The American Panel Survey (TAPS)

The director of TAPS, in consultation with other scholars, established a battery of demographic and other questions that are asked of all panelists. In addition, a large set of recurring questions were asked about economic and political subjects. 

Technical Description

Sampling Methodology and Recruitment

The sample of addresses was drawn from the U.S. Postal Service’s computerized delivery sequence file (CDSF).  The CDSF covers some 97% of the physical addresses in all 50 states including P.O. boxes and rural route addresses.  Homes that are vacant or seasonal are identified as are other categories that help to refine the efficiency of the sample to be mailed.  Using data from available U.S. Census files plus from a variety of commercial data bases, such as White Pages, Experian, Acxiom, etc., MSG can add names to these addresses, match with landline telephone numbers, and—with some level of accuracy—tag on information regarding race/ethnicity, age of householder, whether there are people of a certain age in the household, presence of children, home ownership status, etc.

Based on recent experience with the recruitment of an online panel with the ABS frame, TAPS strata were designed to specifically break out young adults (ages 18-24) and Hispanics, in addition to the balance of the population.  Young adults and Hispanics may be strategically oversampled because these groups have a tendency to under-respond to surveys.  Four mutually exclusive strata were used:

  1. 18-24 year-old Hispanic adults
  2. All other Hispanic adults ages 25+ or age unknown
  3. 18-24 year-old non-Hispanic adults
  4. All other adults that are non-Hispanic or ethnicity unknown and ages 25+ or age unknown
  • The estimated yield from each of the above strata was 5.6%, 6.4%, 14.4% and 9.4%, respectively.
  • Within-household selection procedures vary by the mode in which the household responds to the initial contact.   
  • A successful recruitment was counted only when a Profile Survey is completed.

The resulting sample design and expected outcome is as follows:

Stratum Mailing count Mailing distribution Yields proportion Yields count Profiled proportion Profiled count Strata distributions sample Strata distributions frame
1. Hispanic 18-24 333 1.0% 0.056 19 0.65 12 0.6% 0.2%
2. Hispanic 25+/unk 6,943 19.9% 0.064 445 0.65 289 14.4% 14.1%
3. Other 18-24 459 1.3% 0.144 68 0.65 44 2.7% 0.7%
4. All Else 25+/unk 27,071 77.8% 0.094 2,550 0.65 1,657 82.8% 85.0%
Overall 34,816           100.0% 100.0%

 

 

Weighting

TAPS calculated weights to make survey results generalizable to the U.S. population of English-speaking adults.   Investigators received these weights as variables in delivered data files.

Adjustments were made to compensate for (a) selection probabilities altered by the stratified sample design and (b) within household selection probabilities associated with the random choice of a panel member from among all eligible adults residing in the household.   These adjustments constituted the base weight that corrected the sample to approximate a simple random sample of the population of adults.

The following weight, w1i|k, for mailing addresses i within stratum k is calculated as follows:

w1i|k = (Pi|k / Ptot) (Stot / Si|k)

where

  • Pi|kis the population or frame count within stratum k,
  • Ptot is the total population count from the frame,
  • Si|k is the sample count within stratum k, and
  • Stot  is the total recruited sample size.

We also adjusted for the selection probability of the randomly selected adult within households.  To adjust for this, we weighted each selected respondent, r, by the inverse of the number of eligible adults, A, ages 18 and older, enumerated as residing in household, h, and called this w2r|h  and calculated as follows:

w2r|h  = Ah / 1

The base weight was the product of  w1i|k and  w2r|h .[1]

_____________

[1]  An additional adjustment was made for our follow-up efforts to recruit panelists by telephone after an initial nonresponse.  Households recruited based on the telephone follow-up may have over-represented addresses in the sample with a successful telephone match.  Based on the natural match rate in the sample, recruited households from the telephone recruitment was weighted downward appropriately.

Response Rates

Response Rates were calculated by GfK/Knowledge Networks for each month's survey. They are available by request.

 

TAPS Reports     Taps Data Archive

People

Steven Smith

Jonathan Rapkin

Melinda Warren

Director, The American Panel Survey (TAPS)
Director, Weidenbaum Center on the Economy, Government,
and Public Policy
Kate M. Gregg Distinguished Professor of Social Sciences
Professor of Political Science

Statistician, The American Panel Survey (TAPS)
Technical Support Specialist and Data Analyst

Coordinator, The American Panel Survey (TAPS)
Director, Weidenbaum Center Forum