The Effect of InterCensal versus PostCensal Weighting: The Case of Fertility

Fertility and the 2010 CensusGretchen Livingston glivingston@pewhispanic.org

Jeff Passel jpassel@pewhispanic.org1Im going to present data examining how the 2010 census results affect fertility outcomes1Sources of Fertility Data NCHSProvides number of births from birth certificatesResults can be combined with ACS (or CPS or pop estimates) to produce rates

ACSProvides info on #births in past yearAlso provides a denominator to produce rates

ACS weights based upon annual population estimates

2Two primary means of estimating fertility are the ACS and of course NCHS vital statistics data.

While these two sources are not quite comparable, both are considered pretty reliable for general estimates of fertilitydifferent time periodsacs is a longer perioddifferent racial/ethnic categoriesnchs is at time of birth; acs is retrospectivenchs measures babies; acs measures new moms

The NCHS data is derived from birth certificates, and measures all births occurring in a calendar year in the US.So it is a good measure of number of births, and in terms of just getting these levels, it is unaffected by the Census 2010 results.

However, you must use additional sourcestypically the ACS, pop estimates, or CPS-- to get a denominator to estimate rates. And all of these sources are, indeed, affected by the 2010 Census results.

ACS data asks women ages 15-50 if they have had a birth in the past year.So obviously allows for estimate of number of births, and also provides a denominator that allows for rate calculation.The weighting of the ACS is based upon the Census annual population estimates, which in turn are based upon Census estimates.

Noteacs data are not fully comparable across years, since thye are not revised when pop estimates are revised*************************************************************************************************Population Estimates -- Census Bureau -- Denominators by Demographics & Geography -- Definitions of Race Differ Somewhat -- Inconsistencies Over Time (by Vintage) -- Each Vintage has Consistent Series to Census

22010 Census, Population Estimates, and FertilityNew Census Data Requires Re-Benchmarking Pop Estimates

Other Recent Changes in Population Estimates Immigration Measure changed in 2007 and 2008Immigrant Age Distribution changed in 2009

3So it is the pop estimates that link fertility measures and the census.

First, as is always the case when a census is completed, pop estimates should be adjusted accordingly.the 2010 pop estimates (based upon 2000 data) do not match up with Census 2010 results

In addition to the basic revisions of the pop estimates and ACS weights;

it is also the case that over the past 5 years several other methodological changes have occurred in the pop estimates, that differnetialy impact estimates of certain groups in the ACS, and thus the fertility measures as well:

In 2007 and 2008, the estimates of immigration in the pop estimates were changed**** immigration estimates were changed from a calculation based upon change in foreign born to a calculation based upon residence one year ago**** the age structure of the foreign-born was also changed, in 2009 and 2010

Vintage 2007 (ACS 2007, CPS 2008) -- Immigration by Old Method (FB Difference) -- Immigration by New Method (ROYA)

Vintage 2008 (ACS 2008, CPS 2009) -- All post-00 Immigration by ROYA

Vintage 2009 (ACS 2009, CPS 2010) -- Changes in Age Structure of Immigration, not Level

Vintage 2010 (CPS 2011) -- More Changes in Age Structure not Level

Census 2010-based (ACS 2010, CPS 2012) -- Large Differences in Hispanic & Asian from 2009 -- Intercensals Provide Revised 2001-2009 -- No Planned Revisions to ACS Weights

And Ill touch upon those in a minute.

So to cut to the chase, what both the census revisions, and the changes in population estimation methology have done is led to:Higher estimates of foreign-born; asian; and hispanic womenHigher estimates of foreing-born; asian; and hispanic births;

When rates are calculated using ACS data for the numerator and denominator, there is minimal change in fertility.However, calculating rates using the nchs data for the numerator and the acs data for a denominator using revised data produce LOWER estimates than those produced using the original ACS dataAnd thus lower estimates of fertility rates for these same three gorups3Revised ACS WeightsApplies Census Methods to IPUMS

Creates Estimates that Match Intercensal Population Estimates

4The census bureau does not typically retroactively update ACS data with revised weights, so jeff passel did so on his own.He used an approach similar to the census bureau to ed revised ACS weights, using an approach similar to that used by the census when they create their annual ACS weights using population estimates.

So the results are a consistent set of data that approximates the weights that would have been obtained from a full ACS weighting based upon intercensal pop estimates.

He creates a bridged race variable

Preserves the age/sex/race composition at the state leveladjustes it to match intercensal estimates4Differences in 2009 Original vs. Revised ACS: Females 15-49#ACSREV- #ACSORIG%Diff15-19270,0002.620-2440,0000.425-29-30,000-0.330-3426,0060.335-3900.040-44-30,000-0.345-4940,0000.35Now we can move to the results, comparing the original or postcensal ACS to results based upon the revised ACS .Im focussing on the year 2009 here, which had larger discrepancies than the earlier years.

For the full population and for many subgroups, the differences between these two estimates are quite small.For instance, here is a comparison of totals for women of childbearing age.

The first column shows the absolute numerical difference between the revised and original estimates, and the 2nd column shows the percent difference.So for females 20-24, the difference in the revised and original estimates is about 40,000 which is a .4% difference.And for most other age ranges the results are similarso overall both estimates are quite similar5Differences in 2009 Original vs. Revised ACS: Females by NativityNative BornForeign Born#ACSREV- #ACSORIG%Diff#ACSREV- #ACSORIG%Diff15-19211,7602.259,8578.520-24-69,631-0.8105,3259.225-29-160,625-1.8137,1198.330-34-56,348-0.782,3544.135-39-56,447-0.755,7492.640-44-70,980-0.849,6182.545-49-12,753-0.152,5812.86Here we show the same data, for women, by nativity.Again for the native born, differences between the revised and original data are minimal.

Looking at the second panel, though, it is apparent that the revised estimates show consistently higher numbers of women, particulalry in the prime childbearing ages. For instnace, among women 20-24, the revised estimates are about 9% higher.6Differences in 2009 Original vs.Revised ACS: Females by Race/Ethnicity#ACSREV- #ACSORIG%Diff#ACSREV- #ACSORIG%DiffWhiteHispanic15-19-27,648-0.415-19185,7109.520-24-196,219-3.120-24136,8597.525-29-236,091-3.725-29157,8728.530-34-104,902-1.830-34119,9386.635-39-87,660-1.435-3987,3985.040-44-96,725-1.440-4478,3324.945-49-69,036-0.945-4960,4264.3BlackAPI15-1953,0863.315-1960,61014.720-24-848-0.120-24103,79122.925-29-27,501-1.925-2985,81614.830-34-3,980-0.330-3421,0833.335-39-11,760-0.935-3910,1801.540-44-12,879-0.940-448,6381.545-4919,9121.445-4921,0353.87Here is the same data setup, but breaking down women by race/ethnicity. The first panels show blacks and whites, and there differences between the revised and original data are small.

The second panel, tho, reveals quite dramatic differences for all hispanic womenfor instance the revised estimates show 8.5% more women ages 25-29 than the original estimates.

And again, if you look at the lower part of that 2nd panel, a similar story emerges for asian womenthe revisions based upon 2010 census suggest that the population of asian women ages 25-29 is about 15% higher than estimated originally.

So again, for most groups the revised estimates look a lot like the original estiamtes. Im going to focus on these three groups where we do see discrepancies tho---foreign-born; hispanic; and asian pacific islander women.7ACS Estimates:Foreign-Born Females 15-498Here Im showing the total number of womenin this case foreign-born womenages 15-49, from 2005 thru 2010.The solid line represents the original acs estimates; the dashed line represents the revised version.

8

ACS Estimates: Hispanic Females 15-499Here I show an analogous chart, but focussing on hispanic women instead. Again its a similar pattern9ACS Estimates:API Females15-4910And here is the same chart, but focussing on asian pacific islander women10Births: Foreign-Born Females 15-4911Births: Foreign-Born Females 15-4912Births: Hispanic Females 15-491313Births: Hispanic Females 15-491414Births: API Females 15-491515Births: API Females 15-491616GFR: Foreign-Born (15-49yrs)1717GFR: Hispanics (15-49yrs)1818GFR: Asian and Pacific Islanders (15-49yrs)1919ConclusionsEffect of Revised Weights on Fertility MeasuresMore Foreign-Born, Hispanic & Asian WomenMore Foreign-Born, Hispanic & Asian Births

Lower Rates when using NCHS/ACS formulationRates unchanged when using ACS to calculate numerator & denominator

Pop Estimate Adjustments Affect ACS ComparisonMethodological changes 2007-2009Affect year-to-year comparisons

Revised Weights Essential Change measures require consistencyCensus NOT planning to reweight ACSMulti-Year datasets also affected

20Impacts on ACS Data & NCHS-based Rates *More Asian, Hispanic, Foreign-Born Women *More Asian, Hispanic & Foreign-Born Births *ACS Internal RATES not Affected *Rates based upon NCHS/ACS reduced

Changed methodsso creates a problem with cross-year comparisonsAffect changes in fertility rates across time20Jeff Passeljpassel@pewhispanic.org

Gretchen Livingstonglivingston@pewhispanic.org

21Annual % Change in API Females 15-492222Adjustments to Pop Estimates Vintage 2007 (ACS 2007, CPS 2008) -- Immigration by Old Method (FB Difference) -- Immigration by New Method (ROYA)

Vintage 2008 (ACS 2008, CPS 2009) -- All post-00 Immigration by ROYA

Vintage 2009 (ACS 2009, CPS 2010) -- Changes in Age Structure of Immigration, not Level

Vintage 2010 (CPS 2011) -- More Changes in Age Structure not Level

Census 2010-based (ACS 2010, CPS 2012) -- Large Differences in Hispanic & Asian from 2009 -- Intercensals Provide Revised 2001-2009 -- No Planned Revisions to ACS Weights

23