Data
This study uses primary data from 439 spouses (878 respondents) in agricultural households in the upper Citarum, the biggest watershed in West Java. This upper watershed is mostly located in mountainous areas and the majority of the study site is used for agriculture and forestry (Agaton et al. 2016). The rapid transformation of the agricultural sector in this area presents a great variety of agricultural activities. Increasing demand for agricultural products and its proximity to Bandung city, a major urban centre, led to rapid agricultural intensification, increased cultivation of horticultural crops, and increased diversification of agricultural and non-agricultural livelihoods (Agaton et al. 2016; Mulyono 2010).
The survey applied a multistage stratified random sampling procedure. First, Bandung and West Bandung Districts were selected purposely because 65% of the Citarum Watershed lies in these two districts. Second, six out of eight sub-watersheds were chosen purposely because it was located in rural areas (two sub-watersheds that are located in the urban area were not included due to the lack of farming activities). Third, 22 villages from both districts were randomly selected, representing 10% of all villages in these two districts. Finally, 20 households were randomly selected from each village. The survey was conducted in Bahasa, the local language of Indonesia by local enumerators not from the study site.
The data were collected in July–August 2019. The data set includes information about household members, household and farm characteristics, access to credit, organisation membership, and farm and non-farm physical assets ownership. The survey instrument further includes a gender-specific decision-making module, with questions about agricultural activities directed to husband and wife separately. The survey is thus unique in providing detailed information on intrahousehold decision-making with respect to 21 agricultural activities in six domains (production, conservation practices, processing and marketing, training, credits, and buying and selling assets).
Methods
Measuring participation in intrahousehold decision-making
The survey asked: “Who makes decisions in the following aspects for most of the time in the past year?” for a total of 21 agricultural activities.Footnote 3 The responses to these questions correspond to a Likert-type scale from 0 to 10, which 0 means that the spouse decides alone, and the respondent has no participation at all over the decision, and 10 means that the respondent has full participation over the decision and the spouse has no participation at all. If the respondent answered 5, it means that the respondent perceives that both participate equally in the decision. This provides finer-scale responses to decision-making questions and goes beyond most existing studies that include only three choices of decision-making: self, spouse, and jointly (Acosta et al. 2019; Seymour and Peterman 2018).
Capturing the role of social norms
To incorporate the role of social norms in intrahousehold decision-making we included a question about the rationale for men’s and women’s reported participation in each agricultural decision. This question, presented after the identification of the decision-maker and the decision-making participation, was: “Why do you think this decision is made this way?” Based on the households’ typologies described in Bernard et al. (2020), the responses options included in the survey were:
-
i.
Whoever has better knowledge about the activity (from now on knowledge).
-
ii.
This is how decisions are made in the family/village (from now on family/village).
-
iii.
Whoever allocates the most resources (from now on resources).
Knowledge The most informed individual is the one making decisions about an activity, corresponds to the Bernard et al. (2020) most-informed typology. As discussed in Mudege et al. (2015), there is a wide belief that men are regarded as the ones with knowledge and women are perceived as their helpers (not as farmers). Agarwal (1997) also mentioned that social norms about gender roles in agriculture affect who gets access to information (e.g. who is invited to extension activities and allowed to interact with extension agents).
Family/village Social norms of the community and/or the functions that men and women are expected to perform within the household affect decision-making, corresponds to three household typologies: dictator (one individual, usually the household head, makes all decisions in the households), separate sphere (individuals within the household are in charge of separate domains), and norms (the person who decides is determined by the community norms). These types are all determined by expected gender roles in society (Lundberg and Pollak 1996).
Resources The individual who contributes the most resources used for an activity is the one making decisions about the activity, corresponds to contributor household typology. It is not uncommon that women and girls, specifically in agriculture, are perceived to contribute less than men or boys (Agarwal 1997). Since the response rate to resources is less than 7%, implying limited variation in the data, we opt not to incorporate it in further analysis.Footnote 4 This limited variation is not surprising since, in the Indonesian context, it is commonly believed that family resources are perceived as belonging to the household after marriage (Akter et al. 2017).
Women’s participation index (WPI)
We constructed a women’s participation index (WPI) in agricultural decision-making to measure men’s and women’s perceptions toward women’s participation in agricultural decisions. We followed a widely used approach to estimate asset indices similar to Smits and Steendjik (2015) for the International Wealth Index and Almas et al. (2018), who applied this method to estimate a women’s empowerment index based on women’s perceptions of partner/spouse violence. We adopt this methodology to reduce the dimensionality of our data on intrahousehold decision-making in agricultural activities (Filmer and Pritchett 2001; McKenzie 2005) and also to account for the different weights of each decision.
Specifically, we perform a principal component analysis (PCA) on the responses to decision-making questions. We conducted a separate PCA for men’s and women’s responses. We generated the weights using PCA and used the loadings from the first component, which explains the largest part of the variation in the data, to weight the components of the indices (see online supplementary materials, Table S1). Using this method, the WPI ranges from 0 to 45. For easier interpretation, we used the squared PCA loadings to transform the WPI to be between 0 and 10, where 0 means that the individual has no participation in the agricultural decision at the household, and 10 means that the individual makes all the agricultural decisions without participation of their spouse.Footnote 5 Two resulting indices: WPIw and WPIm, where w means women and m means men, respectively capture women’s and men’s perceptions on women’s participation in decision-making in agricultural activities.
We understand that we can lose some information by aggregating the data in an index. For this reason, we present sex-disaggregated descriptive statistics for the 21 decisions and the WPI in the results section of the paper.
Multivariate analysis
We analyse the correlation between participation in agricultural decisions and social norms while controlling for individual and household characteristics likely to influence this correlation, using ordinary least squares (OLS) as follows:
$$WPI_{xij} = \alpha + \beta_{1} socia\ln orm_{xij} + \beta_{2} individual_{xij} + \beta_{3} diffspouses_{ij} + \beta_{4} household_{ij} + \beta_{5} enumerator_{ij} + \beta_{6} district_{ij} + \varepsilon_{ij}$$
(1)
where \(W{PI}_{xij}\) is the women’s participation index for x = women, men of individual \(i\) in household j, \({socialnorm}_{xij}\) represents individual i’s perceptions of social norms in household j from the perspective of x, \({individual}_{xij}\) represents individual i’s characteristics in household j from the perspective of x, \({diffspouses}_{ij}\) represents characteristics differences between spouses for individual i in household j, \({household}_{ij}\) represents household characteristics for individual i in household j, \({enumerator}_{ij}\) represent the gender of enumerator that interviewed individual i in household j and is used to capture any systematic effect of the enumerator gender, and \({district}_{j}\) represent the district location of household j, to capture the regional effect. \({\upbeta }_{1}, {\upbeta }_{2}, {\upbeta }_{3}, {\upbeta }_{4}, {\upbeta }_{5}, {\upbeta }_{6}\) are parameters to be estimated and \({\upvarepsilon }_{ij}\) as the error term.
Social norms variables are measured using knowledge, and family/village. Knowledge indicates respondent's perception related to the reason on the decision is made based on the person who has better knowledge, and family/village measures respondent’s perception on the reason is made because it is commonly done that way in the family or village. In our study we have 21 activities, thus if a respondent answers knowledge for all 21 activities, then his/her knowledge value will be 21; and 0 if the respondent answers none on knowledge for all 21 activities (see Table 1 for further details).
Table 1 Definitions of variables used in analysis and summary statistics The first set of covariates capture a variety of observed individual characteristics which are usually hypothesized to play a role in determining women’s participation in decision-making. These include age (e.g. Anderson et al. 2017; Frankenberg and Thomas 2001; Reggio 2011), years of education (e.g. Doss 2013; Kabeer 2005; Sen 1999), agricultural organisation membership (e.g. Agarwal 1997; Lyon et al. 2017) and off-farm activity involvement (e.g. Bayudan-Dacuycuy 2013; Maligalig 2019). Interestingly, findings regarding these factors are usually mixed, where some studies suggest a significant effect while others do not, offering a further reason to test these factors in the current study.
In addition to individual characteristics, differentials of certain observed characteristics are also included. As suggested by Agarwal (1997), because “inequalities among family members in respect to determinant factors would place some members in a weaker bargaining position relative to others”, affecting the level of participation in the decision-making. In this study, differentials in age, years of schooling, and agricultural organisation membership between husband and wife are used, based on literature findings (Brown 2009; Doss 2013).
Household characteristics are further included to capture variations at this level. Following literature findings, women family farm labour participation (Bokemeier and Garkovich 1987; Rosenfled 1986), the total number of young children up to five that live in the household (in the spirit of Anderson et al. 2017), men to women ratio in the household (e.g. Brown 2009; Quisumbing and Malucio 2003), whether or not parents/parents-in-law living in the householdFootnote 6 (e.g. Anukriti et al. 2020; Bayudan-Dacuycuy 2013), land size (e.g. Alwang et al. 2017), and household asset index (e.g. Doss 2013) are used as variables to capture household characteristics \(.\)
Finally, we control for the gender of the enumerator. Alwang et al. (2017) found a tendency that men respondents that are interviewed by women enumerators to give a more positive response to wife's participation in decision-making. We also control for location in West Bandung district, since it is relatively closer to a major metropolitan area (Bandung city). Such proximity provides off-farm paid labour opportunities to women, and more urbanised settings, with less tight-knit communities, “may demonstrate a relaxation in social and gender norms” (Bradshaw 2013).
In identifying the determinant of WPIw and WPIm, we conducted two separate estimations. To adjust for potential heteroscedasticity, standard errors are clustered at the village level (Wooldridge 2002).