The major committee of Table cuatro presents the newest baseline outcomes for the fresh instrumented impoverishment regression

Estimates

Very early adolescent marriage and you can shedding away from senior school each other keeps large outcomes towards opportunities a female will become when you look at the impoverishment. New prices mean that marrying young is actually of this a thirty.six fee section increase in the chances of staying in poverty. Dropping out-of high school are of the an eleven.dos percentage area rise in poverty.

I today expose a series of choice estimate ways to evaluate the new robustness of the standard result. Desk step one revealed that aggregation produced a large change having OLS estimates: the person-height overall performance recommended a little impression out-of very early teenager marriage with the poverty, as class-height OLS prices suggested a massive effect. The initial line for the Table 5 repeats the new baseline IV investigation, however, this time with categorized research. The information and knowledge try aggregated at state out of delivery ? season away from birth ? census seasons peak. The fresh classified-study IV estimates is amazingly similar to the individual-top IV rates (0.314 versus 0.306 having very early adolescent relationship and you will 0.112 instead of 0.112 to possess dropout). This new similarity of your coefficient prices isn’t alarming given that instruments is actually lingering for all individuals in a state-cohort class, effortlessly aggregating both personal-level and you may group-height quotes. The product quality errors together with alter very little with all the grouped study, increasing by regarding ten% to possess very early teen wedding and never anyway to own dropout. Once the aggregated investigation supplies very similar point estimates and you will quite even more traditional practical mistakes, in what observe, We present outcomes for aggregated study except if otherwise noted. 16

Dining table 5

Notes: Standard errors, adjusted for clustering of the county away from delivery, get from inside the parentheses. All of the regressions tend to be dummy details to have census seasons, competition, age, county regarding birth, and you can cohort of birth, and you will area for delivery trend. Investigation was aggregated to state from beginning ? season of delivery ? census year cell mode. See the cards so you can Dining table step one . New migration-adjusted approach was described in the text in addition to handle mode approach are explained from the text while the appendix.

The following activity is to gauge the impact migration is wearing this new task away from condition rules having relationships, schooling, and you may works plus the further IV estimates. While the a note, statutes is actually tasked based on a beneficial woman’s condition away from delivery, even in the event essentially we want to make use of the state a lady lived-in on age 15. Because particular girls keeps moved out of their delivery condition and you may to the your state that have yet another band of guidelines because of the years 15, the instruments are counted which have error. I determine exactly how which impacts the brand new IV prices inside column 3 from Dining table 5 .

To see how I examine the issue, notice that the expected value of the ideal (but unobserved) state laws can be calculated if migration probabilities are known. Let z_ij * be a dummy variable indicating the state law woman i faces at age 15, given she was born in state j. The asterisk indicates that this variable is not observed, given that she may have moved from her birth state by age 15. However, if migration probabilities are known, the expected value of this variable can be calculated as

where p_jk represents the probability that a woman will live in state k at age 15 given that she was born in state j, and w_k is the law in force in state k for the relevant year. The same logic applies when there are several variables for the state laws.

It is straightforward to show that substituting in E[z_ij * ] for z_ij * yields consistent estimates in an IV framework. The remaining issue is how to consistently estimate the conditional migration probabilities, p_jk. Although this information is not available for all women, the migration patterns for women who were age 15 at the time of the census enumeration can be estimated because the census records both state of birth and state of current residence. I use 15-year-old women in the 1960 census to estimate these migration probabilities. I then calculate the expected value of the laws based on the state a woman lived in at age 15 as outlined above and use these expected laws as instruments. 18 The migration-adjusted estimates in column 3 of Table 5 are very similar to the baseline estimates, indicating that the assignment of state marriage laws based on state of birth is a reasonable approach.