Risk creep: outings during a pandemic semester

If people feel the risk is high, do they go out less? We followed students for eight weeks to see how risk perception, seeing others, and time related to what they did.

My role: Contributing author (third of three)
Status: Under review. Working paper, July 2026

Logg, J. M., Tinsley, C. H., & Leitao, M. R. (2026). Risk creep: A COVID-19 longitudinal field study. Working paper, July 3, 2026.

Who: 304 returning students in three waves of 38, 144 and 122. 14 of 303 tested positive.
Design: A baseline survey plus seven weekly pulses, eight weeks in all. Counts of outings in each category.
Data: Not public. Figures are rebuilt from the printed tables, and a derived percentage is labeled.

Try it: dial the risk

Skip to the results

If students felt more at risk from the virus, you might expect fewer outings. Pick an activity, turn up how risky people feel, and see what the paper’s model implies. Guess first, then drag.

The paper reports the coefficient b on the log scale. I turn it into a percent change for a chosen step. The interval uses b ± 1.96 standard errors. Both are derived, not printed. The reliable/not reliable call is the paper’s own. Perceived risk is measured in points on the survey’s scale.

Main findings

The more risk people felt, the more outings they reported

304 returning students reported their outings and how risky they felt the virus was, once at the start and then weekly for seven weeks. Each person was a repeated measure, about 1,634 observations in each model.

For discretionary outings, higher perceived risk went with more outings, not fewer. The link was clearest for large events (b = .020) and exercise (b = .005).

b = .020

perceived risk and large events

About 22% more large events per 10 points of perceived risk, my transformation of the coefficient.

Seeing others go out went with going out

The number of others people saw doing an activity went with doing that activity more. This held for exercise, small social events, large events, food outings and school.

Errands, a necessity, did not show it.

b = .006

others observed and large events

Reliable at p < .01.

Habituation does not explain it

Outings fell week by week for every activity (the week coefficient is negative in all six models). That rules out a simple story where people just got used to the risk and went out more.

School outings fell fastest (b = −.206).

6 of 6

activities with a falling trend over the weeks

Four of six reliably so.

The evidence

What went with outings

What to look for. Filled circles are reliable results. The top two rows are mostly filled for discretionary outings. The bottom row shows time moving in one direction.

Reliable (p below .01)
Weaker (p below .05)
No reliable association

Activity	Exercise	Small social	Large events	Food	Errands	School
Your own perceived risk	.005	.002	.020	.003	.002	−.001
Others observed doing it	.001	.001	.006	.003	.000	.005
Week of the study	−.025	−.045	−.035	−.018	−.050	−.206

Coefficients from count models, as printed in the paper’s Tables 1 and 2 (Model 2). Lagged outings were controlled.

Size of the perceived-risk link, in percent

What to look for. Large events stand apart. Errands and school sit near zero.

This is my transformation of the printed coefficients: the percent change is 100 × (exp(10 × b) − 1), with the interval from each printed standard error. The paper reports the coefficients.

Trend over the weeks

Table 1. Perceived risk, others observed and week

What to look for. The first row is the main result. The last column shows the largest decline over time.

Predictor	Exercise	Small social	Large events	Food	Errands	School
Perceived risk	.005***	.002*	.020***	.003**	.002	−.001
Others observed	.001**	.001***	.006***	.003**	.000	.005**
Week	−.025*	−.045***	−.035	−.018	−.050***	−.206***

* p < .05, ** p < .01, *** p < .001. From Tables 1 and 2.

What this means

Higher perceived risk did not go with fewer outings. For discretionary outings, risk perception went with more outings. Seeing others do the same thing went with more of it too.

Limits. This is a working paper. The results are associations in one student group, so I cannot say what caused what.

How this page was made

Estimates are from the printed tables. The percent figure and the intervals on the coefficients are derived from the printed standard errors. I have not re-run the models.