Democratizing access to the best data on religion since 1997

Archive Navigation

View Question Bank

Archive Categories

Sort the above Archive Category by:

Browse Alphabetically
Browse All Categories
Browse Newest Additions

File Summaries

Search Data Archive

Philanthropy Panel Study, 2003




Wilhelm, M. O., Brown, E., Rooney, P. M., Steinberg, R., Stafford, F. P., Schoeni, R. F., & McGonagle, K. (2021, August 23). Philanthropy Panel Study, 2003.


The Philanthropy Panel Study is the Philanthropy Module of the Panel Study of Income Dynamics (PSID). The 2003 Panel contains data on the charitable giving, volunteering, and religious service attendance of 7,822 American families. The charitable giving data describe the giving done by the family unit as a whole. The volunteering and attendance data are separately available for both "Heads" and "Wives" (PSID terminology) in married couples and co-habiting families. The Panel also contains a question about who in married couples makes decisions about charitable giving.

The charitable giving data include religious giving and the volunteering data include religious volunteering. The religious giving and volunteering data- along with the religious attendance and religious affiliation data- make the Panel well-suited for the study of important religious behaviors within the PSID's rich context of families' economic, social, health, and demographic circumstances.

The ARDA has added three additional variables to the original data set to enhance the users' experience on our site.

Data File

Cases: 7822
Variables: 138
Weight Variable: FAMWGT01 (NOTE: Frequencies and means in codebook are unweighted)

Data Collection


Funded By

Atlantic Philanthropies
Bill and Melinda Gates Foundation

Collection Procedures

The PSID was collected in face-to-face interviews using paper and pencil questionnaires between 1968 and 1972. Thereafter, the majority of interviews were conducted over the telephone. In 1993, the PSID introduced the use of computer-assisted telephone interviewing. In the 1999 wave, 97.5% of the interviews were conducted over the phone, and all interviews were conducted using computer-based instruments.

For more information concerning the PSID, click here:

Sampling Procedures

The PSID sample, originating in 1968, consisted of two independent samples: a cross-sectional national sample and a national sample of low-income families. The cross-sectional sample was drawn by the Survey Research Center (SRC). Commonly called the SRC sample, this was an equal-probability sample of households from the 48 contiguous states and was designated to yield about 3,000 completed interviews. The second sample came from the Survey of Economic Opportunity (SEO), conducted by the Bureau of the Census for the Office of Economic Opportunity. In the mid-1960's, the PSID selected about 2,000 low-income families with heads under the age of sixty from SEO respondents. The sample, known as the SEO sample, was confined to Standard Metropolitan Statistical Areas (SMSA's) in the North and non-SMSA's in the Southern region. The PSID core sample combines the SRC and SEO samples.

Principal Investigators

Mark O. Wilhelm (PI of the Philanthropy Panel Study)
Department of Economics
Indiana University-Purdue University at Indianapolis (IUPUI) and
the Lilly Family School of Philanthropy at Indiana University
Indianapolis, Indiana 46202

Co-PIs of the Philanthropy Panel Study:
Eleanor Brown
Department of Economics
Pomona College
Claremont, CA 91711

Patrick M. Rooney and Richard Steinberg
Department of Economics
Indiana University-Purdue University at Indianapolis (IUPUI) and
the Lilly Family School of Philanthropy at Indiana University

Frank P. Stafford (PI of the Panel Study of Income Dynamics).
Survey Research Center
Institute for Social Research
University of Michigan
Ann Arbor, MI 48106-1248

Co-PIs of the Panel Study of Income Dynamics:
Robert F. Schoeni and Katherine McGonagle
Survey Research Center
Institute for Social Research
University of Michigan

Related Publications

Wilhelm, Mark O. 2006a. "New Data on Charitable Giving in the PSID." Economics Letters 92(1, July): 26-31.

Wilhelm, Mark O. 2006b. "The Quality and Comparability of Survey Data on Charitable Giving." Nonprofit and Voluntary Sector Quarterly, forthcoming.

Wilhelm, Mark O. 2006c. "Volunteering Data in the 2003 Center on Philanthropy Panel Study: A Supplement to the User's Guide." Mimeo, IUPUI.

Wilhelm, Mark O. 2006d. "The 2003 Center on Philanthropy Panel Study User's Guide." Mimeo, IUPUI.

Using the Data

It is strongly advised that you consult the User Guide for this data before you begin your analysis. It can be found at:

The documentation you are reading is in draft form, so there will likely be new drafts with minor changes in response to readers' comments. Periodically check the following web site: for new drafts. It is much less likely that there will be changes in the extract data themselves, but if there are changes you will be able to download the revised data from the web site.

The extract data are easy to use, but it falls to the user to use the data responsibly-this means learning about the PSID. For example, say the user calculates simple averages from the extract data. The user is already making a big mistake because he or she is ignoring the fact that the PSID has a low-income oversample (see Section 5). A good place to start learning about the PSID is Martha Hill's (1992) excellent user's guide. Also, the on-line tutorials are excellent (see Section 4).

The extract data are provided as a favor to those interested in the analysis of giving, volunteering, and religious service attendance. I have worked to find and fix problems in the data (especially in the volunteering data), but I do not guarantee that all problems have been found and fixed. If after making a serious effort to understand the data (e.g., reading all of this documentation, reading the questionnaire, reading Martha Hill's (2002) introduction to the PSID, and reading about the PSID on the PSID's web site), you think there is a problem with the data in the extract, contact me (This email address is being protected from spambots. You need JavaScript enabled to view it.). Do not contact the PSID help e-mail with questions about the data in this extract (the PSID has agreed to the posting of this extract with the understanding that I am the person to whom questions about the extract will be sent). Finally, you use these data "as is"-the final responsibility of any conclusions you draw in research using these data rests with you.


The suggested citation for this file is:

Wilhelm, Mark O., Eleanor Brown, Patrick M. Rooney, and Richard Steinberg. 2003. The Center on Philanthropy Panel Study [machine-readable data file] / Director and Principal Investigator, Mark O. Wilhelm; Co-Principal Investigators, Eleanor Brown, Patrick M. Rooney, and Richard Steinberg; Sponsored by Atlantic Philanthropies. In the Panel Study of Income Dynamics Wave XXXIII [machine-readable data file] / Director and Principal Investigator, Frank P. Stafford; Co-Principal Investigators, Robert F. Schoeni, Jacquelynne S. Eccles, Katherine McGonagle, and Wei-Jun Jean Yeung. Ann Arbor: Institute for Social Research, The University of Michigan.

In your acknowledgments, you can also thank Atlantic Philanthropies for funding the collection of data in the 2001, 2003, and 2005 waves, and the Bill and Melinda Gates Foundation for funding the 2007 and 2009 data collection as well as this data dissemination work. Data are a public good, but someone has to pay for them-by expressing appreciation, users encourage funders to continue supporting data collection.


FAMWGT01: This weight variable is used for analysis of all 2003 families, including the immigrant sample families that were added in 1997 and 1999 as well as the PSID core families.

Families that include at least one sample person with a non-zero individual weight are assigned a positive value for this variable. This positive value is the average of the individual weights of all members of the FU, whether or not they are sampled. Certain families, however, have values of zero for this variable. There are several reasons for the zero weight assignment:

(1) Families that were initially dropped from the SEO sample in 1997 but reinstated because of the intense interest in children and child development have zero weights.

(2) Families that are comprised only of non-sample individuals have zero weights. Most of these non-sample individuals are being interviewed because they are parents of sample children (but the children reside in another panel FU).

(3) Families whose only sample people have zero individual weights are assigned zero family weights. In most cases, these sample people have zero individual weights because they were recontacts at some earlier point. For further information on the criteria for these zero-value weight assignments, see the 1993 documentation, Part 5 "PSID Analysis Weights", Section 1.C, pp 23-24.

Charitable Giving Variables

The charitable giving variables are of two types: incidence and amount. An incidence variable is a binary indicator: Did the respondent give or not? Incidence variables are prefixed by "G." The second giving variable type is the dollar amount given, prefixed by "A." For example, Grelig is a dummy variable indicating whether the family unit gave to religious purposes or spiritual development and Arelig contains the amount given. All missing data have been imputed, but corresponding accuracy codes (Greligac and Areligac) indicate whether and what imputations were made. Missing data are not a big problem in the Center Panel (see Wilhelm 2006b).

There is not a "total giving" variable in the data (because the religious giving-independent variables relationship differs from the secular giving-independent variables relationship). However, total giving can be constructed by adding religious giving to all secular giving (Arelig + Asec10).

There was a change in the secular questions between 2001 and 2003. In 2001, after reporting giving to the four main types of secular organizations-organizations that have a combination of purposes (e.g., United Way), organizations that help people in need, educational organizations, and health organizations-the respondent was asked a series of yes/no questions about giving to five other types of organizations (youth and family services, the arts, neighborhoods and communities, environmental, and international) and a mop-up "any other" category. If the respondent said "yes" to, say, four of these six, the respondent was asked the aggregate amount given to all four. In 2003, the respondent was asked the separate amounts given to each of the five other types of organizations and the mop-up other category. Not surprisingly, reports of amounts given to these six are higher in 2003 than in 2001. The variable ASEC04 adds together giving to the four main types: combined purpose, help the needy, educational, and health organizations. ASEC04 is comparable to the similarly named variable from 2001. ASEC06 adds together giving to the remaining five types of organizations and the mop-up other category. The final question in the giving section (M13) asks about decision-making when the family unit decides about charitable giving.

I have analyzed the quality of the 2001 Center Panel giving data in a pair of papers that are included in the Archive.

Who Decides?
The final question in the giving section (M13) asks about decision-making when the family unit decides about charitable giving. The respondent is asked "who in your family was involved in decisions about how much support to give individual charities in 2002." Many respondents name multiple people from their family units. People who could have been named are: any current wave family unit members (including children age six and older) and any people in the family unit last wave who have since moved out. A dummy variable in the 1968-2003 Individual File identifies whether the individual was named as a charitable giving decision-maker by the respondent. I have summarized those dummy variables into three family-level variables: the head is a decision-maker, the wife is a decision-maker, and the number of other decision-makers in the family unit (M13h, M13w, and howmanyr).

In addition, two variables directly from the 2003 Family File give additional detail. If the respondent named both Head and Wife as decision-makers, the respondent was then asked if one person made most of the decisions, if the decisions were made together, or if the decisions were made separately; the responses are contained in the variable M13A_WHO. If the respondent said "one made most decisions" (M13A_WHO = 1) then the respondent was asked which (head or wife) made most of the decisions. The responses are contained in the variable M13B_ASQN. Remember, M13A_WHO was asked only if the respondent said both Head and Wife made decisions about charitable giving. If only one of the Head-Wife pair made decisions, then you have to look at M13h and M13w to figure which one.

Volunteering Variables

The 2003 volunteering questions were designed to be easy for the respondent to answer, but thereby generate data that are complicated for the analyst to use. This extract contains cleaned volunteering variables: annual hours volunteered through six different types of organizations and a seventh mop-up variable for hours volunteered through all other organizations. All problem data are imputed and there is a companion accuracy code variable for each volunteering variable.

For example, the head's hours of religious volunteering ArelgHLB and the accuracy code is ArelgHaccb. In the variable names "H" indicates "Head," LB indicates lower bound (because the imputations are a lower bound to the actual annual hours-for example: if the response is hours response is entirely missing the lower bound is zero hours), and "accb" indicates the brief version of the accuracy code (a more detailed version of the accuracy codes will be described below). Parallel variables for the wives are ArelgWLB and ArelgWaccb.

My recommendation for using the volunteering data is to get arelgHLB (continuing the example), the companion accuracy code ArelgHaccb, and drop the observations for which ArelgHaccb >= 890. Warning: If you do not drop the observations for which ArelgHaccb >= 890, you must look at these observations one-by-one and decide how you want to handle them. To make sure you do this one-by-one checking I have set arelgHLB = -999. Volunteering data without imputations are available in a separate file (see Wilhelm 2006c). The separate file contains a more detailed accuracy code that explains exactly what inconsistencies occurred (if any) when the respondent answered the volunteering questions.

Three other volunteering variables are available in the present extract. Two variables are from the initial screening questions (M14) about whether the respondent did any volunteering last month and last year: HMONTH and HYEAR. The corresponding variables for the wife are WMONTH and WYEAR. The raw variables in the PSID Family File or from the Data Center for volunteered last month/year (ER23550-ER23553) are for the respondent and respondent's spouse. The variables in this extract (e.g., HYEAR and WYEAR) have been switched from respondent/spouse to head/wife using the variable indicating whether the head or wife was the respondent (respondent2003).

The last volunteering variable adds together the annual hours volunteered through the six types of secular organizations. The secular annual hours variables AsecularHLB and AsecularWLB do not have companion accuracy codes, but are set to "-999" if any of the six component types had an accuracy code >=890.

Finally note that the 2003 volunteering questions are very different from the 2001 volunteering questions. The 2001 questions asked about volunteering ten or more hours, while the 2003 questions had no threshold. Probably more importantly, the 2001 hours question simply asked the respondent to estimate how many hours were volunteered last year, while the 2003 hours questions were completely redesigned.

This can be compared to the 2001 data. The percentages volunteering are higher in the 2003 data if the percentages are derived from the single question about whether any volunteering was done last year. For example, the percentage volunteering from the 2003 data is 31.1, compared to 27.4 from the 2001 data. The percentages volunteering in the 2003 data are closer to those in the 2001 data if the percentages from the 2003 data are derived from the hours questions. For example, the 2003 percentage volunteering in this case is 29.1. The median hours volunteered (conditional on volunteering) in the 2003 data is 80, much higher than the median of 52 hours from the 2001 data. In short, the "any volunteering" indicators may be comparable across 2001 and 2003, but the hours certainly are not. Volunteering questions in 2005 and subsequent waves will be identical to the 2003 questions.

The quality of the 2003 Center Panel volunteering data have not been analyzed at the same level of detail as the quality of the giving data. What we know is that the Center Panel volunteering data indicate similar volunteering incidence as does the Current Population Survey September supplement, but much higher hours volunteered (see Table 3). This makes sense considering the longer time spent in the Center Panel getting respondents to think about their hours of volunteering.

Religious Attendance

The religious attendance variables in the PSID Family File or from the Data Center (questions M53 and M54, ER23699-ER23702) are for the respondent and respondent's spouse. The religious attendance variables in this extract (attendH and attendW) have been switched from respondent/spouse to head/wife using the variable indicating whether the head or wife was the respondent (respondent2003).

Each attendance variable has a brief accuracy code indicating whether an imputation was made, and if an imputation was made, a subjective indication of the imputation accuracy.

Religious Affiliation

The religious affiliation questions are not re-asked in every wave but only asked when either the head or wife changed from the previous wave (head/wife changes occur when a family "splits-off" from another family or when a previously non-response family is recontacted and brought back into the study). For the heads and wives who did not change from the previous wave, the religious affiliation variables are "brought-forward" from the last wave in which the affiliation question was asked. The PSID staff did this "bringing-forward." The variables newhead2003 and newwife2003 tell you which heads and wives changed from the previous wave and therefore which were asked the religious affiliation questions in 2003.

Merging Other PSID Data

To add more data from the PSID 2003 Family File to the extracts, use the Data Center to draw the variables to be added (the 2003 family unit identifier ER21002 is automatically provided). Merge the drawn variables into the extract using ER21002 (after you rename ER21002 to fid2003).

To merge the 2003 extract into the 2001 Family File (or the Family File from any other year) read Tutorials 2 or 3. You cannot simply merge the 2003 family unit identifier to the 2001 family unit identifier.

Advice for New Users of PSID

Drop the family units with notasked=1 - these family units were not asked the giving, volunteering, and attendance questions, but their giving, volunteering, and attendance variables are still coded to "0."

When calculating descriptive statistics either (i) use the weights famwgt01 or (ii) work with the SRC subsample (subsamp=1) to avoid over-representation of low-income families (the SEO subsample).

Our Sponsors

Our Affiliates

© 2024 The Association of Religion Data Archives. All rights reserved.