First, many of the differences in reporting across studies are clouded by concerns about missing data. Finally, it is very difficult to measure giving at the top of the distribution without a high income surplus. Only one of the studies without such an overshoot produced data at the ninetieth percentile similar to those obtained in the only study with a high income overshoot.
First, many of the differences in giving measured across the studies are clouded by missing data. Only one of the studies without such an oversample (PS) provided data at the top of the distribution similar to that obtained with the high-income oversample in the NSP. The latest published analysis of attrition in the PSID concluded that it is still representative of the U.S.
In the same way as with the calculation of sharp limits, a lower limit is calculated for e.g. the median contribution by assigning all the missing components to zero (the lowest value they could have been) and. Likewise, the upper bound of the median is calculated by assigning total give to be at the sample maximum for those respondents with any item non-response in the components and determining the median of the resulting "upper bound" distribution.
The Data
A similar situation arises in the CSGVP where only 40 percent of married respondents reported all of their spouses' donations. For these reasons, lower bound amounts in the GVC and CSGVP may be less than those in the other surveys. This procedure identifies 86 percent of the CSGVP sample as heads, a percentage similar to that in the GVUS, GSS, and GVC.
Similarly, missing incidence responses for 13 of the 17 input components in the CSGVP were imputed, apparently “no”. Thus, the upper limit of incidence in CSGVP is somewhat underestimated. Missing data occurs in the determination of total award if the respondent does not know or refuses to answer the case or amount question for any of the award components that are asked. The frequency of missing incidence responses varies by specific component, of course, but on average 17.9 percent of responses are missing (ie, this is the average of the 12 percent missing in the GVUS component questions).
Because the GVUS and GSS questionnaires were identical (except for the use of expanded brackets), the lower rate of missing data in the GSS can be attributed to interviewer training methods used in university-affiliated studies. The missing data advantage of CSGVP, NSP and PS results in fewer respondents having missing data, as can be seen in the lower part of table 1.
Results
This also implies that the sharp upper limit of GUVS shown in the figure is somewhat underestimated. Overall, the figure shows the advantage of better response rates achieved in GSS, CSGVP, NSP and PS: the sharp edges are closer to each other. Indeed, this may explain the higher incidence measured in GVC compared to that measured using a similar amount.
These intervals are very narrow compared to the distance between the lower and upper limits in the GVUS, GSS and GVC, but in the same order of magnitude for the other three surveys. Large gifts make about a $100 difference in the means in the GVUS, GVC, and PS and have a negligible effect in the CSGVP and the NSP. 20Of course, some of this difference is due to the higher incidence of giving in the NSP.
But as the bottom portion of Table 4 shows, most of this difference remains in the average contingent amounts. This is entirely due to a single observation in the GSS (the two largest gifts in each survey are listed in the last two rows of the table). Regarding the conditional donations in the lower part of Table 4, the NSP and PS distributions are relatively close to each other.
22The number of positive gifts included at the bottom of Table 4 is slightly less than the incidence of giving reported in row 1. In contrast, the NSP and PS amounts fall more frequently in the upper GVUS deciles, especially the ninth (Figures 4.4 and 4.5). It is interesting to ask whether the differences between the other data sets and the NSP at the top of the distribution are simply a matter of the NSP measuring larger gifts over all, or, to control for differences in the central locations, the shapes of the distributions differ.
First, assume that the higher incidence in the NSP is accurate, consists of types of respondents who gave very small amounts, and that such respondents were not noticed by the PS. However, this adjustment would be too extreme if the higher incidence in the NSP is, at least in part, due to the high income oversample. In that case, the “extra” givers detected in the NSP would come from higher up in the conditional distribution, and not all from the bottom up.
Conclusions
The incidence of giving $500 or more to SCF is likely underestimated because it is based on a single question with very few. 25 Initially, it can be concluded that the higher incidence in PSK may be evidence that the surveys with lower incidence (GVUS, GSS and PS) have missed donors of large amounts (NSP, after all, measures giving largest of six polls). However, we recall that the SP matches the large giving in the PSK through the 92nd percentile.
The higher giving in the NSP after this point is probably due to the high income oversample rather than the higher incidence rate it achieved. This assumption is supported by the better missing data performance and higher incidence of giving in the input cue of GVC relative to GVUS and GSS (all of these studies used interviewers with limited dollar experience). This is similar to the finding that the PS measure of wealth matches that obtained with the high-income supersample in the SCF, although in the case of wealth the match rises above the 98th percentile.27.
Second, remember that the amount of missing data in the GVUS, GSS and GVC is extensive. For example, the comparison in the current paper assumes that the differences in populations due to time (in the case of the NSP) and country (in the case of the CNSGVP) affect the level of giving, but not the distribution of giving. In addition, smaller gifts may have been partially reported in the CSGVP because many were married.
One way to check whether the giving in PS and NSP is overestimated is to ask whether the giving at the top of these distributions corresponds to it. This figure taken from Kirsch, McCormack and Saxon-Harrold's (2001) general discussion of response in the GVUS series. There are 12 component questions on submission of GVUS and two parts (incidence and quantity) for each component.
631 Approximately how much money or cash property equivalent have you contributed to each of the areas listed above in the past twelve months. 53 In the last 12 months, have you (and your spouse/partner) made a charitable donation using payroll deductions. CQ03A In the last 12 months, have you made a charitable donation by responding to a postal solicitation.
CQ03B IN THE PAST 12 MONTHS, have you made a donation by paying to attend a charity event. Appearance in the PS indicates whether the respondent gave more than $25; in the other surveys there is no such threshold. Tests of equality of conditional distributions of giving with giving and volunteering in the US: Differences in forms conditional on equivalent location.
If the shift is listed in the table, it means that the comparison distribution and the GVUS distribution shifted in the indicated manner had similar shapes at those quantiles.
