In Chapter 1, which is co-authored with Rossella Calvi and Arthur Lewbel, we show that a local average treatment effect (LATE) can sometimes be identified and consistently estimated when treatment is mismeasured, or when treatment is estimated using a possibly misspecified structural model. Our associated estimator, which we call Mismeasurement Robust LATE (MR-LATE), is based on differencing two different mismeasures of treatment. In our empirical application, treatment is a measure of empowerment: whether a wife has control of substantial household resources. Due to measurement difficulties and sharing of goods within a household, this treatment cannot be directly observed without error, and so must be estimated. Our outcomes are health indicators of family members. We first estimate a structural model to obtain the otherwise unobserved treatment indicator. Then, using changes in inheritance laws in India as an instrument, we apply our new MR-LATE estimator. We find that women's empowerment substantially decreases their probability of being anemic or underweight, and increases children's likelihood of receiving vaccinations. We find no evidence of negative effects on men's health. Then, using changes in inheritance laws in India as an instrument, we apply our new MR-LATE estimator. We find that women's empowerment substantially decreases their probability of being anemic or underweight, and increases children's likelihood of receiving vaccinations.In Chapter 2, which is co-authored with Alexander Wolf, we take the Dunbar et al (2013) (DLP) model and explore its strength and weaknesses at recovering information regarding household sharing of resources. DLP develop a collective model of the household that allows to identify resource shares, that is, how total household resources are divided up among household members. We show why, especially when the data exhibit relatively flat Engel curves, the model is weakly identified and induces high variability and an implausible pattern in least squares estimates. We propose an estimation strategy nested in their framework that greatly reduces this practical impediment to recovery of individual resource shares. To achieve this, we follow a shrinkage method that incorporates additional (or out-of-sample) information on singles and relies on mild assumptions on preferences. We show the practical usefulness of this strategy through a series of Monte Carlo simulations and by applying it to Mexican data. The results show that our approach is robust, gives a plausible picture of the household decision process, and is particularly beneficial for the practitioner who wishes to apply the DLP framework.Finally, in Chapter 3, which is co-authored with Bram De Rock and Tom Potoms, we exploit the experimental set-up of a conditional cash transfers (CCT) program in Mexico to estimate a collective model of the household and to investigate how parents allocate household resources. This is important to understand because the success of policies aimed at fighting poverty depends crucially on how parents respond to monetary incentives. If parents allocate resources inefficiently (or non-cooperatively), the resulting level of well-being is likely to fall behind the socially efficient optimum. This is undesirable given the prevalence of CCT programs over the last two decades which have occupied a large percentage of governments' annual anti-poverty budgets. Although there is evidence that they have been beneficial, their effectiveness may still be limited. Our aim is to tackle this research question by estimating a theoretically-consistent demand system and by applying at best a powerful test of household efficiency developed by Bourguignon et al (2009). Contrary to previous results, we show that households make efficient decisions only at the beginning of the program, but fail to cooperate later on. In order to rationalize these results, we propose a simple model of household behaviour where decision makers may change their preferences as a result of a treatment that gives information about the importance of a public good. / Doctorat en Sciences économiques et de gestion / info:eu-repo/semantics/nonPublished

<p>Causality is central to scientific inquiry across many disciplines including epidemiology, medicine, and economics, to name a few. Researchers are usually interested not only in knowing how two events are correlated, but also in whether one causes the other and, if so, how. In general, the scientific practice seeks not just a surface description of the observed data, but rather deeper explanations, such as predicting the effects of interventions. The answer to such questions does not lie in the data alone and requires a qualitative understanding of the underlying data-generating process; a knowledge that is articulated in a causal diagram.</p> <p>And yet, delineating the true, underlying causal diagram requires knowledge and assumptions that are usually not available in many non-trivial and large-scale situations. Hence, this dissertation develops necessary theory and algorithms towards realizing a data-driven framework for causal inference. More specifically, this work provides fundamental treatments of the following research questions:</p> <p><br></p> <p><strong>Effect Identification under Markov Equivalence.</strong> One common task in many data sciences applications is to answer questions about the effect of new interventions, like: 'what would happen to <em>Y</em> while observing <em>Z=z</em> if we force <em>X</em> to take the value <em>x</em>?'. Formally, this is known as <em>causal effect identification</em>, where the goal is to determine whether a post-interventional distribution is computable from the combination of an observational distribution and assumptions about the underlying domain represented by a causal diagram. In this dissertation, we assume as the input of the task a less informative structure known as a partial ancestral graph (PAG), which represents a Markov equivalence class of causal diagrams, learnable from observational data. We develop tools and algorithms for this relaxed setting and characterize identifiable effects under necessary and sufficient conditions.</p> <p><br></p> <p><strong>Causal Discovery from Interventions.</strong> A causal diagram imposes constraints on the corresponding generated data; conditional independences are one such example. Given a mixture of observational and experimental data, the goal is to leverage the constraints imprinted in the data to infer the set of causal diagrams that are compatible with such constraints. In this work, we consider soft interventions, such that the mechanism of an intervened variable is modified without fully eliminating the effect of its direct causes, and investigate two settings where the targets of the interventions could be known or unknown to the data scientist. Accordingly, we introduce the first general graphical characterizations to test whether two causal diagrams are indistinguishable given the constraints in the available data. We also develop algorithms that, given a mixture of observational and interventional data, learn a representation of the equivalence class.</p>