Fighting the learning loss: Evaluating C-SEF for university students and staff.

2022-06-01

2022-06-30

Posterior probabilities of infection, Perceived stress scale, Subjective well-being, performance and productivity.

posterior probabilities of infection: For each individual, we hypothesise a uniform prior probability of infection and corresponding probability of being healthy. After each test, we update the probability of infection according to test results; perceived stress scale: measured via the validated 4-item Perceived Stress Scale by Sheldon Cohen (1994); subjective well-being: we use a variation of the European Quality of Life Survey measure of subjective well-being, using a life/subject evaluation approach; Performance and productivity: we use a composite score for the evaluation of performance and productivity, based on a 5 item self-reported approach from poor to high.

We implement a mechanism to ensuring the safety of an educational institution (C-SEF) on the academic population at IPICYT in San Luis Potosí, Mexico. We randomly assign individuals into a treatment and a control group. Subjects in the treatment group follow a pool-testing protocol and are allowed to freely use the facilities of the university so long as they test negative. Individuals in the control group are expected to work remotely and only use the facilities for emergencies [the definition of an emergency is at the discretion of the head of department to which the student/staff belongs]. The physical isolation of treatment group and control groups is essential for our protocol to work; both to avoid health spillovers and disentangle psychological dynamics. The IPICYT campus lends itself, for the most part, to a two-group randomisation approach. The campus consists of two similar buildings for research in natural sciences, two similar buildings for research in computer science and mathematics, one building for classes, and one building for administration. The individuals participating in our trial are randomly assigned to the treatment and control group. Note however that some individuals in the control group will still work on-site based on exceptions. Therefore, we use the following strategy: one of the natural science buildings is randomly assigned to be treated and one of the CS and math buildings is randomly assigned to be treated. Crucial for this approach to be effective is that individuals in both buildings are comparable. Given IPICYT's reports about their staff, we know that both research students and researchers are assigned to work in each of the buildings contingent only on their academic discipline, based on no individual characteristics important for the evaluation of our framework. Hence, we may consider the assignment as pseudo-random. Nevertheless, we further collect a number of covariates to conduct a balance analysis [Should there be a clear imbalance, we will subscribe to a post-experimental matching approach as proposed by Bruhn and McKenzie (2009)]. Within the staff building, we also need to physically separate the space allocated to treatment and control group. Our strategy here is to separate the four-storied building by floors. We randomly select either the top two or the bottom two floors for treatment, and the remaining two floors for control. Similar to the other two buildings, we accept that the assignment is pseudo-random. Students can be assigned to classrooms depending on availability and suitability for the course taught. Moreover, students in the control group will exclusively participate in online learning. Therefore, each individual in the student population is assigned randomly to either treatment or control.

Individuals assigned to the treatment group are invited to participate in the C-SEF protocol. All information about the institute-wide study is shared with the IPICYT population via an online communication campaign co-produced by the principal investigators and the communications department of the Institute. Consent is obtained from each individual via a form, which precedes a survey that collects baseline data. There exists no natural disincentive to participate in the study, and on the contrary, participation may indirectly give participants more freedom of movement, so we confidently estimate study compliance to be at nearly 100%. The C-SEF protocol uses a utility-weighted selection algorithm that groups individuals for pool testing. Individuals who test negative are allowed to attend their work or study activities on campus. Individuals who test positive are asked to isolate and work remotely until their next test. For a more detailed explanation about the testing mechanism please refer to the Pre-Analysis Plan document.

Cluster randomization (computerized).

Unit of cluster: academic discipline (buildings); unit of analysis: students and staff.

Yes

1 university population with 5 clusters (disciplines).

500 students and staff

250 students and staff per treatment arm

To perform the clustered-randomised power calculation we use an ICC of 0.22 and a cluster standard deviation of 0.5, as per the literature on educational institutions. Moreover, we assess power on the basis of odds ratio for the likert-type outcomes mental health and productivity, which can be thought of as follows: an odds ratio of 1.48 can be considered a small effect size or the equivalent to a Cohen's d of 0.2; an odds ratio of 3.45 can be considered a moderate effect size or the equivalent to a Cohen's d of 0.5; finally, an odds ratio of 9 can be considered a large effect size or the equivalent to a Cohen's d of 0.8. Our final results are as follows: probabilities of infection with n = 500 and alpha = 0.05, has a cohen's d of 0.47 and power of 0.80; mental health outcomes with n = 500 and alpha = 0.05, has an odds ratio of 2 and power of 0.80; productivity and performance measures with n = 500 and alpha = 0.05, has an odds ratio of 2 and power of 0.80. For a more detailed explanation please refer to the Pre-Analysis Plan.

Evaluating C-SEF for university students and staff

Large-scale testing is crucial in pandemic containment, but resources are often prohibitively constrained. We study the optimal application of pooled testing for populations that are heterogeneous with respect to an individual's infection probability and utility that materializes if included in a negative test. We show that the welfare gain from overlapping testing over non-overlapping testing is bounded. Moreover, non-overlapping allocations, which are both conceptually and logistically simpler to implement, are empirically near-optimal, and we design a heuristic mechanism for finding these near-optimal test allocations. In numerical experiments, we highlight the efficacy and viability of our heuristic in practice. We also implement and provide experimental evidence on the benefits of utility-weighted pooled testing in a real-world setting. Our pilot study at a higher education research institute in Mexico finds no evidence that performance and mental health outcomes of participants in our testing regime are worse than under the first-best counterfactual of full access for individuals without testing.

Finster, Simon, Michelle González Amador, Edwin Lock, Francisco Marmolejo-Cossío, Evi Micha, and Ariel D. Procaccia. "Welfare-Maximizing Pooled Testing." arXiv preprint arXiv:2206.10660 (2022).

https://doi.org/10.48550/arXiv.2206.10660