 Hello everyone, my name is Lidrych Kroyser and today I'll be presenting my work with Rulof Berger, entitled Unequal Partners, the determinants and consequences of intra household inequality in South Africa. So usually when we do microeconomic research, our main unit of account is the household. In inequality research, what this means is that we are focusing on the inter household inequality and not really focusing on the intra household dimensions. Some estimates suggest that we are greatly underestimating consumption inequality when we do this. So understanding household origin making can help us design policies to better target the most vulnerable members of society and it also can give us insights into some other important microeconomic aspects such as the labour supply, human capital investments and fertility and so on. So like with these motivations in mind, our research research are much less ambitious. So we're asking four questions. First is the unitary model of household behaviour making valid for South African households? If it's not, is the collective model valid? If the collective model is indeed valid, can the effect of bargaining power be seen in the expenditure on consumption items and then also which factors affect the bargaining power of different household members and can gender preferences for goods be observed? So the outline of this discussion will be pretty fast followed. I'll start with the theoretical model. I'll continue through the econometric model that will discuss the data and the results. So we start by considering a two adult household with a wife and a husband. We have a household member that consumes a vector of private consumption goods. These are goods that are excludable but not necessarily assignable and this is mainly a assumption we make due to the data. The two members jointly consume public goods. Each member has their own preference relation, preferences denoted by vector A and each member has their own utility which may be dependent on her own consumption of private goods, dispasses consumption of private goods, public goods and preferences. And then consumption is then constrained by the household budget as normally. So the household function can be used to express as the weighted average of the member's utility and we see this new term, this state over here and then this is dependent on, this may be dependent on total income of the household, the preference factors of the household and distribution factors. This Pareto weight represents the decision about the utility weight of the member in total household utility. So the Pareto weight may be potentially determined by then this vector of distribution factors. Distribution factors are variables that do not enter household, the household budget constraint nor do they enter preferences of individual members but they may affect the bargaining power of each household members. Where we assume some separability between private and public goods along with the usual technical assumption on individual utilities we may rewrite private good demands as in this equation here. So private good demands are then just some function of income, preferences and distribution factors. I merely distinguish between these two things, these two forms suggest show in cross-section data where we do not have any price variation. We can't really distinguish between variation and the Pareto weight that is due to the income or household preferences but we do know that these distribution factors are not supposed to enter the household demand function individually and directly. So the unitary model of household behavior assumes that the household behaves as if individual preferences can be aggregated into a stable household preference relation and this is very convenient for economic analysis because this means that when we observe the household and we observe aggregate data we can just use this and go on and all our results would be very accurate. But this also implies a very strong and testable restriction for household behavior namely distribution factor independence. This means that the household demands are not dependent on changes in distribution factors so rather factors that only affect the Pareto weights of the two household members. This is also known sometimes as the income pooling hypothesis and has been overwhelming rejected in empirical studies. Against the collective model of household behavior we assume that individual members have their own preferences and that the outcome of household decisions are Pareto-efficient. So this basically means that the bargaining power of household members can affect household consumption outcomes but only through one dimensional effect on decision weights and this is the Pareto weight. This basically means that for all distribution factors that yield the same Pareto weight we should expect the same effect on household consumption. This provides a cross equation restriction known as the proportionality condition that can be used to test the model. This restriction over here has been known to be necessary and sufficient to identify the collective model of household behavior at least since Pergonion's 1993 paper. So maybe a bit easier way to understand the collective model is as though households behave as if making decisions according to a skewed stage process. In the first stage the household decides what proportion of expenditure is allocated to each member according to the Pareto weight and in the second step it is how is individual demands for private goods determined then by this Pareto weight. So this is basically senior so in this case the effect of the female share and expenditure is basically the change in the Pareto weight due to a change in the distribution factor and this is the effect of the Pareto weight and then the difference in the two shares due to changes in the Pareto weight is then distribution factor invariance. So we would expect, so this is commodity invariance so we would expect this change to be the case for all commodities and then this for distribution factors. So the initial empirical studies used relative incomes as distribution factors but there might be some entogenity there as we might think that a household that has a member that might be more efficient or more productive in employment, the household would just prefer that member to do the work and the other household member to be in the household work. So aid in education difference are similarly problematic so more recent studies tend to look at distribution factors that affect outside opportunities of the spouses or give us an indication of the spouse's wealth before marriage. So there's widespread empirical support for the curriculum model so at least in France, Canada, India, Mexico, I'm also aware of England and a few other countries that has received support just to name a few and there's also two studies by Browning and others that attempt to estimate relative gender preferences for different commodities. It's shown that wives generally have a stronger preference for clothing, personal services and recreations whereas husbands care more about food, alcohol, tobacco and transportation. So on to the econometric model. So we modeled the demand for good eye in terms of the specification one over the other. So the demand for good is dependent on preference factors, household income in this quadratic term and then a quasi quadratic specification of the distribution factors. We use status seemingly unrelated regression estimator to estimate the model parameters and we control for factors such as children in the household, ownership of home or car, the rural or urban status of the household, the race of the household and they will also include variables for the individual household members including their age, education level, the hours work and whether or not they are employed. Our distribution factors in our preferred specification is the local gender share of the husband and of the other local gender share of males in the area and the maternal education share. So local gender share is just basically the unmarried men in the district council over unmarried women in the district council and then the husband and the maternal education share is the male mother's years of education over the sum of both fathers' years of education. If the unitary model is indeed an accurate depiction of the data, we would expect all of these coefficients to be zero. The proportionality condition implies that either equation two or three must be nested within the equation I just showed you. So we re-estimate these equations using the non-narrative seemingly unrelated regression estimator and then we use the likelihood ratio test to test whether or not they are or whether or not they are indeed nested in equation I. So I'm going to focus mainly on equation II due to time. And if this equation is valid, then it's convenient to or perhaps convenient to interpret the results in terms of the sharing rule and individual demands. So we would see that since Lambda's distribution factor independence, this is this variable year, we expected to be equal to the impact of changes in the sharing rule on household demand for the item and then if the effect of distribution factor and the effect of the distribution factor on the sharing rule is given by what is inside the brackets over here. So the data we use, we use wave one of the national income dynamic study and we try to restrict the sample to control for as much variation that we won't necessarily be able to control for by just popping in dummies into the equations. We restrict the household to two adult household members that are different gender either married or co-opening partners between the ages of 25 and 65 and where the household head is male. This is also done because this is what we see mostly in the data. We include households with up to three children since the observation of all of the variables we use restricts the sample greatly. And we use a seven broadly defined consumption categories namely communication, clothing, entertainment, food, medical expenditure, personal care and alcohol and tobacco. Locals general shares then calculated by using the data from the 21st, 2001 census which we then merged by district council to the NED data as this data is available. So this is just some of the summary statistics that I won't go over now. So for the unrestricted model I'm going to go through the preference factors and distribution factors since they are broadly similar or the preference factor is broadly similar to the results we see for the restricted model. So we see broadly that children are correlated with higher food and clothing expenditure, lower entertainment expenditure. We see that households residing in rural areas spend less on lower clothing and personal care. We see that asset ownership is associated with increased expenditure on entertainment, communication, medical and personal care. And we see that households with better education than household heads spend more on medical entertainment and communication. So in terms of the distribution factors we see that the local sex ratio along with its quadratic term and interaction with total income is jointly significant. And we see that the maternal education share and all its interactions also jointly significant along with and all distribution factors are also significant. So we fairly come to reject the unitary model of household behavior. So this is a bit more difficult so I'm going to start here at the bottom. So this is the proportionality condition. So this is the LRT statistics of whether or not equation two that we saw earlier is nested within equation one. So we cannot reject the hypothesis that the collective model of household behavior or we cannot reject the proportionality rule and thus we cannot reject that the collective model of household behavior is accurate in South African data. We see that household maternal education share, local sex ratio and both distribution factors are jointly significant. And then just to give an idea of what's going on here, we see we normalize the seemingly unrelated regression on clothing so that these variables over a year give us an indication of the relative impact of the different variables that affect bargaining power on the outcome. So we would expect communication to be more affected but in the same direction as clothing is affected by the distribution factors. Personal care and medical care is expenditure are also positively correlated and very close to clothing. With NC entertainment and food expenditure are insignificant and we see Elk on the back is negatively correlated with it. Negative and I think this is just above the 10% level of significance. So what for this one what it means is that if husband internal education share and household increases for example we would expect Elk on tobacco expenditure to increase. So yeah this is just going through what I just said and this is the relative impact of the sharing rule on consumption item expenditure. So this maybe gives a more intuitive presentation of the results. So what this graph shows us is we use the mean values of all of the variables including the other distribution factors to draw these graphs and this is normalized on clothing still. So what we see here is as husband internal education share increases we would expect female bargaining power which we assume is the case since most of the variables seem to indicate that because we would normally expect from other literature that is associated with female bargaining power or female preferences is correlated with this. We see a dramatic reduction in bargaining power every year in terms of clothing expenditure and as outside options which we model by local sex ratio increases we see that it slightly female bargaining power also slightly increases. We also see that the model is dramatically almost in favor of male bargaining power. We see a much more dramatic drop from year to year and from year to there. So we cannot really say that this is all causal but we do test some other specifications on these results. So the net dataset is very informative as it gives us some indication of which household member is responsible or if the household identifies which member is responsible for let's say day to day expenditure items and we can show that in the probe that the distribution factors are significantly correlated in the same direction on the probability that the female is the main decision maker on day to day household expenditure. Further the collective model also in the collective model of household behavior distribution factors should affect consumption patterns of married couples but not singles. So what we did is we basically we redid the effect of distribution factors on singles as well. We normalized husband maternal education share to the unit interval and then we just looked at the significance levels and we see that for both single men and single women the distribution factors are not significant in explaining their consumption of items. We then also go on to test other distribution factors so we added a distribution factor in the model to gauge how female bargaining power may be affected and whether the collective model is still valid. So we look at the income difference of the household during childhood, whether the husband's mother worked, marital status, living in a rural area, crown share, hourly wage share, and so on, all the rest. And this is the result we see. So we only, it seems like it got a bit there but I think it's an important one. So what we see here is the only ones that are really significant is the number of young children. So the average partial effect and the total effect of the distribution factor still remains significant although it's not as high as in the previous regression and we do not reject the proportionality test but this is also again at a fairly low level just above the 5% level. We're unable to reject the proportionality hypothesis. We see significance for large wage difference of the spouses and we see significance for the child support grant which is already received by the female household member. So with conclusions, the unitary model of household behavior is rejected for South African households and we see evidence in favour of the collective model. We see that household bargaining power is determined by various factors and it is important that it affects consumption outcomes in an observable way. We see that male household members are estimated to have the strongest relative preference for alcohol, tobacco, then followed by food and entertainment while increases in female bargaining power is associated with increases in communication followed by clothing, personal care and medical care. Yeah, thank you. It's my presentation. It's a bit faster than I thought.