Sanitation dynamics: toilet acquisition and its economic and social implications in rural and urban contexts

This paper uses primary micro-data from Indian households residing in rural villages and poor urban neighbourhoods to shed light on household sanitation decision-making. We use a theoretical economic model to reduce dimensionality and complexity of this process. Beyond the most commonly analysed motivator, health, we consider economic and non-pecuniary bene ﬁ ts. We provide empirical evidence that each of these margins matter, and do so in both rural and urban contexts, and discuss how our ﬁ ndings can be explored in sanitation policy and programme design. (cid:129) The paper uses primary data from India to gain a deeper understanding about households ’ sanitation investment decisions. (cid:129) The paper places particular emphasis on economic considerations, which previously received less attention. (cid:129) The paper, in particular, identi ﬁ es that households perceive and experience important economic returns, including an increase in their dwelling value. (cid:129) Sanitation seems to be a pre-marital investment strategy and to shift children ’ s time allocation within the household. (cid:129) Sanitation seems to serve as a pre-marital investment strategy.


INTRODUCTION
Safe sanitation, a means of isolating human waste, has been recognised as an indispensable element of disease prevention and primary health care programmes (e.g. the Declaration of Alma-Ata, 1978). The worldwide consensus of its importance led to 'access to adequate and equitable sanitation' becoming part of the Sustainable Development Goals (UN ). Yet, with an estimated 1.3 billion people lacking basic sanitation, the scale of the problem is huge (Mara & Evans ).
An important challenge to increasing sanitation coverage is its costly provision. According to the World Bank, an estimated US$19.5 billion a year is needed globally to meet nationally defined WASH targets (Hutton & Varughese ). An under-acknowledged contributor to investments is households themselves: based on survey responses by 35 national governments in 2018/2019, households contribute an estimated 66% of US$52 billion of annual WASH expenditures (WHO ). These figures have triggered calls for a stronger emphasis on research that enables a better understanding of household investment in WASH (Danert & Hutton ). Novotný et al. () highlight that research aimed at understanding how contextual factors and motivations affect different sanitation outcomes is currently underdeveloped and that the current programmatic focus provides a narrow understanding of sanitation dynamics.
In this paper, we respond to this identified gap. We use primary micro-data from households residing in two Indian states to shed light on household sanitation decisionmaking, exploring the association of household characteristics with revealed preference for toilet uptake as well as outcomes resulting from the acquisition choice. Since such sanitation dynamics are characterised by complex humanenvironment interactions (Novotný et al. ), we present and structure our analysis around a theoretical economic model, which helps to reduce dimensionality and complexity. In addition to health and non-pecuniary benefits, the model highlights the importance of economic factors as motivators, a category that Novotný et al. () identify as under-represented in the sanitation literature.
Our data include two survey rounds implemented in both rural villages and poor urban neighbourhoods, allowing us to provide a rich picture of the main correlates with, and potential outcomes of, sanitation uptake in different environmental contexts (Note that we clarify the distinct socio-economic-cultural contexts and resulting differences in motivations for, and impacts of, toilet construction throughout our analysis but do not intend to use aggregates to inform policy or programming.).
India is a particularly apt context to study household investment in sanitation, having contributed over 50% of the close to 700 million people who defecate in the open globally in 2017 (UNICEF and WHO ). The Government of India has shown a significant commitment to achieving SDG 6 of clean water and sanitation for all by 2030, including via its ambitious Clean India Movement.

Data and study population
We use data collected as part of an evaluation effort for a sanitation intervention (The original study's baseline and endline report (Augsburg & Rodríguez-Lesmes )   This set-up guides our empirical analysis, which is split into two parts. For the first partan analysis of toilet acquisitionwe consider, for example, the role of the household's current wealth/income, sanitation ownership of its neighbours and proxies for anticipated marriage decisions.
For the second part of our empirical analysisunderstanding potential benefits of owning a toiletthe model provides testable hypotheses for households that are not liquidity constrained and hence are able to make the sanitation investment. For these households, the investment moves resources from the present into the future. As a result, according to the model, we should observe an increase in the value of assets owned, either because the investment itself adds value or because additional assets are purchased.
This part of the model leads us to look at the information on owned assets as outcomes. As mentioned above, such an increase in assets might allow households to obtain access to credit markets, which we consider through information on household borrowing. Additionally, improved health and productivity (potentially in combination with productive investments) can increase household income.
Changes in the household's permanent income and wealth have, in turn, implications for consumption patterns, leading us to look at households' consumption expenditures. Finally, sanitation investment might affect the balance between consumption and labour in two ways: via productivity or via the marginal utility of labour and consumption, induced directly through the impact on health. This implies that the impact of sanitation investment on labour supply is ambiguous. We will bring it to the data through information on children's time use. The analysis section will discuss the full set of outcomes that we consider.
Finally, although households in this model can borrow and save, some might remain liquidity constrained given imperfect financial markets with borrowing limits that depend on a household's wealth, and hence only indirectly on its income potential.

Empirical strategy
Our analysis is structured around two main objectives of (1) assessing determinants (correlates) of toilet acquisition and (2) understanding the potential benefits of toilet ownership for several outcomes.
We analyse correlates of toilet acquisition through a logistic regression where we constrain our sample to households that had no toilet at the time of the first survey round.
It establishes the conditional correlation between a set of covariates X and toilet ownership status T.
Variables are at the level of the household i, the community j and time t, with t ¼ {1,2} representing the two survey rounds. The vector of estimated parametersβ 1 gives us an idea of the correlation between each variable on the right-hand side and toilet ownership. We cluster the error term u at the community level. We run this regression for the entire sample, as well as separately for rural and urban 'slum' areas (We confirm the robustness of our findings by, first, assessing the sensitivity of estimates to the inclusion of a different set of covariates, X, and, second, using a linear instead of a logistic regression model.).
For our analysis of the link between toilet ownership and outcomes, we can consider potential impacts by regressing outcomes on toilet ownership, conditional on the above determinants, which include confounding factors, such as income and education. We use a linear panel model, which includes a time (survey round) dummy (γ t ) and household fixed effects (α i ).
Other variables and subscripts are defined as above.
Monetary outcomes are transformed using the inverse hyperbolic sine transformation (IHS), which is similar to a logarithmic transformation but allows for zeros in its domain. The resulting coefficientsδ andω 2 are to be inter- Although findings are in line (as discussed later), Equation (2) remains our preferred specification since the matching exercise results in a significantly reduced sample size. Table 2 presents the results of the analysis of determinants of toilet acquisition. Our sample is households that changed toilet ownership status from having no toilet at survey round 1 to having one at round 2, 3-4 years later. These  (1). Column 7 presents the difference between the estimates in columns 5 and 6, obtained by interacting toilet ownership (as well as all other independent variables) with the 'slum' indicator.

Correlates of toilet acquisition
The first two panels provide results on wealth and status determinants. We observe a positive gradient of toilet ownership and social standing: the means of toilet ownership presented in column 3 show that households with higher income and higher caste are more likely to own a toilet. In terms of acquisition, our results reveal that those from the third income quartile are significantly more likely to invest in a toilet between the two survey rounds. Rural areas, where toilet ownership rates are generally much lower than in urban 'slums', drive this result. Interestingly, we also find that in 'slums', it is those in the second income quartile or above that are significantly more likely to acquire a toilet over time. At the same time, households in a scheduled caste or tribe are less likely to improve their sanitation ownership than those in a forward castea finding that is more pronounced in 'slums' than in rural areas.
Correlations of toilet acquisition with other proxies of household wealth are in line with those of income: Notes: Columns 1 and 2 present the mean at survey round 1, in rural and 'slum' areas, respectively, of each of the independent variables in the rows for households that did not have a toilet in their home at the first round (base mean). 'Rural' is peripheral villages of Gwalior and villages in rural Tamil Nadu; 'slums' are in the city of Gwalior. As a reference, column 3 presents the percentage of households that own a toilet conditional on the specific category (for discrete variables only). Columns 4-7 correspond to average marginal effects (AME), for each of the independent variables, after logistic regressions where the dependent variable is having a toilet in the house at the second round, conditional on not having one in the first round. Column 4 is for the entire sample, and columns 5 and 6 for rural areas and 'slums', respectively. Column 7 presents the difference between the estimates in columns 5 and 6, obtained by interacting 'toilet ownership' (as well as all other independent variables) with the 'slum' indicator. The four regressions include as controls all the variables presented in the table plus: a binary variable for the presence of any major shock to the household in the last 12 months; the age of the woman who is the household head or the spouse of the household head; an indicator of whether she lives with her in-laws; an indicator of whether the dwelling is owned by the household; and the value of household elements. Standard errors, clustered at the village/'slum' level, are shown in parentheses. Significance: *10%, **5%, ***1%.
households with dwellings of semi-strong (rather than strong or weak) structures are significantly more likely to invest in a toilet between the two survey rounds, again driven by rural areas.
We further find that certain household compositions and changes are driving sanitation investments. Specifically, the arrival of a new female (an increase in household size, conditional on the number of males) in a rural household increases the likelihood of constructing a toilet significantly, a finding not driven by a newborn member. Further, a male close to the legal marriage age (21) in the household makes investment more likely.
Finally, we find that conditional on contextual characteristics, such as wealth or income, an increase in sanitation coverage is positively associated with greater sanitation uptake in the village, as predicted by the theoretical model. We also find a negative association between the coverage of water service connections in rural villages and toilet ownership, in line with Bennett () results, which are consistent with the hypothesis that clean water can serve as a substitute for sanitation. Table 3 shows estimates of parameter δ (Equation (2)) in columns 4-6. Column 7 shows the difference between rural and 'slum' associations, similar to column 7 of Table 2. Inspired by the model, we group outcomes around health, household consumption, wealth and finances and time allocation.

Health
Health is typically considered a key motivator for improved sanitation and also highlighted in our model as one margin along which households might benefit. Within this analysis, we do not find evidence for reduced illness associated with sanitation, measured by diarrhoea incidence among children in the last week. We do, however, find evidence that households perceive health benefits of owning a toilet, particularly in urban 'slums', where the main respondent is 11 percentage points more likely to perceive herself as healthier than peers in the community and 10 percentage points more likely to perceive her family as healthier than other families in the community (significant at the 5 and 1% level, respectively).

Consumption
The model predicts that toilet ownership can result in increased income. Since income is an important confounding factor, which we account for throughout our analysis, we consider consumption expenditure, rather than income, as an outcome variable. A change in consumption expenditures is typically a result of a permanent change in income (Jappelli & Pistaferri ) and hence a valid proxy. We find a large, positive and significant correlation between sanitation ownership and household consumption, observed in rural and urban 'slum' areas alike. Particularly, nondurable consumption (such as transport, utilities, fuel, education and cosmetics) is higher for those with a toilet.
We stress that this relationship holds despite accounting for income in our analysis, which in fact could downwardbias estimates. The higher expenditures are suggestive that households with a toilet are more productive, potentially driven by better health, ownership of more productive assets or differing time allocation.

Wealth and finances
And indeed, we find significant relationships between sanitation and likely productive assets the households own, including vehicles (bicycle, scooter, motorbike, fourwheeler) and other household items (We also observe an increase in the value of agricultural assets for 'slum' households. Further analysis reveals that these assets are uncommon and small, with the results driven by about 2% of the sample.).
We further find that sanitation investments are reflected in the value of the dwelling, much above the investment  Interestingly, the significant associations of toilet ownership and consumption and wealth are relevant in both rural and 'slum' contexts but are generally quantitatively larger in rural communities.
Finally, results on credit outcomes suggest no changes in borrowing on either the extensive or intensive margin.

Time allocation
Our model predicts a potential change in labour supply. We do not have the relevant data to analyse this prediction. We

Robustness checks: matching
As stressed previously, while the presented data account for a large set of covariates and we consider consistency across time and context, results might still be driven by pre-existing trends. In other words, households that were already improving along several margins for reasons that we cannot observe might acquire toilets. We, therefore, employ as a robustness check a PSM exercise, which ensures we compare households that were as similar as possible along observable characteristics. Appendix A4 presents evidence of a successful PSM procedure ( Figure A2), providing us with a comparable sample. We find that our main conclusions hold true (Cross-sectional analysis with both survey rounds further confirms our conclusions.): coefficients remain in line (Table A3), but at times with reduced Notes: Each row represents the dependent variable of a set of fixed effects linear panel models. Columns 1 and 2 present the mean at survey round 1, in rural and 'slum' areas, respectively, of each of the dependent variables in the rows, for households that did not have a toilet in their home at the first round (base mean). 'Rural' is peripheral villages of Gwalior and villages in rural Tamil Nadu; 'slums' are in the city of Gwalior. Column 3 presents the number of observations of the regression in the 'full' sample. Columns 4, 5 and 6 present the coefficient estimate for the independent variable 'toilet ownership' when considering the full, rural and 'slums' sample, respectively. Column 7 presents the difference between the estimates in columns 5 and 6, obtained by interacting 'toilet ownership' (as well as the other independent variables) with the 'slum' indicator. All regressions include as controls: log of income, the presence of any major shock to the household in the last 12 months, the numbers of total and male household members, dummies for the presence of marriageable-age-male/marriageable-age-female/ under-5 members, and the proportion of water that comes from piped drinking water in the community. Consumption expenditure, amount of debt and asset values are in 1,000 Indian Rupees of 2013 (i.e. round 1 numbers were adjusted by a factor of 1.32, which was calculated based on national figures for 2011, 2012 and 2013). For the regressions, these values were transformed using the inverse hyperbolic sine (IHS), which for interpretation is similar to the logarithmic transformation. Standard errors, clustered at the village/'slum' level, are shown in parentheses. Significance: *10%, **5%, ***1%.
significance driven by the reduced sample (the matching procedure reduced the analysis sample by almost half) and hence larger standard errors. We lose significance altogether for the perceived health of the family in urban 'slums', consumption and the value of transport assets in rural settings, dwelling value, the value of household items and some timeuse variables.

DISCUSSION
The theoretical model has provided a useful foundation to support our analysis. Here, we use it to guide the discussion of our results and explain their broader relevance. We start with the motivating factor health for sanitation investment. and/or that the bride-to-be brings additional resources that allow the investment in a toilet to be made.
Finally, while our model suggests that increased wealth can improve access to borrowing, we do not find any statistically significant evidence for this. While we show a catch-up in toilet acquisition along income and caste gradients, our study also reveals that a large percentage of the population, in particular, those at the lower end of the income distribution in rural areas and those of lower caste, do not invest in sanitation. Households' own reporting that toilets are too expensive hints at liquidity constraints being important drivers.

CONCLUSIONS
Gaining a deeper understanding of contextual factors and motivations that affect sanitation outcomes is important given the current push and aim to provide safe sanitation for all, as manifested in Sustainable Development Goal 6.2.
While not experimental, our findings can be used to contextualise interventions better and to find different angles to promote sanitation. The most commonly It is interesting to note, though, that a significant percentage of the richest households in our context still do not own a toilet at the time of the second survey round (46% in rural areas and 16% in urban 'slum' areas). It is likely that these households do not perceive a high enough benefit from making the investment, rather than them being liquidity constrained. At the same time, Abramovsky et al. () show that CLTS is not effective in richer areas, making this a potentially challenging part of the population to motivate for sanitation investment.
Our findings suggest that a policy addressing nonpecuniary benefits, such as the 'No Toilet No Bride' campaign successfully implemented by the Government of Haryana, India (Stopnitzky ), might work well in our study context, particularly the urban 'slums'.
Identifying and assessing which policies are effective in further closing the sanitation gap, particularly when working at scale, should be a priority for researchers, policymakers and implementers in the field.

DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.