Analysis of twins

In cross sectional and time series data, the number of children in a family tends to exhibit a negative association with both the educational attainment of children and the labour market outcomes of women. [...] To ensure that the treated and control samples are balanced on birth order they propose estimating the model by the birth order at which the potential twin birth occurs, and conditioning the sample on having at least as many children as are implied by the particular birth order. [...] (2010), where the endogenous variable of interest is multi valued, rather than binary, the LATE must be cast in terms of the variation which the instrument generates in the endogenous variable of interest. [...] The analysis is now implemented at the level of the parent, so there are no sample restrictions but the approach still allows the relationship to vary with the parity at which twins occur. [...] The fact that the twin instrument does not allow for the consideration of impacts of high fertility in a family on the later born children (in cases where the later born are born after twins) is a limitation.