Can teachers learn online? – evidence from Armenia during the COVID-19 pandemic

ABSTRACTThe COVID-19 pandemic has forced a shift from traditional face-to-face instruction to online learning. We analyze how this shift has affected learning outcomes, using a rich data set from a financial literacy training of schoolteachers in Armenia. Online training worked well for relatively simple skills (acquiring theoretical financial knowledge) but less well than in-person training for more complex tasks (learning how to teach financial literacy to students). We also found that the deterioration of training success in the online cohort is stronger among social studies teachers than among math teachers. AcknowledgementsWe are very thankful to the editor and two anonymous referees whose comments helped us to substantially improve this paper. We are also grateful to Andrew Dustan, Kayleigh McCrary, Pedro Sant'Anna and Zaruhi Sahakyan, as well as attendees at the 2022 meeting of the Armenian Economic Association for helpful discussions, and to the staff of the Consumer Rights Protection and Financial Education Center, in particular Araks Manucharyan, for providing the data used in this article. The views expressed in this paper are those of the authors and do not necessarily represent the views or policies of the Central Bank of Armenia.Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 Even after the pandemic ends, online learning will have an important role in the future, as it can be more cost-effective (OECD Citation2020), or meet diverse learning needs (UNESCO Citation2020).2 Armenia has about 1420 primary and secondary schools, so each cohort consists of about 355 schools. The average school in Armenia is much smaller than in the United States, serving around 270 students in the relevant age range, and sending about 4 teachers to the financial literacy training.3 Almost all teachers also participated in the pre-test before the training took place. Those who did not (<10 across both years combined) were dropped from the dataset.4 The test questions contain a ‘don't know’ answer option.5 The 2020 Covid death rate was 1180 per million population in Armenia as a whole (1572 per million population in Yerevan). The 2021 Covid death rate was 1810 per million population in Armenia as a whole, and 2400 per million in Yerevan. Measured by these death rates, the pandemic was approximately 50% more severe in the in-person year 2021 than in the online year.6 Approximately 50 percent of our sample are ‘rich’ under this definition. Because teacher salaries are relatively flat, this variable depends mostly on the teacher's partner's income.7 For example, some clusters outside Yerevan are composed of schools from relatively urban areas, e.g. from the second-largest city, while other clusters contain primarily rural schools.8 Both the predicted and the actual pre-scores in 2021 are also quite close to the average pre-score of the 2020 cohort (46.3).9 Since teachers receive a salary that varies only slightly with job experience (except for teachers who are part-time), most of the difference in household income is due to the absence or presence of a second earner. The most plausible path how household income affects financial literacy is that richer households interact more with formal financial institutions (e.g. have savings accounts in banks).10 It is interesting to consider the results of a model without any demographic controls. While the treatment effect remains significantly negative (for TS and TMS; and insignificant for FLS), the size of the effect is reduced by about 10 to 25 percent, for example, for the first regression in Table 3, from −4.26 to −3.08; and, in the third regression, from −6.22 to −5.27. The direction of this change is intuitive because there are somewhat better types of teachers in the treatment group than in the control (e.g. there are more math teachers and more rich teachers in the treatment cohort). Thus, if we were not taking that change in the demographic composition into account, the size of the estimated negative online effect would be smaller.11 While positive correlations between general ability and gains from instructions have been reported (e.g. Kliegl, Smith, and Baltes (Citation1990); Kwon and Lawson (Citation2000); Verhaeghen and Marcoen (Citation1996)), negative correlations are also common (e.g. Gaultney, Bjorklund, and Goldstein (Citation1996); Traut, Guild, and Munakata (Citation2021)).