Electricity Consumption of Singaporean Households Reveals Proactive Community Response to COVID-19 progression
Research Fellow: Gururaghav Raman | Advisor: Jimmy C.-H. Peng
Understanding how populations' behaviours vary during the COVID-19 pandemic is critical to evaluating and adapting public health interventions. In this study, we use residential electricity consumption data to unravel behavioural changes within peoples' homes during the early stages of the pandemic in Singapore.
Pandemic response of the Singaporean populace
We analyse over 10,200 individual Singaporean households' electricity consumption patterns to uncover previously unknown behavioural trends during the period of January 2020 to May 2020. In particular, we assess if there are links between publicly-available information about the progression of the pandemic and the peak domestic electricity demand, which occurs in the evening; see Fig. 1. We consider the latter because an increase in the evening residential electricity consumption would likely correlate with people staying at home more during this period, or shifting behaviours by doing more activities at home rather than outside in a bid to avoid crowded public spaces. For the former, we consider two metrics—daily new COVID-19 cases, and daily recovered cases reported by the Ministry of Health Singapore.
Figure 1. Relationship between COVID-19 case data and residential electricity consumption in Singapore in January-May 2020.
The results are shown in Fig. 1. We find strong positive correlations between the peak aggregate demand and both new and recovered cases. In addition, we find that these data are co-integrated as well, suggesting that there is indeed a link between the response of the society and the progression of the pandemic.
Proactive community response before the Circuit Breaker
Under normal circumstances, an important factor influencing the Singaporean electricity consumption is the weather. To test whether weather could have been a factor causing the observed changes in the household demand, we constructed a vector error correction model (VECM), while considering weather as an influencing factor. Overall, five weather parameters were obtained and reduced to two principal components that explain > 99.9% of the variance. In addition to the weather, the daily new and recovered COVID-19 cases were fed as inputs to the VECM.
We trained this model with data from the period 23 January 2020—when the first COVID-19 case was detected in Singapore—until before 7 April 2020 when the government implemented the lockdown, called the "Circuit Breaker". With this trained VECM, we performed forecast error variance decomposition to assess how changes in each influencing factor contribute to the changes in the peak aggregate demand, the results of which are shown below in Fig. 2. Clearly, we see that while both the weather and COVID-19 case data influence the electricity demand, the predominant influencing factor is the new COVID-19 cases. Note that the government did not implement any mobility restrictions in this period; therefore, the response observed here is from people proactively responding to the progression of the pandemic. The increased peak demand implies that people either stayed in to a greater extent or performed more activities at home rather than outside in the evenings.
Figure 2. VECM results showing the relative influence of various influencing factors on the residential electricity consumption.
Impact of the Circuit Breaker
To study how the lockdown changed peoples' behaviours, we consider three different time periods, and train VECMs for each; see Fig. 3(a) and 3(b). These figures illustrate how the relative influence of the new COVID-19 cases progressively declines as time progresses. In contrast, the overall contribution of the weather components increases with time, and during the Circuit Breaker, weather becomes the most dominant influence on the peak aggregate demand. Clearly, the behaviour of residents did not change significantly during the lockdown due to the progression of the pandemic, suggesting that the residents have settled into their new lifestyles.
As an alternate to considering specific time periods, we also performed the same analysis using a moving training window that is 10 weeks long and moves forward in steps of 2 days each; these results are presented in Fig. 3(c). We find that the impact of the new COVID-19 cases remains high as long as the training window does not overlap with the Circuit Breaker period, after which its influence progressively declines.
Figure 3. Impact of the Circuit Breaker. (a) We consider three different periods in 2020, before and during the Circuit Breaker. (b) For each period, we construct and train a VECM. (c) Results of FEVD of the VECM considering a rolling time window, showing the relative influence of new COVID-19 cases on the peak demand.
Influence of demographics
Social response during the pandemic can be very different depending on demographic factors. To study if demographic factors played a role in the Singaporean context, we classified the households into 6 different dwelling types based on their average electricity consumption: 1-/2-Room HDB, 3-Room HDB, 4-Room HDB, 5-Room/Executive HDB, Private Apartment/Condominium, and Landed Property. These dwelling types exhibit clear disparities in their family composition, average number of residents in a household, and average income. The results of the classification are shown in Fig. 4(a).
We repeat our VECM analysis for the aggregate demand of households belonging to each dwelling type, see Fig. 4(b). This shows that there are no significant differences in the reactions of the households based on the dwelling type.
For more details, and implications of this work, see the full paper.
Figure 4. Influence of demographics on Singaporean's response to COVID-19. (a) Classification of the households into different dwelling types, and (b) VECM results for the different dwelling types.
It is vital for policymakers to understand how populations react during a pandemic. Here, we use domestic electricity consumption data, which can capture peoples’ daily behaviors accurately and dynamically. By studying over 10,200 Singaporean households, our analysis underscores the proactive response from Singaporean residents during the initial stages of the COVID-19 pandemic, and implies that they took steps to protect themselves despite there being no government-mandated lockdowns. Importantly, we find a surprisingly cohesive response across all demographics, a factor which may have contributed to the effectiveness of Singapore's response to COVID-19.
Supplementary data file