, just uncertain. That two-humped pattern also showed up in how empowered people felt to avoid infection. Uncertainty — not just fear — was a character in this story. Behavior tracked with those feelings. About eighty percent of respondents said they washed their hands more often, which dwarfed everything else. Avoiding school or work? Much rarer. Wearing a mask? Also rare in that early week. And where did people get their information? The internet led the pack, with television, radio, and health officials close behind as regular sources; social networking tools were used less than the other channels. That matters because it sets the stage for how messages travel and which messages nudge behavior. If you stitch those days together, you see an arc. Anxiety and the sense of immediacy peak early, then slacken as the week rolls on, and protective actions move with them. Jones and Salathé checked this not by eyeballing a line but by asking, statistically, whether day-by-day patterns in emotion and protection were just noise. They weren’t. The departures from what you’d expect by chance were big — the association between protection level and day, and between affect and day, were both highly significant, with chi-square statistics well over one hundred and p-values effectively zero. That’s the data version of saying yes, the mood and the behaviors shifted across the week in a way that’s very unlikely to be random. Now, that’s the descriptive picture. The tougher question is the one you probably care about most: when you control for who people are, what they know, where they live, who they see, and where they get their information, does feeling anxious still predict what they do? Here, the authors turn to a tool that’s both simple and powerful. They model the protection index as a set of nine coin flips — nine chances to take a protective action — and ask how the log-odds of flipping a yes change with each factor. If that sounds mathy, the key idea is straightforward: each covariate nudges the odds up or down by an amount you can add on a linear scale, and then a logistic curve turns that linear sum into a probability between zero and one. And the headline is clear. Anxiety matters, a lot, even after you account for everything else. In their protection index model, moving up one step on the anxiety scale is associated with a jump in the log-odds of taking more protective actions by about zero point nineteen. That’s bigger than the bump from most information sources and larger than the effect of being a year older. Risk perception also nudges protection upward, and so does feeling confident you can avoid infection, but anxiety is the heavyweight here. Some patterns are common sense, but it’s nice to see them quantified. Older respondents reported more protection. Men reported less; being male pulled the log-odds down by roughly zero point fifteen in the protection model and by about zero point fifty-one for the specific outcome of increased hand-washing. More contacts outside the home were associated with lower protection, stepwise, from five to ten contacts all the way to more than one hundred — a steady, monotone relationship that suggests people with busier social days were less likely to add precautions. Information exposure helped. Using the internet, radio, television, or health officials often was each associated with higher protection; print had a slight negative signal in the protection model, an interesting outlier in a week when print lagged the firehose of online and broadcast coverage. Geography also told a story. Living in Mexico — the epicenter of early attention — was associated with a large positive push on protection, with a coefficient around zero point sixty-five. Europe leaned the other direction on average in this early week, with a negative association with protection. For hand-washing specifically, Europe showed a large negative coefficient, and Australia and New Zealand did too, suggesting that local context and messaging shaped which non-pharmaceutical steps people took. That hand-washing model is worth a beat on its own. Here the outcome is a simple yes or no: did you wash your hands more often? The same cast of predictors shows up. Anxiety again looms large — a step up in anxiety corresponds to about a zero point thirty increase in the log-odds of washing more — and using mainstream information sources is linked to more frequent hand-washing. Age helps; male gender hurts. It’s a microcosm of the broader protection index, but focused on the one behavior almost everyone said they changed. There’s one more intriguing lens the authors bring in. They asked people not just about swine flu, but about perceived threats across eight health and security issues, and then they looked at how those threat perceptions hung together. Correlations were modest overall, which is what you’d expect if these are genuinely distinct ideas. But a principal components analysis — think of it as finding the main axes along which people’s fears line up — surfaced a striking second component. Swine flu, bird flu, and terrorism loaded strongly together, with coefficients right around zero point five. Three threats with heavy media coverage and deep uncertainty braided into the same psychological strand. The authors don’t overplay this, but it hints that media-salient, uncertain hazards may share a common emotional tone that spills over into behavior. If you’re wondering about representativeness, you should. The sample was not a census; it was an internet snapshot. Education was skewed high, and the Western United States was overrepresented. The team is candid about that. But the size — more than six thousand responses — and the timing — literally the first days after the World Health Organization’s alert shift — give this dataset a kind of ecological validity that’s hard to get any other way. And the picture it paints is consistent with other contemporaneous work, including a telephone survey in the United Kingdom, even if you wouldn’t bet public policy on exact magnitudes from these coefficients alone. So what do we take away? Two things I think are durable. First, emotions move fast and they move behavior with them. In the earliest days of a novel pathogen, anxiety isn’t just background noise. In this study, it’s a strong, independent predictor of the simplest, most effective non-pharmaceutical behaviors we have — like hand-washing — even after you control for age, gender, social contacts, where people live, and how they get their news. Second, information channels matter, not just for knowledge but for action. People who reported using the internet, radio, television, or health officials as regular sources were more likely to take protective steps. Social networks, at least in that two thousand nine moment, were used less and didn’t show the same punch. There’s also a quiet systems insight here. The number of people you see in a day and the choices you make — wash more, avoid crowds, skip a commute — are the behavioral levers that shape transmission before vaccines arrive. When those levers are connected to emotional states that can swing over a few days, the early communication environment becomes part of the epidemic’s dynamics. That’s not a causal claim from this dataset alone, but it is a plausible mechanism supported by the pattern Jones and Salathé captured. Finally, a word about speed. This was approved by an Institutional Review Board, or IRB, and launched the day the alarm bell got louder. Within a week it had mapped a rise-and-fall arc in worry and action. We learned in two thousand twenty how much those first days matter. If you can see behavior changing in real time, you can calibrate messages, amplify the right channels, and give people clear, doable actions when their attention is highest. That’s not science fiction. It’s what this two thousand nine study actually did: measure, quickly, how people were feeling and acting, and show that those early feelings — especially anxiety — were tightly linked to what people did with their hands, their schedules, and their social lives. In a pandemic, that’s where the curve starts to bend.