The impact of the intervention is calculated for various periods, namely: ... to note that these treatment effects were realised at the end of a period which has seen.
Green Nudges in the DSM toolkit: Evidence from Drought-Stricken Cape Town Kerri Brick, University of Cape Town, Martine Visser, University of Cape Town
December 2017 DRAFT MANUSCRIPT
Abstract
Green nudges are an additional demand-side-management tool that can be used to reinforce government policy and improve policy outcomes. In this context, social comparisons have been shown to be particularly effective in the environmental space. Against the backdrop of a severe and sustained drought, this study estimates the impact of a social-norm inspired behavioural nudge implemented by local government in Cape Town, South Africa. The insert – which combined social comparison with a punitive threat – was sent to households consuming excessive amounts of water. The letter admonished recipient households for unacceptable, excessive and above-average water consumption. This paper estimates the treatment effect for a subsample of these households. The impact of the intervention is calculated for various periods, namely: 2-months, 3-months, 4-months, 5months, 6-months and 7-months after receipt of the letter. Throughout this timeline, the treatment effect remains relatively consistent: treated households reduced consumption by around 1.1 kl per month for the first six months after receipt of the letter; the treatment effect reduces to 0.9 kl seven-months post intervention. We use some back-of-the-envelope calculations to contextualise this saving against the current crisis. Assuming an average reduction of 1.1 kl per month over seven months implies a saving of 7.7 kl per households. Under Level 5 restrictions where individual consumption is capped at 87 l per person per day, this saving is equivalent to the monthly consumption of a three-person household. Households that consumed greater than 50 kl in January 2017 were eligible to receive the letter. In terms of the percentage reduction in water consumption achieved by the intervention, when only considering eligible households (i.e. a sample of households consuming at least 50 kl in January 2017), the letter induced a consistent 3% reduction in consumption (decreasing to 2.5% seven-months post the intervention). The implication is that the nudge was extremely effective in eliciting a response in a group of households that were price insensitive and slow to respond to higher tariffs and physical restrictions. Moreover, when considering the treatment effect against the entire sample (eligible and non-eligible households), the letter generated a consistent 6% reduction in consumption (decreasing to 5.3% seven-months post the intervention). It is important to note that these treatment effects were realised at the end of a period which has seen increasingly stringent physical restrictions and tariff hikes: prices in the 5th and 6th tariff blocks increased by 293% and 689%, respectively. As the warning letter induced additional water savings on top of the savings generated by these more conventional demand-side-management tools, green nudges present as an effective tool for reinforcing government policy, particularly in a time of crisis. Index Terms—Behavioural Economics, Social Norms, Demand Side Management, Water Conservation, Drought
I. INTRODUCTION Cape Town is besieged by sustained drought. At the time of writing, as South Africa heads into the dry summer season, dam levels stand at a dire 37% of capacity, with the last 10% being unusable. Against this background of water austerity, the local municipality has adopted a number of demand-side-management (DSM) interventions to reduce residential water consumption. These include the introduction of physical restrictions on water usage, substantially higher tariffs and a widespread media offensive. In addition, in a nod to less traditional DSM tactics, the municipality has nudged households consuming excessive amounts of water with a social-norm inspired intervention. This social feedback was incorporated into a letter of warning that was sent to residential households consuming over 50kl of water in January 2017. The letter informed errant households that their consumption was above the average of 20kl, cautioned against excessive and unacceptable water usage and threatened the household with the installation of a flow restrictor. This intervention is emblematic of the growing academic and political traction around the use of nudges in the environmental space; and a prime example of how local municipalities can use nudges to reinforce government policy, particularly in time of crisis. Serving as a case study of how policy makers can integrate green nudges with traditional DSM initiatives to both encourage normative behaviour and leverage behaviour change, this paper estimates the overall impact of this intervention on residential water consumption. Traditional DSM’s focus on price changes and information disclosure as levers for behaviour change is a legacy of the neoclassical economic model of unbounded rationality. In contrast, behavioural economics emphasizes bounded rationality – with the applied literature showing that decision-making is affected by what is salient (focal), the current default, the social norm and other-regarding preferences such as a preference for fairness (Allcott & Mullainathan 2010; Croson & Treich 2014; Chetty 2015). While bounded rationality and cognitive limitations can reduce the efficacy of government policy, behavioural nudges small changes in context which do not alter economic incentives – increase the policy tools available to policymakers and can reinforce government policy (Thaler & Sunstein 2008; Chetty et al. 2009; Nyborg et al. 2016). The ubiquitous policy goal of increasing retirement saving rates provides an example of how nudges can be used to reinforce policy and improve policy outcomes. Specifically, while governments typically incentivise retirement savings via subsidies (tax concessions), a large proportion of individuals prove unresponsive to price signals. In fact, recent research around the impact of subsidies on saving behaviour found over 80% of the sample to be passive savers who are unresponsive to changes in financial incentives (subsidies) (Chetty et al. 2014; Chetty 2015). In terms of the role of nudges, a growing literature has shown that a default opt-out system, where employees are automatically enrolled into a savings plan and are given the option to stop contributing, substantially increases participation (Thaler & Sunstein 2008). Importantly, Chetty et al. (2014) show that default plans increase total saving: employees do not fully offset pension contributions by reducing saving rates in other accounts. Moreover, default options are a mechanism to target passive savers who are less responsive to price signals. Social comparisons have also been found to be effective at promoting behaviour change with the literature showing that people behave normatively when the social norm is made apparent (Nyborg et al. 2016). Social norm comparisons, widely applied in the environment space, have been particularly effective as
green nudges. In this context, a household’s consumption is typically compared to the average for their neighbourhood. Allcott (2011) evaluates a social-norm intervention run by US energy company OPOWER. With a sample of 600 000 residential households, treated households received a Home Electricity Report which compared their electricity usage to the average in their neighbourhood. The intervention reduced electricity consumption by 2%. Ayres et al. (2012) analyse the impact of the OPOWER experiment in a different US city and find that the feedback reports significantly reduced energy consumption by between 1.2-2.1%. Furthermore, they find this effect to be sustained for between 7-12 months (depending on the utility company). Ferraro & Price (2013) utilise norm-based messages to reduce residential water consumption in Atlanta, USA. The authors found that, in the month after the intervention, households treated with the strong social-norm treatment used around 5.6% less water relative to the control group. Datta et al. (2015) conducted a randomized field experiment in Costa Rica. Depending on the treatment, residential households were either compared to the average household in their neighbourhood or city. Households in the “neighbourhood” treatment reduced average monthly water consumption in the first two months after the nudge by between 3.7 and 5.6% of control group consumption. While this discussion shows that social comparisons are effective at changing behaviour, is it possible to contextualise the magnitude of their effect relative to more traditional DSM measures? Allcott (2011) estimates the 2% decrease in energy consumption to be equivalent to a short-run electricity price increase of between 11-20% or a long-run price increase of 5%. Ayres et al. (2013), who find a 1.2-2.1% effect size, speculate that the 3-7% tax increase that would yield similar levels of short-term energy reduction would not be politically feasible. Finally, Ferraro & Price (2013) estimate that their social-norm effect is equivalent to a 12-15% price increase. The implication is that social-norm comparisons can have effect sizes that are comparable to large price changes. This is a powerful benefit of nudges: their impact can be equivalent to large price changes that would not be politically feasible. Nudges can be both appropriate and effective at opposite extremes of the income spectrum. Firstly, nudges do not feel punitive and regressive to low-income households in the same way that price increases often do. In the local context of Cape Town, tariffs have significantly increased in recent years with the escalating drought crisis. For example, since 2010, tariffs have increased by 440% and 1470% in the second and third tariff blocks, respectively. Even when considered on a shorter timeline, tariffs have increased substantially; between December 2015 and July 2016, tariffs in the second, third, fourth, fifth and sixth tariffs blocks increased by 60%, 64%, 86%, 293% and 689%, respectively. Moreover, and of even greater significance, in July 2017 the local municipality historically broke with the policy of free basic water for consumption in the first tariff block. While low- and even middle-income households adjust to real increases in their water bills, nudges present as non-pecuniary measures for promoting behaviour change. Secondly, and moving to the opposite end of the spectrum, a number of studies have found that nudges are most effective among wealthy households who are least price sensitive (Ferraro & Price 2013; Allcott 2011; Datta et al. 2015). Ultimately, behavioural economics expands the set of policy tools available to policy makers to influence behaviour. For example, Ferraro & Price (2013) who tested the impact of technical advice (tips on how to save water) in addition to social comparisions, found the provision of technical advice to reduce
consumption by around 1%. However, combining techical advice with a social comparision more than quadrupled the effect – with households reducing consumption by nearly 5%. Against this background, this paper estimates the impact of a DSM initiative that integrates a social comparison with a punitive threat (installation of the flow restrictor). While it is not possible to unbundle the effect of the social-norm comparison from the warning, we exploit the fact that a subset of eligible households did not receive the intervention and estimate the cumulative effect of the intervention. The analysis thus illustrates how one progressive municipality has been able to combine green nudges with more traditional DSM measures. The results indicate that the intervention decreased water consumption in treated households by around 1.1 kilolitres per month relative to control households, decreasing to 0.9 kl seven-months after households first received the insert. When only considering the subsample of eligible households (i.e. those households consuming a minimum of 50 kl in January 2017), the intervention generated a consistent 3% reduction in consumption (with the treatment effect decreasing to 2.5% sevenmonths post intervention). When considering the entire sample (eligible and non-eligible households), the intervention consistently reduced consumption by 6% (decreasing to 5.3% at the seventh-month mark). This is a significant finding given that, as previously mentioned, tariffs in the 5th and 6th tariff blocks have increased by 293% and 689% since July 2015, respectively. The implication is that this intervention was able to effectively elicit a response in a group of households that were price insensitive and slow to adhere to the physical restrictions. The document proceeds as follows: Section 2 provides further background around local government’s response to the drought. The data and estimation method is described in Section 3. Results are provided in Section 4 and Section 5 concludes with a discussion.
II. DROUGHT CRISIS IN CAPE TOWN As discussed, Cape Town is besieged by sustained drought. Over the past few years dam levels have systematically decreased: from 100% of capacity in 2014 to 37% at the time of writing (City of Cape Town 2017). Between January 2016 to today, water restrictions have been systematically increased from level 1 water restrictions (normally in place) to level 5 (the highest level, applicable under extended drought conditions and periods where dam levels are critically low). Under level 5 restrictions, individuals are expected to consume no more than 87 litres per person per day. In addition, household consumption is capped at 20 kilolitres per month – irrespective of household size, with the local municipality warning that contravening households will be subject to excessive fines. Furthermore, the use of municipal drinking water for anything other than essential use is prohibited (this means that any domestic irrigation, use of water features or topping up of swimming pools is prohibited). As previously mentioned, throughout this period, residential water tariffs have also been increasing. Cape Town has a stepped tariff system with six tariff blocks. While consumption in the first tariff block (0-6kl) has historically been free, the local municipality introduced a charge of R4.56 per kl in July 2017. In addition, tariffs in the remaining tiers have increased substantially: water charges in tariff block 2 (6-10.5kl) have increased from R11.07 per kl in July 2015 to R17.75 in July 2017 (60% increase); over the same
period, tariff rates in tariff block 3 have increased from R15.87 per kl to R25.97 (64% increase), tariff block 4 has seen an increase from R23.51 per kl to R43.69 (86% increase), in tariff block 5 rates have increased from R29.03 per kl to R113.99 (293% increase) and, finally, tariff block 6 has seen an increase in rates from R38.30 per kl to R302.24 (689% increase!). In addition to rising tariffs and increasingly stringent restrictions, the municipality has adopted other demand-side measures. For example, since July 2015, the local municipality has led a widespread information and social media campaign. In addition, between November 2015 and June 2016, the local municipality rolled out a behavioural intervention with a sample of 400 000 residential households. During this randomized experiment, several nudges were tested, including: loss and gain framings, social-norm comparisons and social recognition for pro-environmental behaviour. More information on this is provided in Brick et al. (n.d.). In addition, in a bid to ratchet up social pressure on non-compliant households, in February 2017, the municipality published a list of the top 100 water users by street (the list provided the street name of the offending households – but not the family name or street number – as well as the level of monthly consumption). This list was published by numerous media outlets with titles such as “Named and Shamed” and “Cape Town’s water wasters…”. Finally, in a response to a spike in consumption in January 2017 (see Figure 1), the City sent letters to households consuming greater than 50kl in February and March 2017. The letter referenced these households' "unacceptable" and "excessive" consumption, noted that their consumption was outside the norm of 20kl per household and, finally, threatened households with the installation of a watermanagement device should households not reduce consumption levels. The full text of the letter is provided in Appendix A. This paper analyses the impact of this intervention. Note that while it is not possible to isolate the effect of the social comparison from the threat of sanction against non-compliant households, the analysis provides a general indication of the benefits using green nudges to reinforce local policy.
III. DATA AND ESTIMATION The municipality sent letters to over 7000 households between February and March 2017. This analysis is conducted with a small subsample of residential households who were eligible to receive the letter. The treatment effect is estimated using monthly consumption data obtained from the local municipality. In addition, by merging this data with other administrative data obtained from the municipality, we are able to link households with their property values (acting as a proxy for income). In terms of the sample, this analysis is restricted to only those households that were in the control group of the large-scale randomised behavioural intervention conducted the previous year (Brick et al. n.d.). The sample further excludes households that receive their utility bills via email – as the researchers have monthly data only for households that receive their bill via the post. Finally, the sample excludes estimated and final readings and households with extended (and thus generally problematic) billing periods. After
these exclusions, the sample consists of 23 571 households and contains consumption information over the period June 2014 – October 2017. As expected, only a small portion of these remaining households were eligible to receive the letter in Feb/March 2017. For example, in January 2017, while monthly consumption varied between 7 kl and 78 kl, around 95% of households consumed less than 50kl (the reader is referred to Figure 1). As such, the sample analysed here consists of 655 households that consumed 50kl or more in January 2017 and were eligible to receive the letter. Of these households, 380 received the warning letter from the local municipality between February-March 2017. The households that consumed 50kl or more in January 2017 but did not receive a letter (275 households) act as the control group. Figure 1. Distribution of monthly consumption in January 2017
Note that the level of consumption in January 2017 was a necessary but not sufficient condition for receipt of the letter. The local government applied additional selection criteria – for example indigent households were not sent the insert (and have thus been excluded from this analysis). While the full set of selection criteria for treatment is unknown, it is evident that selection into treatment was not completely random. To account for this non-random allocation into treatment, we estimate a difference-in-difference (DiD) model, comparing observed consumption of treated households (who received the insert in Feb-March 2017) relative to non-treated households. Instead of simply taking a before and after estimate of the impact of the intervention, DiD compares the before-and-after outcomes for the households that received the insert (the first difference) and the before-and-after outcomes for the households that did not receive the insert but were exposed to the same set of economic and environmental conditions (Khandker et al. 2010). Then the
difference between the difference in outcomes for the treated and the comparison groups is calculated. In a two-period setting, the average program impact is estimated as follows: . . ( ( 𝐷𝐷 = 𝐸 𝑌%&' − 𝑌%&* 𝑇' = 1 − 𝐸 𝑌%&' − 𝑌%&* 𝑇' = 0
(1)
where t = 0 before the intervention is implemented and t = 1 after implementation, YtT is the outcome for treated households at time t, YtC the outcome for non-treated households at time t, T1 = 1 denotes treatment and, finally, T1 = 0 indicates allocation to control (Khandker et al. 2010). As evident from equation 1, the DiD estimator is based on a comparison of treated and control households both before and after the intervention – with the average treatment effect being calculated as the difference between the outcomes for the treatment group after netting out the difference in outcomes for the control group. A central assumption of any DiD strategy is the parallel trend assumption, whereby the outcome in treatment and control groups would display the same trend in the absence of treatment (Angrist & Pischke 2008). In this case, the outcome change in the control group (𝐸 𝑌'. − 𝑌*. 𝑇' = 0 ) acts as the appropriate counterfactual (Khandker et al. 2010). While unobserved heterogeneity (differences in ability, income or motivation across treatment and control groups which might affect the outcome) could result in selection bias, if unobserved heterogeneity is time invariant, then, once again, the outcome change in the control represents the counterfactual (as will be discussed, Figure 2 is supportive of the parallel trends assumption). The DiD estimate can also be calculated within a regression framework. Following Abramitzky & Lavy (2014), the difference in pre- and post-intervention consumption in treated households relative to the control is modelled with the following DiD regression: 𝑌0% =∝ +𝛽' 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠0 + 𝛽A 𝑃𝑜𝑠𝑡 + 𝛽C 𝑃𝑜𝑠𝑡 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠0 + 𝛽E 𝑋G + 𝜀G
(2)
Where 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠0 is a treatment variable indicating if the household is in the treatment group and 𝑃𝑜𝑠𝑡 is a time variable indicating whether treatment has commenced. The coefficient, 𝛽C , the interaction between the treatment and time variables 𝑃𝑜𝑠𝑡 ∗ 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠0 gives the average DiD effect of the program. The variable 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠0 captures the underlying differences between treatment and control groups while 𝑃𝑜𝑠𝑡 captures the underlying differences between the twotime periods. Finally, 𝑋G are a vector of controls. The control variables are informed from the balance and pre-intervention trend regressions conducted in the following subsections. These include tariff block, suburb, seasonal and income quintile fixed effects and an indicator for length of billing period. All standard errors are clustered at the suburb level. A fixed effects model is estimated to control for the unobserved heterogeneity across treatment and control that does not vary over time. A. Balance in treatment and control groups We test whether treatment and control groups were balanced in terms of several key observable outcomes and demographic characteristics in January 2017 – when these households were “allocated” to treatment
and control. Participants in the treatment group are those who received the letter from the local municipality in the first quarter of 2017 while the control group are those participants who do not receive a letter. Note that all households in the subsequent analysis consumed 50kl or more in January 2017 (one of the criteria for selection into treatment). Given seasonality in water usage (to be discussed in the next subsection), the pre-intervention period is the same period in the previous year: April - October 2016. Column 1 and 2 of Table 1 provide the mean and standard deviation (in parenthesis) of monthly consumption (kl), daily average consumption (kl), property value (ZAR), number of billing days and tariff block for treatment and control cohorts, respectively. Columns 3 and 4 reflect the estimated difference between the treatment and control groups. The estimated coefficients in both columns 3 and 4 are estimated by regressing the treatment indicator (explanatory variable) on the characteristic or outcome variable (i.e. consumption) (dependent variable) (Abramitzky & Lavy 2014; Bhanot 2015). Standard errors are clustered at the suburb level. The estimates provided in column 3 reflect regressions with no additional controls. As evident from the results, characteristics do differ significantly across the two groups. Specifically, water consumption in the treatment group is significantly higher relative to control; also, treated households live in homes with significantly higher property values. As expected, given these results, treated households are also in a higher tariff block relative to households in the control. Column 4 replicates the balance regressions, but includes the participants' tariff block and suburb as additional controls. With the inclusion of these controls, mean characteristics no longer differ significantly across the treatment and control groups.
Table 1. Balance tests for treatment and control cohorts for January 2017
Monthly cons. Daily avg. cons. Property value Billing days Tariff block
(1) Treatment
(2) Control
(3) Difference
(4) Difference
61.09 (7.57) 1.78 (0.26) 2136974.3 (2628348.8) 34.56 (2.89) 5.63
58.99 (8.24) 1.73 (0.29) 1199055.4 (1742357.2) 34.3 (2.77) 5.54
2.101*** (0.656) 0.047** (0.023) 937918.92*** (213531.88) 0.265 (0.257) 0.084**
1.058 (0.814) 0.007 (0.028) 161045.8 (273043.0) 0.417 (0.271)
(0.48)
(0.5)
(0.037)
-
Note: Standard deviations in parenthesis. Period of analysis: January 2017. Standard errors clustered at the suburb level (columns 3 and 4). Sample: 655 households that consumed 50 kl or more in January 2017 and were eligible to receive the Mayor’s letter. *, ** and *** signify significance at the 10%, 5% and 1% level, respectively.
B. Pre-intervention time trend One of the key identifying assumptions of the difference-in-difference model is that water usage trends would be the same in the control and treatment groups in the absence of the treatment and that the intervention induces the deviation from this common trend (Angrist & Pischke 2008). In this section, we consider whether treatment and control groups have comparable pre-intervention trends. Figure 2 graphically depicts water use trends among eligible households. The figure plots consumption for the period January 2016 – February 2017. The figure indicates that consumption in treated households is generally higher than control households – although this differential in consumption decreases between December 2016 and February 2017. Furthermore, it is evident that there is a seasonal component to water usage: specifically, mean monthly consumption is higher during the warmer summer months (from November) and peaks in January. Thereafter, consumption begins to decline as South Africa moves into the cooler winter period. As the summer months of December and January coincide with the holiday season, the increase in water usage over this period is due to both an absolute increase in consumption given warmer weather (for example, filling pools and increasing the frequency of irrigation) as well as an increase in the billing period (number of days billed in a month) as the municipality has a lower staff contingent over the holiday season (pers. comm.). Note that while there is an element of seasonality in water use, this seasonal trend is common across treatment and control groups. Because of this seasonal component to water usage, we use month fixed effects (via a seasonal dummy variable) in all regression specifications to control for seasonal effects. In addition, when analysing the impact of the intervention, the corresponding month in the previous year is used as baseline consumption. As the intervention was conducted in mid-Feb to mid-March, we analyse the impact of the treatment over the period April - October 2017. As such, the commensurate pre-intervention period is April – October 2016. As evident from the figure, and lending credibility to the parallel trends assumption needed for DiD estimation, treatment and control groups appear to have a similar trend in the pre-intervention period. While Figure 2 suggests that treated and control households have comparable pre-intervention trends, following Abramitzky & Lavy (2014), we use pre-intervention data from April - October 2016 (the preintervention period) to examine whether the treatment and control groups have differential time trends with respect to water usage. The regression results are reflected in Table 2. The estimates in column 1 provide the results of a constant linear time trend model which allows for an interaction of the trend with the treatment indicator; in columns 2 and 3, the linear time trend variable is replaced by a series of month dummies as well as an interaction of the treatment indicator with each of these time dummies (Abramitzky & Lavy 2014). The difference between column 2 and column 3 is only the baseline, with July 2016 and October 2016 acting as the baseline dummy variable in columns 2 and 3, respectively. While the results from both models confirm the presence of a time trend with respect to water usage, in general this trend is identical for treated and non-treated households.
Figure 2. Pre-intervention average monthly consumption in treatment and control cohorts
The result in column 1 suggests that, on average, water consumption decreased by 0.7 kl per month between April and October 2017. This decline in consumption is expected given that consumption generally peaks in January/February and declines as the weather cools before picking up again towards the end of the year (Figure 2). As evident by the interaction term (Treatment x Time trend) this trend does not differ significantly between treatment and control groups. The Treatment variable is significant, confirming that average consumption of treated households was higher during the pre-intervention period. The estimates from the models with time dummies (columns 2 and 3) are generally consistent with the linear trend model. Both models indicate that treated households consumed significantly more than control households in the pre-intervention period. However, for both sets of regressions, the interaction terms (of the treatment indicator with the month dummies) are insignificant (with the exception of September 2016 for column 3). Overall the models indicate that, while treated households had a higher average consumption, treatment and control households followed the same trend with respect to water usage in the year preceding the intervention (April - October 2017). As a robustness check given the significance of the September month in column 3, we estimate the treatment effects for various periods - some including and some excluding September 2016. Our findings are extremely consistent across all these timeframes.
Table 2. Differences in the time trend of water usage in treatment and control in April – August 2016 (3) (2) Monthly consumption 4.438*** 3.896*** 5.271*** 1.548 1.405 1.425 -0.699*** 0.235 -0.142 0.261 9.793*** 6.466*** 1.33 1.646 2.365** -0.963 0.95 1.07 1.132 -2.196* 1.038 1.246 -3.328*** 1.151 0.838 -2.490** 0.98 0.965 2.562** -0.766 1.102 0.939 3.328*** 1.151 (1)
Treatment Time trend Treatment x Time trend April 2016 May 2016 June 2016 July 2016 Aug 2016 Sep 2016 Oct 2016 April 2016 x Treatment May 2016 x Treatment June 2016 x Treatment July 2016 x Treatment Aug 2016 x Treatment Sep 2016 x Treatment Oct 2016 x Treatment
1.015 1.584 1.363 1.101 -0.831 1.136 -0.51 1.168 -1.748 1.249 1.375 1.327
-0.36 1.816 -0.012 1.28 -2.206 1.518 -1.375 1.327 -1.886 1.176 -3.123*** 1.153 -
Note: Pre-intervention period: April – October 2016. Standard errors clustered at the suburb level. Baseline dummy variable: July 2016 in Column 2 and October 2016 in Column 3. Sample: 655 households that consumed 50 kl or more in January 2017 and were eligible to receive the Mayor’s letter. *, ** and *** signify significance at the 10%, 5% and 1% level, respectively.
IV. RESULTS In this section, we estimate the treatment effects of the social-norm inspired nudge. Ultimately, we conclude that the letter induced a significant and sizable reduction in consumption among treated households – a subgroup of high-consuming and price-insensitive households who have been slow to respond to water restrictions. The regression estimates provided in Table 3 demonstrate the effectiveness of the insert. To determine the impact of the letter, we restrict the analysis to the first meter reading after receipt of the insert. As the date of receipt of the insert is between mid-Feb and mid-March, the analysis is conducted for the period April – October 2017 (as April is the first month we can conclude with certainty that all households received the letter). Regressions are estimated for various timelines. The estimates in columns 1, 2 and 3, reflect the short-term impact of the letter. For example, column 1 displays regression estimates for the period April-May 2017, two-months after the intervention.1 Similarly, columns two and three provide treatment effects for the first three and four-months after households first received the letter. The remaining columns indicate the long(er)-term impact of the intervention with column six eventually reflecting the impact seven-months after the letter was sent. For all these regressions, and as previously discussed, to control for seasonal water use, we use the same month in the preceding year as baseline water consumption (April – October 2016). To control for unobserved heterogeneity, all regressions in Table 3 are fixed effects regressions. As evident from the table, the regressions include dummy variables for tariff block, month (via a seasonal dummy variable), property quintile and billing period. The results from Table 3 indicate that errant households, that received the social-norm inspired warning letter, reduced consumption by between 0.9-1.2 kl per month relative to control households and depending on the time period. More specifically, considering the first two-months after the intervention (column 1: April-May): the Post variable indicates that consumption in both groups decreased by an average 1.3 kl per month relative to the previous year. However, treated households additionally reduced consumption by an average 1.1 kl per month relative to control households (Treatment x Post). Table 4 provides the mean consumption values for two subgroups: namely, eligible households (who consumed in excess of 50 kl in January 2017) and the full sample. The time-periods in Table 4 mirror those from Table 3. As mean consumption for eligible households over the April-May pre-intervention period was 36.88 kl, this implies a reduction in consumption of 3.0% for eligible households. Looking at the wider sample, where mean consumption for the period was 18.03 kl, the letter resulted in a 6.1% reduction in water consumption.
1
Households received the letter between mid-Feb and mid-March. Thus, mid-March to mid-April signifies the first month after the intervention and mid-April to mid-May reflects the second post-intervention month. However, for simplicity, we assume the end of April to denote the first month after the intervention and the end of May to reflect the second month after the intervention and so on.
As evident from Table 3, the negative treatment effect (Treatment x Post) persisted over all time periods and a significant effect was still evident seven months after the intervention (column 6, April-October 2017). Figure 3 graphically depicts the treatment effects. The figure confirms that treated households consistently reduced consumption relative to control households across the timeline. Furthermore, Figure 4 plots the percentage reduction in consumption associated with the letter for both eligible households as well as the entire sample. As evident from the figure, the subgroup of eligible households reduced consumption by between 3.5% (three-months post intervention) and 2.5% (seven-months post). For the whole sample, the percentage reduction in consumption oscillates around 6.3%, reaching as high as 7% three-months after the intervention and then decreasing incrementally to 5.3% at the seven-month mark,
Table 3: Short and long(er)-run treatment effects
Treatment Post Treatment x Post Constant Tarff block fixed effects Month fixed effects Property quintile fixed effects Control for billing period R-squared Observations Treated Control Clusters Fpvalue
(1)
(2)
(3)
(4)
(5)
(6)
Consumption
Consumption
Consumption
Consumption
Consumption
Consumption
(KL/month)
(KL/month)
(KL/month)
(KL/month)
(KL/month)
(KL/month)
APR - MAY 2-months post
APR - JUN 3-months post
APR - JUL 4-months post
APR - AUG 5-months post
APR - SEP 6-months post
APR - OCT 7-months post
0 . -1.346*** (0.466) -1.102* (0.568) -16.868*** (4.28) YES YES YES YES 0.825 1875 1134 741 243 0.000
0 . -1.249*** (0.383) -1.207** (0.474) -16.498*** (3.182) YES YES YES YES 0.835 2754 1675 1079 244 0.000
0 . -1.193*** (0.331) -1.095*** (0.421) -15.520*** (2.593) YES YES YES YES 0.838 3599 2215 1384 245 0.000
0 . -1.130*** (0.297) -1.101*** (0.379) -13.490*** (2.299) YES YES YES YES 0.842 4462 2744 1718 246 0.000
0 . -1.216*** (0.28) -1.076*** (0.352) -11.401*** (1.977) YES YES YES YES 0.842 5351 3288 2063 246 0.000
0 . -1.361*** (0.252) -0.948*** (0.328) -11.992*** (1.946) YES YES YES YES 0.848 6244 3816 2428 246 0.000
Note: Fixed effects regressions. Standard errors clustered at the suburb level. Regressions estimated for various post-intervention periods. Pre-intervention baseline period is always the commensurate period of the previous year. Sample: 655 households that consumed 50 kl or more in January 2017 and were eligible to receive the Mayor’s letter. *, ** and *** signify significance at the 10%, 5% and 1% level, respectively. Results for control variables available on request.
Table 4: Pre-intervention mean consumption Period APR-MAY 2016 APR-JUN 2016 APR-JUL 2016 APR-AUG 2016 APR-SEP 2016 APR-OCT 2016
Sample Eligible All 36.88 18.03 34.34 17.17 33.68 16.90 34.16 17.17 35.12 17.35 37.74 17.88
Figure 3. Graphical representation of the treatment effect
Figure 4. Percentage reduction
V. DISCUSSION In mid-February to mid-March 2017, a subsample of households consuming in excess of 50 kl of water were sent a letter of warning by the local municipality. The letter admonished these households' "unacceptable" and "excessive" consumption, noted that their consumption was outside the norm of 20kl per household and, finally, threatened households with the installation of a water-management device should households not reduce consumption levels. This paper has evaluated the cumulative impact of this intervention over both the short and long(er)-run. With respect to the treatment effects, two-months after the intervention, treated households had reduced consumption by an additional 1.1 kl relative to control households. This treatment effect remained relatively consistent as the intervention period was extended, and still registered a reduction of 1.1 kl six-months after the intervention. At the seventh-month check, treated households had reduced consumption by an average 0.9 kl relative to control households. How significant is this reduction? Using some back of the envelope calculations, assuming an average reduction of 1.1 kl per month for a period of seven months implies a saving of 7.7 kl per household. To import some context, under the current Level 5 restrictions where individual consumption cannot exceed 87 l per person per day, the monthly consumption of a three-person household would be 7.8 kl. Thus, under the current restrictions, each of the treated households in the sample achieved savings (over the seven-month period) equivalent to the monthly consumption of a three-person household. We emphasise four important points: Firstly, households targeted for treatment were both the biggest consumers and the most resistant to the physical restrictions and tariff hikes. Figure 5 depicts the mean consumption of treated households relative to the rest of the sample (i.e. eligible and non-eligible households that did not receive a letter) for the extended timeline of June 2014 – Feb 2017. As evident from the figure, treated households consumed substantially more than the broader (non-treated) sample over this period. Moreover, the figure signals that, in contrast to non-treated households that began to reduce consumption from June 2016 onwards, treated households were much slower to adjust their consumption. The finding of a persistent 3% reduction in consumption among the top consumers highlights that the nudge was able to elicit a response in a group of households that were price insensitive and slow to adhere to the physical restrictions. Secondly, when considering the treatment effect against the broader sample of households (i.e. households eligible and not eligible for treatment), the social-norm inspired letter resulted in a consistent 6% reduction in consumption. This is a significant finding given that such a large reduction was achieved in a period where further price increases and physical restrictions would have been less politically palatable. Thirdly, these treatment effects were achieved at the end of a period which has seen increasingly stringent physical restrictions and tariff hikes: prices in the 5th and 6th tariff blocks increased by 293% and 689%, respectively. As the warning letter induced additional water savings on top of these more conventional
DSM tools, green nudges present as an effective tool for reinforcing government policy, particularly in a time of crisis. Figure 5. Average monthly consumption for treated and non-treated households, Jun 2014-Feb 2017
Finally, the intervention was extremely cost effective. Using information on pricing provided by the city, we estimate a cost of around 31c per insert. Again, using back of the envelope calculations and assuming a saving of 1.1 kl per household per month, we calculate a total water saving of 2926 kl. The cost per kl saved is thus calculated to be 4c per kl.
REFERENCES Abramitzky, R. & Lavy, V., 2014. How Responsive Is Investment in Schooling to Changes in Redistributive Policies and in Returns? Econometrica, 82(4), pp.1241–1272. Available at: http://doi.wiley.com/10.3982/ECTA10763. Allcott, H., 2011. Social norms and energy conservation. Journal of Public Economics, 95(9–10), pp.1082–1095. Available at: http://linkinghub.elsevier.com/retrieve/pii/S0047272711000478. Allcott, H. & Mullainathan, S., 2010. Behavior and Energy Policy. Science, 327(5970), pp.1204–1205. Andreoni, J. et al., 2004. Public goods experiments without confidentiality: a glimpse into fund-raising. Journal of Public Economics, 88(7–8), pp.1605–1623. Angrist, J. & Pischke, J.-S., 2008. Mostly Harmless Econometrics : An Empiricist ’ s Companion. Massachusettts I nstitute of Technology and The London school of Economics, (March), p.290. Ayres, I., Raseman, S. & Shih, A., 2012. Evidence from Two Large Field Experiments that Peer Comparison Feedback Can Reduce Residential Energy Usage. Journal of Law, Economics, and Organization, 29, pp.992–1022. Ayres, I., Raseman, S. & Shih, A., 2013. Evidence from Two Large Field Experiments that Peer Comparison Feedback Can Reduce Residential Energy Usage. Journal of Law, Economics, and Organization, 29(5), pp.992–1022. Available at: https://academic.oup.com/jleo/articlelookup/doi/10.1093/jleo/ews020 [Accessed August 8, 2017]. Bhanot, S.P., 2015. Rank and Response: A Field Experiment on Peer Information and Water Use Behavior. , (May). Brick, K., Martino, S. De & Visser, M., Behavioural Nudges for Water Conservation: Experimental Evidence from Cape Town, South Africa, Carlsson, F. et al., 2016. Spillover Effects from a Social Information Campaign, Chetty, R. et al., 2014. Active vs. Passive Decisions and Crowd-Out in Retirement Savings Accounts: Evidence from Denmark *. The Quarterly Journal of Economics, 129(3), pp.1141–1219. Available at: https://academic.oup.com/qje/article-lookup/doi/10.1093/qje/qju013 [Accessed September 29, 2017]. Chetty, R., 2015. Behavioral Economics and Public Policy : A Pragmatic Perspective. American Economic Review: Papers & Proceedings, 105(5), pp.1–33. Chetty, R., Looney, A. & Kroft, K., 2009. Salience and taxation: Theory and evidence. American Economic Review, 99(4), pp.1145–1177. City of Cape Town, 2017. Residential water restrictions explained. Available at: http://www.capetown.gov.za/Family and home/residential-utility-services/residential-water-andsanitation-services/2016-residential-water-restrictions-explained [Accessed January 31, 2017]. Croson, R. & Treich, N., 2014. Behavioral Environmental Economics: Promises and Challenges. Environmental and Resource Economics, 58(3). Datta, S. et al., 2015. A Behavioral Approach to Water Conservation Evidence from Costa Rica. , (June). Ferraro, P.J. & Price, M.K., 2013. Using Nonpecuniary Strategies to Influence Behavior: Evidence from a Large-Scale Field Experiment. Review of Economics and Statistics, 95(1), pp.64–73. Available at: http://www.mitpressjournals.org/doi/abs/10.1162/REST_a_00344. Khandker, S.R., Koolwal, G.B. & Samad, H. a., 2010. Handbook on Impact Evaluation: Quantitative Methods and Practices, Available at: https://openknowledge.worldbank.org/bitstream/handle/10986/2693/520990PUB0EPI1101Official0 Use0Only1.pdf?sequence=1. Nyborg, B.K. et al., 2016. Social norms as solutions. Science, 354(6308), pp.6–8. Thaler, R. & Sunstein, C.R., 2008. Nudge, London: Penguin Group. Yoeli, E., 2009. Does social approval stimulate prosocial behavior ? Evidence from a field experiment in the residential electricity market.
Appendix A