Eco572: Research Methods in Demography
|
Eco572: Research Methods in Demography | |
| ||
File sipp01w3741.dat in the datasets section
has data on age at marriage for women aged 37 to 41 in the 2001 panel
Wave2 of the Survey of Income and Program Participation (SIPP).
Note that age at marriage is coded as missing for single women.
(These data were used in the course last year and have been analyzed
by Josh Goldstein.)
(a) Compute a life table of entry into marriage including qx, dx, lx, and Lx. Treat women married at ages 37 to 41 completed years as single at exact age 37.0 to avoid problems with incomplete exposure. Estimate time lived in the single state up to age 37.0.
(b) Use the estimated life table to answer the following questions: (i) What's the probability of marrying by (exact) age 20 ? (ii) by (exact) age 25? (iii) by (exact) age 25 if the woman is single on her 20th birthday? (iv) by (exact) age 25 if the woman is single on her 20th birthday but she will marry eventually?
(c) Fit a Hernes model. here is a simple way to do this. In the model the ratio of the hazard to the cumulative marriage rate is h(a)/(1-l(a)) = e-r(a-15). Estimate the left hand side using the discrete equivalent qx/(1- lx), take logs, and regress that on age-15 (using age midpoints) to obtain a preliminary estimate of r. The model also says that
where p is the proportion who ever marry. Use your estimate of r to construct the exponential term (using exact ages) and regress the logit of the marriage function estimated in part a on the constructed variable to estimate p and A/r. Plot the observed and fitted survival (or its complement, cumulative marriage) and density functions.
(d) Fit a Coale-McNeil model. Start by estimating the conditional mean and standard deviation of age at marriage for women who marry by exact age 37.0. How would the eventual mean and standard deviation differ given that the experience of this cohort is almost but not quite complete? Goldstein and Kenney predicted using 1995 data that about 88.7% of this cohort would marry. Use this value and your best guess for the mean and standard deviation to plot observed and fitted survival and density functions, preferably in the same graphs as part b. (You should find that neither model fits very well, particularly in terms of densities.)
Files dr75x.dta and dr80x.dta have
extracts from the 1975 and 1980 WFS surveys in Dominican
Republic. The data include the date of interview, date of birth
of respondent, date of first union, dates of birth of up to 24
children, and childhood and current type of place of residence.
I called the date of first union udat and
the dates of birth bdat1 to bdat24.
I think the former is less confusing than m012 and
the latter are easier for looping than the actual WFS names as
used in my Colombia example.)
Use one of these datasets to fit a Page model of fertility since first union by age and duration since first union, allowing for effects of urban/rural residence on the spacing and limiting components of fertility.
Make sure you count events and exposure in a window that starts three years before the survey or with first union, whichever happens later, and ends the month before the survey. Compute age and duration since first union as of the mid-point of the window.
Use Poisson regression, treating the count of births as the outcome variable and the product of exposure time in years and natural fertility at the woman's age as an offset or fixed part of the linear predictor.
Treat type of place of residence as a dichotomy using a dummy variable for urban residence (so rural is the reference category). Make sure you introduce a main effect and an interaction with duration.
Write a short paragraph interpreting the results in terms of urban-rural differences in spacing and limiting behavior.
Use the same extract from part 2 to construct a life table showing the transition from second to third birth for all women who had a second birth in the ten years before the survey, separately by childhood type of place of residence. Exclude cases where the second and third births were twins.
Plot the estimated birth functions for each place of residence for intervals of up to 72 months (all on the same graph to facilitate comparison).
Summarize your results in terms of the quantum and tempo of the transition to third birth. By comparing results obtained by the students working on the 75 and 80 surveys we will also get a handle on time trends in quantum and tempo for the third child.
The DHS surveys conducted in the Philippiness have tracked a declined in the TFR from 4.09 in 1993 to 3.73 in 1998 and 3.50 in 2003.
Use Bongaarts's proximate determinants framework to asses the roles of marriage, breastfeeding, and contraception in this decline, combining abortion with the residual for lack of reliable data.
You will find most of what you need in the DHS reports. However, they report marital fertility by duration of marrige (as one should) not age, although you will find proportions married and time spent in marriage, so use your ingenuity.
You will also need estimates of the effectiveness of various contraceptive methods. You may use the data in Bongaart's original paper or more recent estimates, applying the same values to all three surveys.
You may collaborate with two other students by splitting the work so each of you collates the data for one of the surveys, but you should each write your own analysis.