Intrahousehold preferences heterogeneity and demand for labor-saving agricultural technology

Evaluations of agricultural technologies rarely consider the implications of how adoption may alter the labor allocation of different individuals within a household. We examine intrahousehold decision-making dynamics that shape smallholder households ’ decision to use mechanical rice transplanting (MRT), a technology that disproportionately influences demand for women ’ s labor. To study the adoption decision, we experimentally estimate the willingness to pay for MRT services both at the individual and household level. We find that women value MRT more than men, especially when they participate in trans-planting on their own farms. This preference heterogeneity is evident in the unconditional differences between women ’ s and men ’ s valuation and differences conditional on their individual observable characteristics. Despite having stronger preferences for MRT, women have less influence on the household ’ s technology adoption decision than men. This differential influence over the MRT adoption decision reflects the intrahousehold power structure: in households where women have less control over assets, they also have less influence over the MRT adoption decision. Our results highlight how technological changes interact with unobserved, gender-based intrahousehold power relations to influence


J E L C L A S S I F I C A T I O N
D13, J43, Q160

| INTRODUCTION
Mechanization in agriculture can be both a cause and consequence of disruptions in rural labor markets.Economic theory has long held that economic growth in the high-productivity manufacturing sector tends to pull labor from the low-productivity agricultural sector, thereby increasing rural wages as part of the overall structural transformation process (Lewis, 1954).Higher wages and labor scarcities relative to other factors of production can, in turn, shift agricultural production towards more laborsaving technologies and a relative increase in the use of capital (Bigot et al., 1987;Binswanger, 1986;Hayami & Ruttan, 1970).At the same time, increased mechanization of farm operations can reduce the demand for manual labor, thus displacing farm workers.These dynamics are captured in models that integrate endogenous structural change with induced innovation, agricultural transformation, and mechanization, many of which have recently gained traction in empirical applications to smallholder production systems in developing countries (Binswanger & McIntire, 1987;Houssou et al., 2013;Takeshima, 2017;Takeshima et al., 2013;Takeshima et al., 2015).
When labor displacement arises as a consequence of technical change-specifically, through the adoption of labor-saving machinery-the distribution of economic welfare between men and women within a household or community depends crucially on the operations for which machinery is substituting for manual labor.An unexplored dimension of this technical change process is how intrahousehold preferences and the distribution of bargaining power within the household affect adoption decisions, especially if mechanization substitutes for men's and women's labor differently.In many research settings, it is often difficult to explore this dimension because of a lack of data, especially because many studies assume a unitary household and direct survey questions about farm management only to the male household head.Yet, even with a general division of labor within such households, significant agricultural decisions are often made in consultation with other household members, including female members.This aspect makes unobserved individual demand for agricultural technologies a potentially important empirical shortcoming.
This study explores this knowledge gap by assessing intrahousehold variation in demand for a new agricultural technology-mechanical rice transplanting (MRT)-along explicit gender dimensions.Each household member's demand for the technology is assumed to be a function of, among other things, the household member's gender as well as his or her participation in the agricultural operation being replaced.As such, this study represents one of the first efforts to understand gendered differences for a labor-saving agricultural technology that has disproportionate gendered impacts and to explore how heterogeneous preferences might converge through a bargaining process to arrive at a household-level adoption decision.
Our focus on MRT is both timely and relevant to Indian agriculture.Although women in rural India contribute to agricultural production in myriad ways, one of their primary contributions in India's rice-producing regions has long been the transplanting of rice seedlings (Unnevehr et al., 1985).Consequently, there are important gender dimensions to consider when evaluating technologies that substitute for labor during this phase of rice production.Manual transplanting is the most labor-intensive activity undertaken during rice production in India, accounting for as much as 20% of all labor employed in rice production (Barker et al., 1985).On average, manual transplanting requires 10-12 labor days per acre, and households may employ a combination of family and hired laborers to complete the task (Malik et al., 2011).The entire transplanting operation entails several distinct tasks that are typically undertaken by household members, such as managing and providing food for hired laborers, in addition to physically transplanting rice seedlings in the paddies.The physical task of transplanting rice seedlings requires long hours working in a stooped position in flooded fields and stagnant water, and is often accompanied by exposure to water-borne pests and diseases (Batliwala, 1988;Dey, 1985;Roberts et al., 1982).
MRT was first introduced in the Indian rice cropping system in 2006 and was heralded as a way to significantly reduce labor costs and improve rice productivity (Kamboj et al., 2013).An MRT equipment can transplant 3-4 acres per day, using a team of two individuals-a driver and a helper-to operate the unit (Malik et al., 2011).These productivity benefits translate into reductions in total hired and family labor used.For family members involved in manual transplanting, MRT adoption could shift individual members' roles and responsibilities, in addition to the total amount of labor supplied to the overall operation.
More importantly, MRT mechanizes rice transplanting and thereby masculinizes transplanting labor through two channels.First, female household members involved in transplanting are more likely than men to perform the task of physically transplanting rice seedlings into the paddies-the task that is replaced by MRT.In the Indian context, the intersection of wealth, land size, and caste are related to the activities women and men perform on the farm, and family women from land-poor, lower caste households may especially contribute to physically transplanting rice seedlings (Chen, 1989;Paris et al., 2000).Women from lower caste households may supply disproportionately higher amount of unpaid labor to this task as compared to men: whereas women may supply 16-20 labor days per hectare to transplanting seedlings on their farm, men may only devote 6-8 labor days per hectare (Paris et al., 2000).Second, male household members are more likely to perform supervisory tasks related to MRT operations than women and therefore would continue being broadly involved in transplanting operations.This is particularly true because women are seldom involved in managerial decisions regarding ownership or use of agricultural machinery in the Indian context (Farnworth et al., 2022;Paris et al., 2015).MRT's potential to reduce drudgery and displace women's labor from this operation makes understanding MRT demand among women as important as understanding the valuation of men who are typically the primary decision makers in rice production.
Based on this differential gender-based family labor reallocation induced by potential MRT adoption, this study attempts to disentangle two unobserved dimensions of a household's technology adoption decision.The first dimension pertains to the preferences of women and men for rice transplanting technologies.The second dimension relates to the power each household member has to influence the adoption decision.We develop a theoretical framework to understand the role of these two dimensions in the adoption decision.Empirically, the key challenge in identification is that only the household decision is observed while individual preferences are unobserved.To disentangle individual members' unobserved preferences and relative influence from the observed household demand for MRT, we use a combination of stated preference and experimental valuation methods.We capture MRT valuation from women and men belonging to same households using a stated valuation elicitation mechanism, and measure the household's revealed demand for MRT services using an incentive-compatible Becker-DeGroot-Marschak (BDM) mechanism (Becker et al., 1964).Structurally, these valuation exercises were identical, with the principal difference being the incentive compatibility of the latter.These valuation exercises provide three comparable measures of willingness to pay (WTP): one for the woman, one for the man, and one for the household.
We use men's and women's WTP to understand the differences in their preferences.Using a Kitagawa-Oaxaca-Blinder decomposition, we separate the unconditional difference in individual WTP into the characteristics differential (arising from differences in the levels of observed characteristics of men and women) and the conditional differential, which explains the differences based on those characteristics (Oaxaca, 1973).Further, we combine these individual WTP measures with women's participation in transplanting operations to assess men's and women's ability to influence the household technology adoption decision, as captured by the household WTP.We also assess the correlation between women's degree of power outside of transplanting with their ability to influence the household's transplanting technology adoption decision.
Despite results indicating that women have a stronger demand for MRT (as measured by their higher willingness to pay), especially when they participate in transplanting, our analysis suggests that women are less able to influence the ultimate technology adoption decision.We also find differences in preferences for MRT based on women's and men's relative bargaining power in decision making.In households where women have less control over the use of assets, they have significantly higher WTP than men.Despite these differences in valuation, men have greater influence over the household's MRT adoption decision-especially when women participate in transplantingsuggesting an imbalance in intrahousehold bargaining power.This power imbalance in technology adoption is directly related to the power structure of the household: in households where women have less control over the use of assets, the gap between men's and women's influence over the MRT adoption decision is substantially wider than in households where women have greater control over the use of assets.The result implies increasing the equity within intrahousehold power dynamics could accelerate the transition to mechanized rice establishment.
These findings are relevant in explaining how technological changes interact with unobserved, gender-based power relations to influence women's welfare.In our context, we find that although women may prefer technologies that shift their on-farm labor allocations, they may not be able to sufficiently influence household decisions in line with their individual preferences.Afridi et al. (2023) studied the expansion of mechanized agriculture in India, and although they found that agricultural mechanization reduced drudgery in tasks typically undertaken by men, such as land preparation, it also indirectly affected women by reducing their employment in agriculture.In both Afridi et al. (2023) and our study, there is substantial evidence of men's preferences being reflected in agricultural production decisions that also have direct and/or indirect implications for women's labor supply on the farm.
This link between intrahousehold gender inequities and mechanization is also noteworthy in light of significant changes in India's rural labor markets in recent years.From 2004From -2005From to 2009From -2010, 17, 17.8 million male workers left agriculture and were primarily absorbed in non-agricultural sectors.Over the same period, 36.4 million female workers left the agricultural sector, but unlike their male counterparts, the majority of these women-many previously involved in unpaid family agricultural work or self-employment-were not absorbed into non-farm employment (Raveendran & Kannan, 2012).Where women lack power to make choices about their labor supply, this broad structural transformation has the potential to worsen gender inequities.
Finally, our work contributes to the literature that examines the influence of women's bargaining power on their labor supply.For example, Heath and Tan (2020) show when women's autonomy increases, their labor supply in market opportunities also increases.Indirectly, our study establishes a similar connection and shows that narrowing the gap in men's and women's relative influence has the potential to improve women's control over decisions that influence their labor supply.

| RESEARCH DESIGN
This study was conducted with farm households located in 28 villages across 13 districts in the Indian state of Bihar.Bihar is one of the poorest states in India, with a multidimensional poverty headcount rate of about 52% in 2015-2016(NITI Aayog, 2021).The study was implemented during the months preceding the kharif (monsoon) rice-growing season in 2015 (a study timeline superimposed on the agricultural calendar is shown in Appendix A).We randomly selected 965 ricecultivating households that had an adult man and woman decision maker residing in the household.The sample size in each village was proportional to the population: about 25% of a village's population were selected, with a maximum of 65 households chosen in any village.
The study was implemented in three phases.First, beginning in March 2015, we conducted a survey to collect information on the household's demographic and social characteristics, as well as information on the labor and capital used in each agricultural task involved in cultivating rice.We also conducted an individual survey with the man and the woman identified as co-decision makers in which we elicited their information on asset ownership, human capital, employment and earnings, contributions to household decision making, and social and familial backgrounds.We collected information on their current status along these dimensions as well as those prior to their household formation, that is, at the time of marriage.
In April 2015, we revisited the man and woman co-decision makers and conducted value elicitation exercises to assess the men's and women's WTP for MRT services.This activity was implemented simultaneously but separately, with female enumerators interviewing women respondents and male enumerators interviewing men respondents.The enumerators introduced the MRT technology to men and women through a brief verbal introduction, followed by an informational video demonstrating MRT operations and introducing them to the MRT service provider who would offer custom-hire services to village members.In all cases, the service provider was not a member of the same village as the study participants but lived in a neighboring village and could have been known to sampled households. 1 Enumerators also read responses to a list of common questions about MRT service provision.We aimed to provide complete, accurate, and uniform information to both men and women, and to ensure that each individual understood that they would have an opportunity to hire the machine at a later time.
Prior to eliciting each member's WTP for MRT services, we also described the differences between MRT and manual transplanting in terms of the use of family and hired labor, productivity, and timing of transplanting.This discussion allowed each individual to form their own opinion about the technology.Eliciting the monetary valuation of MRT entailed a dichotomous "yes" or "no" response to 14 discrete price points, ranging from INR 600 to INR 1600 per acre for each of the plots the household intended to cultivate (see Appendix F for the prices used in the exercise). 2Because some women do not participate in their household's agricultural activities, we also provided participants with an approximate range of per-acre manual transplanting costs as a point of reference.
Throughout the valuation exercise, we used three strategies to minimize hypothetical bias (Aadland & Caplan, 2003;Cummings & Taylor, 1999;Jacquemet et al., 2013).First, we employed honesty priming to inform the subjects that there would be no benefit in stating WTP that deviated from their true valuation because their household would actually have an opportunity to receive the services if they desired (Jacquemet et al., 2013).Second, we used "cheap talk" and told both men and women repeatedly to state their valuations as if they were the ones responsible for making the final transplanting decision for their household (Cummings & Taylor, 1999).Finally, we elicited individual valuations and household WTP as a dichotomous choice for each price point to more accurately reflect the decisions individuals make (e.g., opting in or out at specific prices).At the end of the individual elicitations, both men and women were informed that researchers would return at a future date to discuss the household's opportunity to procure MRT services, and that they should use this intervening period to make their final decision about whether to hire MRT services.
The final phase involved implementing the household-level BDM valuation exercises in May 2015.Members of sampled households who self-identified as the "household head" in the initial survey were invited to participate in a village-level activity where they would have an opportunity to bid for custom-hiring MRT services. 3After reviewing the terms of MRT service provision with the participants, we elicited WTP for MRT services on each of the households' plots using the same 14 price points as the individual elicitation.Because participants would actually be able to hire MRT services 1 In our sample, about 34% of men and 8% of women reported knowing the MRT service provider.The heads of households were invited to the auction in order to mimic the way MRT services would have been offered in real life if an MRT service provider visited the village independently to provide services.In such a situation, the service provider would have interacted with the household head.The vast majority (92%) of those attending the auctions were men.Because BDM valuation exercises were organized at the village level (rather than on an individual basis), attrition during this phase of the study was high.From the original sample of 965 households, only 608 households had a representative attend the auction.The most frequently cited reasons for not attending were working on farm or being away from the village due to a prior engagement.Sample attrition appears to be random on most observable characteristics, as shown in Appendix B. However, households that were present for the BDM exercises cultivated slightly larger plots and had a lower valuation of women's individual WTP.If we assume that these differences are nonrandom, these results would imply that women in households who had a higher valuation for MRT were disproportionately underrepresented.When we regress the household's participation in the BDM on men's and women's WTP, we find women's WTP is statistically significant and negative, but the magnitude is very small and not economically meaningful.
if their WTP was greater than or equal to the village-drawn MRT price (that is, because the BDM auction was incentive compatible), WTP is a revealed measure of the household's demand for MRT. 4  The BDM valuation mechanism and individual-level valuation activities provide us with comparable measures of women's, men's, and households' WTP, with three key differences.First, the MRT service provider was present during the BDM elicitations but not during the individual elicitations.Second, the initial components of the BDM elicitation exercise-the description of MRT customhired services and the question and answer session that followed-were held in the presence of other participants, whereas other aspects of the household valuation were done in private.Third, despite our best efforts to mitigate bias, individual elicitations were ultimately hypothetical in nature.We control for these differences in our econometric specifications when we examine the influence of individual members' valuation in household's MRT demand.

| CONCEPTUAL FRAMEWORK
Here we provide the theoretical motivation for our research design and subsequent analyses.We show how household decisions are shaped by not only individual members' preferences but also by the extent of power-sharing among household members.We define the extent of power sharing as the relative weight placed on individual members' preferences in household decisions (Anderson & Eswaran, 2009).While the framework is general enough to apply to any labor-saving agricultural technology, we will use MRT as shorthand for the technology because that is our specific context.We build the model as a two-stage, decision-making process.Men and women reveal their individual preferences for the two transplanting technologies in the first stage, and the decision about which technology the household selects is made in the second stage.
In the two-member household comprising a woman and a man, i w, m f g, the household chooses between using traditional and mechanical transplanting, j trad,mrt f g , respectively.Using a set-up similar to Miller and Mobarak (2013), we assume that v i j is member i's valuation for technology j, such that the valuation premium, v i mrt À v i trad , is drawn from a distribution F i Á ð Þ with density f i Á ð Þ. 5 Theoretically, this valuation premium arises from utility maximization in which individual preferences map the different technology attributes (such as the use of female and male labor, use of hired labor, and speed of transplanting) into a scalar utility, and each individual selects the technology based on income and technology prices (Lancaster, 1966). 6In our research design, this valuation premium is captured by individual valuations for MRT relative to traditional transplanting.
Our next set of assumptions pertain to implementing transplanting tasks.We assume that adopting MRT entails hiring MRT services (an operation managed by the man), and the cost of custom-hiring MRT services is given by p mrt .We assume that the man is responsible for managing 4 Participants were informed that their bids would be binding, and a practice round also helped the participants understand this aspect.As described in the BDM protocol in Appendix F, we emphasized that participants freely and independently express their true valuation for MRT services in a manner most consistent with their households' unique needs.The elicitation mechanism was exhaustive in the checks to ensure that participants felt autonomous in truthfully revealing their demand.The participants did not pay for MRT services when the auctions were conducted but only after MRT services were provided on their farm plots.Participants with a nonwinning bid could not obtain services from the contracted service providers but could obtain MRT services from other MRT operators.Participants who were selected to receive MRT services could choose to not obtain them when the service provider visited the villages to discuss actual service provision.

5
F i Á ð Þ may not necessarily be statistically independent among members of the same household.However, given that marital relationships in rural India are primarily arranged by families and rarely on the basis of individual choice, the independence assumption is plausible (Rosenzweig & Stark, 1989).

6
Using the random utility framework, we can express individual i 0 s utility (u i j ) for technology j as: McFadden., 1974).Here, x i jk represents the kth attribute for technology j, β i jk is the marginal utility derived from this attribute, and ε i j is the stochastic term in the utility function.Assuming the use of unpaid female family labor is an attribute of each of the two technologies, the woman and the man co-decisionmakers may derive differential marginal utilities from this attribute.In addition, the price of traditional transplanting may also be different for women and men if women supply unpaid family labor, as reflected in the shadow value of women's time.
agricultural income and pays the price, p mrt , if MRT is adopted. 7Traditional transplanting is implemented by hired and family female labor.The price of traditional transplanting includes the cost of hiring labor and/or women's unpaid family labor.A key assumption we make pertains to the price of women's unpaid family labor: women and men value women's unpaid transplanting labor at p i fem .If the household transplants manually, the price of woman's family labor is internalized by her as she is supplying labor for the task and the man incurs the cost of hiring transplanting labor, p hire .The woman incurs the price, p w fem , which may not be the same as man's perceived price for her labor, p m fem .We also assume the household decision about which technology to adopt in the second stage is made by the man after learning about the woman's valuation premium. 8Implicitly, we also assume that the man supplies the same amount of managerial labor when adopting either of the two technologies.Finally, we assume the man weighs his and the woman's valuation premium by α and 1 À α, respectively, in making the household's technology adoption decision.In summary, two asymmetries may exist between the man and the woman in relation to the transplanting technology decision.
1.The shadow value of woman's unpaid labor supplied to transplanting could vary between the man and the woman.This difference could arise from two related reasons.First, MRT is a new technology, and prior to the technology's availability, the woman and the man may have never discussed the woman's shadow value of time devoted to transplanting.Second, and more importantly, even if the woman and the man discuss the two technologies, the opportunity cost of woman's time is subjective, and the man and the woman could value it differently due to differences in woman's and man's perceptions of the drudgery (and health costs) associated with transplanting and the opportunity cost of woman's nonmarket and off-farm market work.This lack of a precise measure of woman's shadow value of labor (that is, it is not equal to woman's wage) is especially relevant in the rural Indian context where significant gender-and caste-based frictions exist in the labor market (Eswaran et al., 2013).Whereas the research design explicitly elicited the valuation premiums men and women would pay, we did not ask men and women the monetary value of women's unpaid labor allocated to transplanting.
2.The weight the man places on his and the woman's valuation premium is especially relevant if the woman and the man diverge on which technology to select in the first stage.This weight, α 0, 1 ½ , could vary for each decision the household makes (specific and endogenous to the transplanting decision) or could be fixed for all decisions the man makes for the household (exogenous to the transplanting decision). 9 In the first stage, the woman and the man learn about the attributes of the two technologies and independently reveal their valuation premiums.

Stage 1-Reveal valuation premium
The woman chooses MRT if it has a greater payoff than traditional transplanting or the valuation premium exceeds the differences in their prices.Equation (1) shows the woman's technology choice decision.Even though the man pays p hire and p mrt , the woman considers all prices in choosing the technologies because she is deciding as if she is the main household decision maker (and as consistent with the research design).

7
This assumption is valid in our context, as men were reported as making the decision about spending on farm inputs in 94% of the households.

8
The assumption about communication follows directly from our research design where the woman and the man were encouraged to discuss MRT after the individual valuation elicitation.The assumption about the man making the household's technology adoption decision is valid in our context, as men were reported as making the decision about transplanting in 91% of the households.9 An implicit assumption is that the weight-especially if it is fixed for all decisions-is known to both the woman and the man.Knowing this weight has no implication on the woman or the man forming their own independent valuation premiums in the first stage.Knowledge of this weight may have implications in the second stage, especially in how the valuation premiums are communicated in the household.For instance, if α ¼ 1, the woman may not communicate her valuation premium to the man because she knows it would have no bearing on the household decision.Our data do not allow us to ascertain whether household members discussed their valuation premiums with one another.
By rearranging terms, this becomes: For simplicity, we denote p mrt À p hire as p.Similarly, the man prefers MRT over traditional transplanting if: Equations ( 2) and ( 3) depict how the woman and the man may choose different technologies.First, their respective valuations may differ, that is (v w j ≠ v m j ), which could result from women and men deriving different utility from various attributes of the two technologies.Second, even if v w j ¼ v m j , the difference in the price woman and man assign to woman's unpaid transplanting labor may lead to heterogeneity in their technology choice.
In the second stage, the man makes the decision about which technology to adopt based on individual members' technology choices.

Stage 2-Man makes household technology adoption decision
The man learns about the woman's valuation premium and one of two cases emerge: either the woman and the man agree or they disagree about their technology choice.

Case (a): The woman and man agree on technology choice
In this case, the man selects the technology they both agree on.Irrespective of α's value, woman's valuation will be consistent with the household decision.The joint probability that the household adopts MRT when both agree is: Similarly, when both value traditional transplanting more than MRT, the household adopts traditional transplanting with joint probability: Case (b): The woman and the man disagree on technology choice When the woman and the man disagree on their technology choice, the man makes the decision to adopt MRT if: It is evident from Equation (4) that in the case of disagreement, the degree to which the woman's technology choice is reflected in the decision depends on the weight the man places on the woman's valuation premium (1 À α) as well as the divergence between p m fem and p w fem .In the extreme case when α ¼ 1, the woman's technology choice is never selected.Even if the man places an equal weight on his and the woman's valuation premium, the extent to which p m fem is lower (or higher) than p w fem will determine whether the woman's technology choice is selected disproportionately less or more by the household.For instance, consider the case when the woman prefers MRT disproportionately more than the man preferring traditional transplanting, such that j In this scenario, even if the man places an equal weight on both valuations, the woman's choice may not be reflected if the average of the man's and the woman's valuation premiums is still lower than p À p m fem , which results from a divergence between p m fem and p w fem .That is, not only could the asymmetry in the shadow value of woman's time result in disagreement about technology choices between the man and the woman, but it could also influence the household choice even if the man places equal weight on both valuations.The probability the household selects traditional transplanting in the case of disagreement is: Similarly, the probability the household selects mechanical transplanting in the case of disagreement is: Overall, the probability with which the man selects traditional transplanting for the household is given by: and the probability with which the man selects mechanical transplanting for the household is given by: The probability the household selects the technology the woman prefers is decreasing in α.The probability the household selects traditional transplanting when the woman prefers it decreases as p m fem increases, and the probability the household selects mechanical transplanting when the woman prefers it decreases as p m fem decreases.Put together, these conditions show how asymmetries in the household may influence decision making.Particularly, when the man places unequal weights on his and the woman's valuation premium or values her shadow value of time differently than the woman, the woman's technology choices are more likely to be not reflected in the household decision, as compared to when these asymmetries do not exist.
Table 1 provides a snapshot of the sample households.The vast majority of households (97%) are headed by men who are, on average, just under 50 years of age.About a quarter of our sample belonged to the upper caste.About 87% of female decisionmakers selected for the study are the wives of the household heads.Despite belonging to the same household, the sampled men and women vary across several dimensions with men being about 4 years older and having two more years of education than women.Women and men do not appear to be significantly different in terms of their stated risk and uncertainty preferences based on index measures constructed from a standard set of questions and exercises we conducted with each participant. 10Financial inclusion among men is also 10 For risk preferences, we simply asked women and men to rank themselves on a Likert scale of 1-10 on their tendency to take risks, with a score of 10 implying being fully prepared to take risks.For constructing the uncertainty index, we asked participants to select a bag for drawing a red ball from two bag options in order to win a hypothetical price.In one bag, they knew the exact number of red and green balls.In the other bag, they were unaware of the distribution of red and green balls.Based on whether they selected a bag with a known or an unknown distribution, we classified their uncertainty preferences.
higher: whereas 66% of men reported having a bank account, only 41% women had one.Despite having access to the same household assets, women believed they would be able to acquire INR 20,000 less in loans compared to men.On average, women supplied more labor to unpaid house work than men and less labor to the farm.
Although all households are involved in agriculture, the extent of individual members' involvement in agriculture varied: about 93% of men and 67% of women reported being involved in agriculture as their primary occupation.Moreover, about 80% of women reported participating in agricultural activities even if it was not their primary occupation.Women reported having less technical knowledge about agriculture than men in our sample, based on a knowledge index constructed from respondents' familiarity with 18 widely used mechanical technologies in India. 11Although overall access to agricultural extension is low for both men and women, it is 18% points lower for women than for men.

| Gendered tasks in transplanting operations
We have alluded to an important distinction between transplanting seedlings as a specific task in broader transplanting operations.In this subsection, we explicitly describe this distinction, which is essential to understanding the gendered division of labor in traditional rice transplanting.As a production activity, manual rice transplanting culminates in physically transplanting seedlings into flooded paddies, but a number of supporting tasks are also required as part of transplanting operations, including hiring, supervising and paying outside workers; coordinating family and hired workers as a work force; transporting seedlings from the nursery to the paddy; and providing food for workers as a common form of in-kind compensation.As shown in Table 1, about 76% of men and 69% of women reported participating in these broader rice transplanting operations in some capacity.About 24% of women also reported working as a hired laborer on other farms in addition to working on their own farms during transplanting (no women in our sample worked exclusively as hired laborers on others' farms).In principle, household members could contribute to any or all of the transplanting tasks, but in practice women are much more likely to transplant seedlings than men.
T A B L E 1 (Continued)

Observations 965
Note: Standard deviation in parentheses for mean outcomes and standard errors in parentheses for men's and women's differences in mean outcomes.Statistical significance of differences in mean outcomes between men and women is based on the t-test.
a Following variables were used in construction of the wealth index using factor analysis: ownership of cellphones, motorcycle, television units, cable television; expenditure on transport, education, and festival donations; ownership of diesel pump, rotavator, knapsack, and tractor; and the size of land owned (in acres).
b This cost only includes the cost of hiring male and female workers during transplanting operations.The cost of family labor is not included.
c Bargaining power index is constructed using a generalized weighting procedure employed in Anderson (2008) and measures the difference in power between men and women as captured by differences between their age, education, the amount of credit they could access, and land owned by their fathers at the time of marriage.d Women's control over assets index is constructed as a simple average of women's ability to sell, give, rent, or purchase the following assets: land, cattle, sheep and goats, farm equipment, agricultural machinery, business equipment, house, large and small durable items, cellphone, nonagricultural land, and means of transport.
e Although 67% women reported agriculture as their primary occupation, 80.41% women (with a standard deviation of 39.7%) reported participating in agricultural activities even if agriculture was not their primary occupation.f These statistics are based on data from 617 households in which women reported participating in transplanting operations.
g Agricultural technology index is constructed as a sum of respondent's familiarity (ranging from not knowing to having adopted in own field) with a set of 18 agricultural technologies: diesel pump electric pump, diesel generator, thresher, traditional leveler, combine harvester, rotavator, seed drill, turbo happy seeder, four wheel tractor, fodder chopper, straw ripper, knapsack sprayer, power sprayer, power tiller, reaper, and mobile-based diesel irrigation pump. h For constructing the uncertainty index, we asked participants to select a bag for drawing a red ball from two bag options in order to win a hypothetical price.In one bag, they knew the exact number of red and green balls.In the other bag, they were unaware of the distribution of red and green balls.Based on whether they selected a bag with a known or an unknown distribution, we classified their uncertainty preferences. i For risk preferences, we simply asked women and men to rank themselves on a Likert scale of 1-10 on their proclivity to take risks, with a score of 10 implying being fully prepared to take risks.*p < 0.05; **p < 0.01; ***p < 0.001.

11
The agricultural technologies about which we elicited information include: diesel pump, electric pump, diesel generator, thresher, traditional leveler, combine harvester, rotovator, seed drill, turbo happy seeder, four wheel tractor, fodder chopper, straw ripper, knapsack sprayer, power sprayer, power tiller, reaper, and mobile-based diesel irrigation pump.
Figure 1 shows the tasks that men and women perform during transplanting operations among the 69% of households in which women report participating in these operations.In these households, nearly all of the women (97%) are involved in physically transplanting rice seedlings, compared to only half of the men.The share of women and men who report managing transplanting operations and providing food are similar.That is, although many women are multitasking during transplanting operations (perhaps supervising transplanting operations while they themselves are in the paddies transplanting seedlings), some men are supervising without actually transplanting seedlings in the flooded paddies.Moreover, as shown in Table 1, both men and women allocate an average of 3 labor days (per acre) to transplanting operations, in households where women report participating in transplanting operations. 12 When a household adopts MRT, the need for labor devoted to transplanting seedlings is eliminated.Because nearly all women participating in transplanting operations are involved in transplanting seedlings, MRT adoption will disproportionately displace female labor.MRT adoption will also likely influence who performs managerial tasks during transplanting operations because the coordination and supervision of custom-hired services, like other machinery-related decisions, in India tends to fall to men in male-headed households (Afridi et al., 2023;Paris et al., 2015).This may be especially true once women are no longer engaged in transplanting operations as seedling transplanters.Thus, given that MRT eliminates manual transplanting of seedlings and mechanizes transplanting operations, MRT effectively F I G U R E 1 Transplanting tasks by men and women in households where the woman reports participating in transplanting operations.The figure is based on men's and women's self-reported participation in different tasks during the process of rice transplanting in the sample of households where women participate in transplanting operations (N ¼ 617 households).
12 We also collected data on labor allocated by other household members to transplanting operations.Appendix C discusses the differences in the labor allocated to overall transplanting operations by all family men and women, based on information provided by the household head.masculinizes rice transplanting.Central to this paper is the disproportionate effect of MRT adoption on women, which raises the possibility of distinctly heterogeneous valuation of this new technology.Because this gendered effect of MRT on transplanting labor is especially evident in and relevant to households where women participate in the household's own transplanting operations, we separately analyze empirical patterns for these households (denoted as "woman participates in transplanting" or, simply, "woman transplants") in the sections that follow.

| Women's bargaining power
As described in the theoretical model, the weight the man places on his and the woman's valuation premium, especially when the woman and the man diverge on which technology to select, constitutes an important dimension of the technology adoption decision.This weight could be unique to the transplanting decision or could apply across all household decisions.In our empirical setting, we capture this notion as a woman's bargaining power, which is defined as her ability to exert her choice in joint household decisions.We expect the weight the man puts on woman's valuation's premium (α) is correlated with the woman's bargaining power in the household; that is, the greater her bargaining power, the higher the weight the man places on her valuation premium.
To understand the relation between this latent measure of power and women's relative say in the MRT adoption decision, we construct two measures based on intrahousehold bargaining models.Bargaining models have typically relied on measuring threat points, which capture the level of wellbeing that would accrue to the individual from exiting the household in the event that conflict arises (Anderson & Eswaran, 2009;Carter & Katz, 1997;Kabeer, 1999;Manser & Brown, 1980;McElroy & Horney, 1981;Quisumbing et al., 2003).A direct consequence of these threat points is that they specify a sharing rule in the household welfare function, which reflects in how strongly individual preferences are reflected in demand decisions made by the household (Manser & Brown, 1980;McElroy & Horney, 1981).Consequently, a change in individual threat points may also alter the sharing rule for the household.The first measure of power we construct is exogenous to the formation of the household and aims to capture this notion of threat points.We use the difference between women's and men's age, education, land owned by their respective fathers at the time of their marriage, and the amount of credit they could access to create an index of relative bargaining power using a generalized weighting procedure employed in Anderson (2008).We expect that the weight a man places on the woman's valuation premium is related to woman's relative power, as captured by this measure of power that is exogenous to household formation.
However, in a developing country like India, threat points may lie within rather than outside marriage or household formation (Anderson & Eswaran, 2009).That is, in a noncooperative scenario, women would rely on using either their unearned income if they have access to it independently or women who were working on the farm would continue doing so (Anderson & Eswaran, 2009).Further, Anderson and Eswaran (2009) posit that women's power share not only depends on the level of their income but also their control over the income they generate.Based on this notion, we construct an endogenous measure of power that is based on women's current role in decision making pertaining to ownership, procurement, and disposition of assets such as land, cattle, sheep, and goats; farm equipment; agricultural machinery; business equipment; house; large and small durable items; cellphone; nonagricultural land; and means of transport.Women who exercise some control over such assets could rely on them in the event of conflict within marriage (and arguably in disagreement over technology choice).We expect their degree of control over assets to be correlated with their share of influence in the transplanting technology decision.

| Intra-and interhousehold differences in WTP
The distribution of men's and women's valuation provides the first evidence of potential heterogeneity in individual MRT demand.Based on a Kolmogorov-Smirnov test, the WTP distributions of men and women are significantly different p À value < 0:01 ð Þ . 13Figure 2 illustrates the distribution of plot-level WTP for MRT services after subtracting the actual, household-specific cost of manual transplanting.First, responses in the southeast and northwest quadrants represent households in which men and women clearly disagreed about the value of MRT operations relative to manual transplanting.For example, in the northwest quadrant, women value MRT services in excess of the actual transplanting costs, whereas men value MRT services less than the actual transplanting costs.In the southeast quadrant, women value MRT services less than the actual transplanting costs, whereas men value MRT services in excess of the actual transplanting costs.About 43% of the WTP valuation pairs (including when either the man or the woman stated their WTP to be exactly equal to the transplanting cost) are in these two quadrants.
The second type of variation arises from the extent of deviation from the 45 line, which signifies perfect harmony in men's and women's valuations for MRT services relative to the manual transplanting costs and which further implies that both household members stated the exact same WTP for a particular plot.Although there were a surprising number of such occurrences (about 19% of women and men stated the same nonzero WTP), the majority of observations deviated from perfect harmony in valuations.Even if both members stated WTPs that were either both greater than or both less than the actual manual transplanting cost (bids in the northeast and southwest quadrants), the further away their valuations are from the perfect harmony (45 ) line, the greater is the difference in their individual valuation.The figure also suggests that men or women are potentially willing to pay more F I G U R E 2 Heterogeneity in women's and men's individual willingness to pay.Adjusted women's and men's willingness to pay is calculated as a difference between their willingness to pay for mechanical rice transplanting services per acre and the actual transplanting cost (cost of hiring labor) per acre that accrues to the household on a given plot.

13
If the WTP values are ranked, the Spearman rank correlation is low (ρ ¼ 0:28).Although the test rejects the hypothesis that the two unconditional distributions are independent, this could stem from the fact that these valuation pairs are from individuals belonging to the same household and who face the same household income and other constraints.
for MRT services as compared to manual transplanting (that is, at least one household member is willing to pay more if the responses are in the northwest, northeast, or southeast quadrants).
Table 2 reports the all-plot unconditional averages of men's and women's WTP for MRT services, with households classified based on the woman's participation in transplanting, women's and men's relative power constructed using exogenous measures (relative bargaining power), as well as women's power based on her degree of control over assets.We also classify households based on whether women and men agree on their transplanting technology choice.Households in which both members' WTP is higher or lower than the transplanting cost (that is, households in the northeast and the southwest quadrants in Figure 2) are in agreement about the transplanting technology.Households in which one member states a WTP higher than the transplanting cost and the other member states a WTP lower than the manual transplanting cost (that is, households in the northwest and the southeast quadrants in Figure 2) disagree about which technology to adopt.
Across these classifications, we observe clear differences in MRT demand between women and men within the same household, especially if the woman participates in transplanting operations, if she has less control over decisions pertaining to use of household assets, or if she disagrees with the man about the transplanting technology adoption decision.The difference between women's and men's valuation is INR 62 among households where the woman participates in transplanting operations, and this difference is statistically significant.However, this difference between women's and men's WTP is only statistically significant in households where the woman participates in transplanting operations exclusively on her own farm and does not also work as a hired laborer on other farms during rice transplanting.Men's WTP in households where the woman participates in transplanting exclusively on her own farm is INR 94 lower than women's WTP.The difference is not statistically significant between the men's and women's WTP when women work on other farms in addition to their own during transplanting.Moreover, the average WTP of women who also work as hired laborers is INR 779, and those who do not work on other farms but participate in transplanting on their own farm is INR 884, and this difference is not statistically significant.Men's WTP in households where women also work as hired laborers and in those where they do not work but only transplant on their own farm is INR 753 and INR 789, respectively, and this difference is also not statistically significant.
The difference between women's and men's WTP is not statistically significant when conditioned on whether women and men have relatively equal power (using exogenous measures), although the magnitude of the difference between women's and men's WTP is INR 90 in households where men have relatively more power.When classified based on women's reported control over assets, women who report having less control over assets have a WTP that is INR 113 higher than men, which is approximately 13% of the average WTP of this group.When women and men disagree, women prefer MRT more than men by INR 174 per acre. 14Note, however, that our data do not allow us to examine the exact reasons for women's higher valuation as compared to men along these dimensions.

| INTRAHOUSEHOLD PREFERENCE HETEROGENEITY
Although we find that women's unconditional average WTP is higher than men's, this difference could be due to differences in the characteristics of women and men, such as differences in education or participation in transplanting.As such, understanding whether women and men prefer different transplanting technologies conditional on these observable characteristics is important because in several contexts, policy design implicitly assumes that differences in agricultural decisions among 14 As shown in Appendix Table D1, the magnitude of these differences holds when we restrict the sample to households who attended the experimental auction.The statistical significance of these differences holds for all subgroups except for the subgroup of households in which women and men disagree about technology choice.The magnitude is similar, but the difference is not statistically significant at conventional levels.
T A B L E 2 Intrahousehold differences in willingness to pay (WTP) for mechanical rice transplanting services., and 10% critical level, respectively.Average auction WTP for each of the categories-women's transplanting participation, bargaining power, control over assets, and agreement-is not statistically different at conventional levels for the subsamples in each category.Women's relative bargaining power is categorized as lower (higher) if the bargaining power index is greater than or equal to (less than) the sample mean.Higher values of bargaining power imply men have greater power as compared to women.Women have lower control over assets if their control over assets index is less than or equal to 0.5.The index ranges from 0 to 2, with higher values signifying greater control over decision making pertaining to assets.Women and men agree on their valuation for MRT if both bid higher or lower than the transplanting cost that accrues to the household.women and men stem from differences in these characteristics and that narrowing such gaps would lead to similar outcomes (Quisumbing & Pandolfelli, 2010).

Women
To understand the conditional difference in WTP between women and men, we employ the Kitagawa-Oaxaca-Blinder decomposition, which measures how much of the mean differences among two groups can be attributed to differences in the levels of observed characteristics (Blinder, 1973;Kitagawa, 1955;Oaxaca, 1973).Suppose that WTP n , n m, f f g is assumed to be a linear and separable function in observable (X) and unobservable (ε) characteristics.
Here β maps individual characteristics (represented by X) into the WTP for MRT services; and X, β, and ε are indexed by gender n ð Þ.We can write the difference in individual WTP between women and men as follows.
We can rewrite this unconditional difference in WTP into two components: a component resulting from differences in the levels of these characteristics and a component due to differences in the coefficients associated with them.That is, assuming that E ε f Â Ã ,E ε m ½ ¼0, the unconditional WTP difference can be written as follows.
Equation ( 9) provides a twofold decomposition of the unconditional WTP difference between women and men.The first component, Δ characteristics (referred to as the level effect in the literature), provides the differential that arises from differences in the levels of the observed characteristics of women and men (Blinder, 1973;Fortin et al., 2011;Kitagawa, 1955;Oaxaca, 1973).The second, unexplained part (referred to as the structural effect in the literature) is the conditional differential, Δ conditional , because it captures the differences conditional on those characteristics being alike for women and men.
A caveat in interpreting the WTP differences is that these differences could arise due to hypothetical bias in the elicitation procedures and because women seldom make decisions about agricultural technologies.Implicitly, we assume that differencing WTP eliminates the hypothetical difference (that is, the hypothetical bias is equal for women and men).Moreover, we also observe that, on average, the WTP measures are increasing in household wealth, which also suggests that individuals stated their WTP while considering income constraints.If we assume that hypothetical bias is higher for women because they may not be involved in performing transplanting tasks, then we would expect that households in which the woman participates in transplanting and is aware of the costs of manual transplanting to have a smaller difference in WTP compared to households in which the woman does not participate.In contrast, we observe a larger difference in women's and men's WTP among households in which the woman participates in transplanting operations compared to households in which the woman does not participate.It could also be the case that the hypothetical bias is higher for women because they assume that their valuations would never be taken into consideration in the household technology adoption decision.In such cases, their WTP could either be higher or lower than their true valuation.We consider the sensitivity of our demand analysis presented in Section 6 to this case of asymmetric hypothetical bias between women and men.For the decomposition, we maintain the assumption of symmetric hypothetical bias between women and men.

| Results: Intrahousehold preference heterogeneity
To decompose the unconditional WTP difference into the characteristics and the conditional differential components, we specify a set of observable characteristics that may reflect the characteristics differential.We use a set of sociodemographic characteristics (age, education, and whether the individual belongs to a group, and credit access), agricultural involvement (participation in transplanting, access to agricultural extension, and technology knowledge index), risk and uncertainty measures, and time allocation (time devoted to housework, farm activities, and leisure every day).Based on Equation ( 9), we decompose the unconditional difference in WTP into the characteristics and the conditional differentials.
Table 3 reports the results from the decomposition of the unconditional WTP difference. 15We decompose the difference for the entire sample (Column (1)), based on women's bargaining power (Columns (2) and ( 3)) and based on women's control over the use of assets (Columns (4) and ( 5)).Several insights emerge from the decomposition.First, the conditional differential (roughly INR 119) for the full sample is higher than the unconditional difference in WTP between women and men, implying women have an even higher valuation for MRT than men after accounting for the differences in observable characteristics.Second, the characteristics differential, arising from differences in levels of these characteristics, makes men value MRT more than women by INR 71 for the full sample.In other words, if we narrowed the gap in the observable characteristics between women and men, men would value MRT more than women by about INR 71.Third, when we disaggregate the sample based on women's power, we find the conditional differential to be significantly higher in households where women have lower power in decision making (regardless of whether this power sharing is captured by bargaining power or the asset control measures).
Turning to the factors contributing to the characteristics and conditional differentials, we find that individual differences in education and access to extension contribute to higher valuation of MRT by men.In the full sample, differences in access to extension predicts men valuing MRT by INR 35 per acre more than women.Differences in involvement in transplanting do not contribute to the characteristics differential in the full sample.However, the contribution of these factors differs in the estimates of conditional differentials.We find women value MRT by INR 148 more than men conditional on being alike in risk preferences, and women value MRT by INR 37 per acre more than men conditional on being alike in uncertainty preferences.The other dimension that significantly contributes to the conditional differential is labor allocation.Men would value MRT by INR 187 per acre more than women if they allocated the same amount of labor to house work as women and by INR 103 per acre more than women if they allocated the same amount of labor to farm activities.In summary, these results suggest that differences in women's and men's valuations of MRT have a significantly higher component that is not driven by differences in their individual characteristics and appear to be rooted in the amount of time allocated to unpaid house and farm work.

| INTRAHOUSEHOLD BARGAINING AND HOUSEHOLD DEMAND
In this section, we use women's and men's WTP measures to assess their relative contribution in the binding household decision.Let WTP f and WTP m represent the woman's and the man's valuation, 15 Full decomposition results and the decomposition based on whether women and men disagree on which technology to adopt are shown in Appendix Tables D2 and D3.INTRAHOUSEHOLD DEMAND FOR AGRICULTURAL TECHNOLOGY 14678276, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12430 by University Of California -Davis, Wiley Online Library on [03/10/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E 3 Kitagawa-Oaxaca-Blinder decomposition of women's and men's willingness to pay.D2 and D3 in Appendix D. Agricultural technology knowledge index, risk index, and uncertainty index use the same definition as mentioned in Table 1.Hours allocated to leisure, home production, and farm labor were self-reported by women and men.
respectively, and let β f and β m represent the relative weights of their individual valuation in the ultimate household decision.The overall household demand for the technology, as captured by the household's WTP WTP h ð Þ, is: In addition to the relative influence of individual valuation in the household decision, we also control for the differences between the auction and the individual valuation procedures, which may have influenced the household WTP in our study's context and as described in Section 2. In order to account for these potentially confounding differences, we include the following variables in the estimation: whether the household head knows the service provider, whether the auction participant understood the auction procedure (as evaluated by an experienced survey enumerator), whether the household is upper caste, and the household's wealth index. 16We also included plot area as a control variable to account for plot size-specific factors affecting WTP (e.g., large plots may be more suitable to MRT use due to technical specifications of the machinery or due to economies of scale in mechanized transplanting).We estimate the following equation.
where X represents the vector of methodological and control variables influencing household demand, α is the vector of coefficients we estimate for these controls, and ε captures the measurement error in the estimation.
Using Equation ( 11) as the basis of our econometric estimation, we explore decision making for technology adoption in the household along several dimensions.In addition to estimating the relative shares of power between women and men in the household demand for MRT services using the full sample, we classify the households based on the following dimensions that may influence their degree of contribution to the decision.First, and perhaps the most critical dimension, is whether the woman participates in transplanting.Women who participate in transplanting operations may either have a higher or lower influence in the decision relative to women who do not participate.It could be that women who participate in transplanting have a lower say in their labor supply decisions to begin with or they could have had a greater say in these decisions if their transplanting labor was deemed indispensable.The second case is expected to arise when women and men disagree on which technology to adopt.Where there are disagreements about the preferred transplanting technology, women's WTP, on average, is higher than men's.However, the relative weights would reflect the weight placed on each individual's valuation, irrespective of which members' valuation is higher in the household.The third dimension relates to the power women have in the household to influence all decisions.Here, we predict that women's relative influence in the technology adoption decision is similar to the overall influence they have in household decisions.We classify households using both the exogenous measure of bargaining power and women's control over decisions related to assets.

| Results: Intrahousehold bargaining and household demand
Table 4 shows the estimation results for the full sample (Column (1)), by the woman's participation in transplanting (Columns (2) and ( 3)), and by whether men and women agree on transplanting technologies (Columns (4) and ( 5)).Among all households (Column (1)), men's influence parameter, (β m ) is higher than women's influence parameter (β f ), with β m À β f ¼ 0:16 , and this difference is statistically significant at the 10% level.Moreover, the sum of these influence parameters is well less 16 Based on enumerators' report on comprehension, over 96% of auction participants understood the auction activities well.than unity, implying that there are other factors that contribute to the final household WTP.Next, we examine the relative influence based on the woman's participation in transplanting (Columns (2) and ( 3)).In households where the woman participates in transplanting operations, men exercise an even greater influence over the household's MRT demand than in either the full sample or in households where the woman does not participate in transplanting.The difference in the influence parameters in households where the female co-decision maker participates in transplanting is 0:28 and is statistically significant at the 5% level, whereas the difference in the influence parameters in households where she does not participate in transplanting operations is only 0:03 and not statistically significant at conventional levels.Despite the low difference in influence parameters in households where the woman does not participate in transplanting operations, the individual influence parameter estimates (β m and β f ) are jointly significant, implying that valuations of either of them would lead to similar household decisions.Despite having a higher average WTP relative to men, women exercise less influence in the ultimate household adoption decision, especially when they participate in transplanting.
A related aspect of decision making is the power to influence decisions in case of a disagreement among household members about preferred technologies.As Columns (4) and ( 5) in Table 4 suggest, in households where the woman and the man agree, both influence parameters are jointly significant, but the difference between their influence parameters is only 0:10 and not statistically significant at conventional levels.In contrast, in households where men and women disagree about transplanting technologies, men have a greater influence in the household demand.Here, the influence gap is roughly twice the influence gap among households that agree.
Next, we turn to examining the estimated influence parameters based on women's bargaining power.Columns (1) and (2) in Table 5 report the estimates for the sample disaggregated by the relative power of women being greater or less than the sample mean, using factors exogenous to household formation.In households where men have an overall greater influence than women, they also have a much stronger influence in the technology demand-more than double-as compared to households where men's and women's power is more equitable.These findings also hold when we disaggregate the sample based on the degree of women's control over assets.In households where women have less control over use of assets, the influence gap (β m À β f ) is 0:22, which is more than double the influence gap in households where women have more control over assets.Combining these results together, the findings suggest that the relative power of individuals is a more relevant dimension of decision making pertaining to adoption of agricultural technologies, perhaps more important than being directly affected by the decisions.
As shown in Tables D4-D7 in Appendix D, these findings are also robust to restricting the sample to spouses only (as opposed to also including men and women related in other ways in the household) or to households that understood the auction procedures well, and including other potential differences between women and men such as the difference in their agricultural technology knowledge index. 17Appendix E shows the sensitivity of our results (that is, the difference in the influence parameters) to lowering or increasing women's WTP.We implement the sensitivity analysis to account for the possibility that hypothetical bias in women's elicitation could be disproportionately higher because women may never be in the position to make the technology adoption decision or because women co-decision makers may not participate in transplanting (or agricultural) activities.As the figures suggest, the difference in the influence parameters (β m À β f ) is robust in magnitude to changes in women's WTP, unless women's WTP is markedly lower than its observed levels.The statistical significance of the influence parameter difference, however, changes when women's WTP is lowered in subsamples where it is shown to be significant in our main results.17 Although we cannot verify the robustness of our findings to sample attrition, the fact that households that attended the auction had a lower average women's WTP as compared to households that did not attend the auction is suggestive that, all else being equal, our results would not have changed in economic significance as the gap in men's and women's WTP would have been higher had those households also attended.
T A B L E 5 Intrahousehold bargaining and household demand for mechanical rice transplanting services, disaggregated by women's bargaining power in household.

Outcome variable:
Women's bargaining power relative to men Women's control over assets Household WTP ) and women's (β f ) willingness to pay and the difference between coefficients of men's and women's WTP.The outcome variable, household WTP, has been elicited using a BDM experimental auction procedure.
In the previous section, we estimated the relative weights of men's and women's valuations in the household's ultimate MRT demand for relevant subgroups.The household demand measures allow us to construct aggregate demand for MRT services should it be offered at different price schemes.
Using the estimated influence parameters, we can also test how the aggregate demand for MRT services changes should the influence parameters differ.That is, we aim to understand if changing the power structure in the household also alters aggregate demand for such labor-saving technologies, assuming men's and women's preferences for the technology remain the same.We consider two particular scenarios that are observable and could be used for specifically targeting women: the woman's participation in transplanting and her access to and control of assets.For the woman's participation in transplanting, Figure 3 depicts the aggregate demand constructed using the regression estimates shown in Table 4 for the subsample of women who participate in transplanting.For the "actual" aggregate demand, we use the predicted values of household demand obtained from the regression coefficients for the subsample of households in which the woman transplants (and as shown in Table 4, Column (2)).Next, we obtain aggregate demand for the same subsample of women who participate in transplanting but assuming that this subgroup of households have the same influence parameters as the subsample of households where they do not participate and have a higher relative influence (depicted as the "As If Do Not Transplant" aggregate demand).That is, we aim to understand, ceteris paribus, what the aggregate demand would be if women who participate in transplanting had a greater relative influence as compared to men in the technology adoption decision.As the figure suggests, changing the influence parameters and narrowing the gap in influence among men and women leads to greater demand for MRT services at all prices per acre than compared with the actual aggregate demand observed in our data.
Similarly, we construct demand measures among the subsample of women who have less control in decisions about using assets.Although an individual's degree of control is not a directly observable characteristic, policies aiming to improve women's access to resources and assets, such as improving women's land rights, increasing women's wages and employment opportunities, and enhancing access to financial resources, could all potentially contribute to increasing women's degree of control in decisions (Quisumbing & Pandolfelli, 2010).By using the influence parameters obtained from the subsample of households where women have greater control over assets and constructing aggregate demand for the subsample of households where women do not, we find that if these women had greater power, the aggregate household demand for MRT services would also be greater as compared to the actual demand for MRT among this subgroup.This finding is especially relevant because the difference in power among the subgroups is not large: the lower control group has little to no control, whereas the group of women who are categorized as having more control only exercise marginally more control over these decisions, on average.
From a mechanization policy perspective, the possibility that intrahousehold bargaining dimensions could also influence demand for labor-saving technologies is especially relevant.If aggregate demand is low for such technologies because of intrahousehold asymmetries, then policies such as production subsidies may not achieve their intended objective.More importantly, a social planner observing the demand for labor-saving technologies particularly suited for women may mistake lower demand for a lack of demand for such technologies.Such insights have been found to be true in the context of other technologies relevant to women, such as improved cookstoves with presumably greater health benefits for women than men (due to greater exposure as a consequence of time spent preparing meals), and a range of other agricultural technologies with labor-saving benefits (Miller & Mobarak, 2013).The analysis also suggests that social welfare policies aiming to empower women could also, in turn, have spillover effects in the form of greater demand for labor-saving agricultural technologies and consequently lead to higher agricultural productivity.4 for this subgroup.The "As If Do Not Transplant" scenario shows hypothetical aggregate demand for the subgroup of households in which women participate in transplanting operations but uses the regression coefficients obtained for women's and men's WTP among households in which women do not participate in transplanting operations.(b) Scenario 2 figure shows the aggregate MRT demand for different prices (INR 600-1600 per acre) under two scenarios based on women's control over assets.The "actual" scenario shows aggregate demand for households in which women have lower control over assets using the estimated regression coefficients shown in Table 5 for this subgroup.The "As If Have Greater Control Over Assets" scenario shows hypothetical aggregate demand for the subgroup of households in which women have lower control over assets but uses the regression coefficients obtained for women's and men's WTP among households in which women reported having greater control over use of assets.
In this paper, we show that women's role in unpaid farm work and their power in household decisions are associated with intra-and interhousehold heterogeneity in demand among potential adopters of a labor-saving agricultural technology.Our theoretical model and empirical analysis suggest that asymmetries in power in the household may result in individual preferences not being equitably reflected in household decisions.These results have three broad implications for mechanization and women's empowerment at the individual, household, and policy levels.First, the results highlight the role that women's bargaining power and their labor market options outside of transplanting play in the emerging market for custom-hired MRT services.Even if MRT adoption implies hired women workers losing wages, an overall improvement of women's influence may allow them to have greater control over their unpaid and paid labor allocation decisions.Second, the heterogeneity in intrahousehold demand suggests the importance of including both women and men in efforts to promote MRT through public extension services, nongovernmental development projects, and commercial marketing strategies.This should be obvious not just because MRT has gendered effects on household labor allocation but also because the differences in MRT valuation between women and men-as shown in this paper-suggest a keen recognition of the implication of these differences on adoption among our participant households.
Finally, if complementary labor market policies to support women's market work are not implemented simultaneously with the expansion of labor-saving technologies, there is a possibility that higher rates of MRT adoption may push women into traditional gendered labor roles.That is, greater adoption of such labor-saving technologies that displace women's labor may also influence women's wage rates and bargaining power in rural labor markets.If households use MRT in order to reduce the drudgery of transplanting, then MRT adoption may limit women's work to only unpaid family housework and, in turn, lower their voice, agency, and mobility.Alternatively, if households use MRT to reduce drudgery and adoption of the technology opens up new employment opportunities for women, then MRT may increase voice, agency, and mobility.Future work on this topic may explore the linkages between women's bargaining power and labor displacement.In fact, Eswaran et al. (2013) find supporting evidence that when women withdraw from agricultural work to engage in "family status" production due to agricultural productivity gains, they lose their individual autonomy.These "family status" activities include providing greater attention to children, preparing meals, and improving the family's social capital.Another extension of the present research can examine the interaction between women's and men's wage rates and women's labor displacement, especially because women's and men's labor are not perfect substitutes in the Indian context.Mahajan and Ramaswami (2017) show that although women's labor supply does not influence male wage rates, men's labor supply has a significant effect on female wage rates.If women's displacement from labor-intensive agricultural tasks implies a greater proportion of men working on the farm, such a shift in farm production technologies can influence the relative wage rates of women and men.Ultimately, this massive exit of women from agricultural labor markets has long-run implications for women's participation in remunerative employment, welfare, and empowerment.
authors.This study was approved by the International Food Policy Research Institute (IFPRI) Institutional Review Board (IRB #00007490; FWA #00005121).

2INR=
Indian Rupee.At the time this study was conducted, the exchange rate adjusted for purchasing power parity was 64.152 INR/USD. 3

F
I G U R E 3 Influence of women's power on aggregate mechanical rice transplanting adoption.(a) Scenario 1 figure shows the aggregate MRT demand for different prices (INR 600-1600 per acre) under two scenarios.The "actual" scenario shows aggregate demand for households in which women participate in transplanting operations, using the estimated regression coefficients shown in Table Summary statistics: household, men, and women.
T A B L E 1 Standard deviation in parentheses shown for women's, men's, and auction willingness to pay.Standard errors for the differences in WTP are clustered at the village-level, and standard errors are shown in parentheses for the WTP differences.***, **, They disagree otherwise.INTRAHOUSEHOLD DEMAND FOR AGRICULTURAL TECHNOLOGY 14678276, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12430 by University Of California -Davis, Wiley Online Library on [03/10/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License The decomposition uses plot-level women's and men's willingness to pay.Robust standard errors shown in parentheses, and are clustered at the village-level.***, **, and * indicate statistical significance at the 1%, 5%, and 10% critical level, respectively.Factors included but not shown are age, credit that women and men are able to obtain, and leisure allocation.Full decomposition results are shown in Tables Intrahousehold bargaining and household demand for mechanical rice transplanting services.The regressions use plot-level household willingness to pay as the dependent variable.Robust standard errors shown in parentheses, and are clustered at the village-level.***, **, Note: The regressions use plot-level household willingness to pay as the dependent variable.Robust standard errors shown in parentheses and are clustered at the village level.***, **, Note: