Why Donors Give and Who They Are
Joan Mount Eugene Kaciak
Laurentian University and Brock University
Sudbury, Ontario St. Catharines, Ontario
Canada Canada
Introduction
Many organizations within our society depend heavily on the largesse of individual and corporate donors. These organizations span a very wide spectrum of activity from health, through education, through partisan politics, through protection of the environment, to name a few of note. A myriad of causes assail the householder, most commonly through mass mail campaigns, door to door solicitations, and telephone appeals. Researchers are curious about why donors give, and why some persons who are approached respond to some causes and not to others (Leslie and others, 1983; Leslie and Ramey 1988; Andreoni 1989; Lindahl and Windship 1992; Prince, File and Gillespie 1993). Some (e.g., Rosenblatt, Cusson and McGown, 1986; Brittingham and Pezzulo 1990; Mount, 1996) have proposed models to explain personal donorship.
The typical fundraising campaign meticulously targets a small number of potential donors who might be persuaded to give a "major gift". Much less attention is paid to the considerably larger number of benefactors who will give small gifts. The latter are often treated as an amorphous group, with little known about their motives for giving and little use made of segmentation strategies.
In order to answer the two title questions: "who are donors?" and "why do donors give?", we must find what sociodemographic and psychographic segments they belong to, as well as what motives can be associated with these segments. This task would be relatively easy if a survey showed that there are distinct clusters of respondents based on their motives for donorship. Unfortunately, in most cases targeted groups of respondents do not form any clusters across motives.
This paper explores an analytical technique that can help. This methodology has been labelled the Alternative Principal Components Method (Kaciak and Sheehan, 1987, 1988), and its potential value is as a tool to identify market segments within a seemingly homogeneous pool. In this paper we analyze such a case and propose a segmentation procedure that links sociodemographic variables with particular motives.
As a basis for discussion we use data from a survey of motives for supporting an educational cause - i.e. alumni giving (Mount and Quirion, 1988). These data were collected in June 1987 from a random sample of two hundred and eighty nine donors and non-donors to an alumni fundraising campaign conducted earlier in the same year at a small regional university in Canada. The 289 respondents to the mail questionnaire assigned a numerical score from ‑3 to +3 to each of 8 motivational factors (liking to be asked = LTBA, obligation = OBLG, nostalgia = NOST, payback = PAYB, tax credit = TAXC, sympathy = SYMP, joy of giving = JOYG, and altruism = ALTR). The scores represent the intensity of their motivating power (score ‑3 corresponds to "not at all motivating", score +3 corresponds to "strongly motivating"). In order to avoid negative numbers we transposed these scores to a 1 (=-3) to 7(=+3) equivalent scale, and thus obtained a 289x8 input data matrix of the motives ratings.
Arranging the motivators from the strongest to the weakest, we obtain:
The strongest donorship motivator for all respondents is ALTR, followed by NOST. The two weakest ones are TAXC and LTBA. We notice, however, that the above method of averaging scores across all 289 respondents produces very small differences among the motives – most of them are close to the trivial 0.125 (=1/8) level as if each motive was equally important for the group of 289 respondents.
Introductory Analysis
In order to have a better insight into the nature of philanthropy in this group of respondents we have to investigate whether distinct clusters of respondents are present, hopefully each one characterized by a different set of motives. Before settling on a particular clustering/segmentation method to use, it is good practice to look at the entire set of respondents as a group of points in space. We can then visually determine whether clusters of respondents are present or not. For this purpose we used the Multiple Correspondence Analysis (MCA) which is a technique for converting a set of multidimensional (in our case – 8-dimensional) points to projections on a 2-dimensional plane. It is as if we forced the invisible seeds in the watermelon to drop on a table under it. Then we could see them. Figure 1 below shows projections of respondents onto the plane defined by the first two principal axes generated by the MCA algorithm. The mechanics of the MCA algorithm are beyond the scope of this presentation. Those interested may find them in Kaciak (2000), Kaciak and Louviere (1990), Greenacre (1984), Lebart, Morineau and Warwick (1984).
Figure 1: MCA Projections of Respondents from the 8-dimensional Space of Motives
The strongest donorship motivator for Respondent No 1 is JOYG, followed by OBLG. The two weakest ones are ALTR and TAXC. Notice that each respondent is represented by an 8-dimensional set (vector) of different intensities of motives. We say that respondents belong to an 8-dimensional space of motives. One could imagine a watermelon with black seeds inside it which are invisible – each seed described by 8 numbers. Obviously, this watermelon would contain 289 seeds. The seeds are invisible not simply because they are inside the watermelon, but because they belong to the 8-dimensional space and the human eye can see objects only in 1-, 2- or 3-dimensional space.
In order to find out which motives are the strongest, and which are the weakest for the entire group of 289 respondents we calculated for each motive its average score across all 289 respondents and then normalized it so that the total of all scores would be 1:
Table 1: Normalized scores for the 8 donorship motives across the random sample of 289 respondents
It is obvious that there are no clusters of respondents visible on the plane, and therefore they are probably not present in the original 8-dimensional space (our watermelon) either. In such case, it does not make too much sense to use any of the clustering methods known in the literature. On the other hand, we would like to be able to partition our watermelon somehow into smaller pieces and to investigate each piece separately, in the hope of finding different patterns of motives among different segments. For this purpose, we will apply the segmentation procedure described for the first time by Kaciak and Sheahan (1988). They used a modified version of the classic Principal Components Method, which they called the Alternative Principal Components Method
(APCM), in order to project the multi- dimensional points on to the plane.
Segmentation Procedure (Step 1)
Figure 2 presents projections of respondents onto the plane defined by the first two principal axes generated by the APCM algorithm. We notice the strange shape of the projections – they are virtually stretched along the second principal axis with no variation along the first axis. In other words, the APCM algorithm converts the 8-dimensional points to 1-dimensional points. Kaciak (1987) proved that this would be the case for any data set that is submitted to the APCM algorithm. Using our watermelon analogy, we can say that the APCM algorithm arranges the seeds on the table in one line according to some pattern. We may suspect that the respondents projected onto the top of the second axis will differ in their donorship motives from respondents on the bottom. Since the projections are virtually 1-dimensional, Kaciak and Sheahan (1988) proposed to use well known one-dimensional techniques for segmenting sets of points, such as the arithmetic mean plus or minus the standard deviation.
Figure 2: APCM Projections of Respondents from the 8-dimensional Space of Motives
The arithmetic mean and the standard deviation of the projections’ coordinates on the second axis are .0004 and .0494, respectively.
According to the method of market segmentation introduced by Kaciak and Sheahan (1988), one can divide the group of 289 alumni into the following three subgroups:
A) alumni whose coordinates on the 2nd Axis are greater than 0.0498 (TOP part)
B) alumni whose coordinates on the 2nd Axis are in range [-0.0498, 0.0498] (CENTRAL part)
C) alumni whose coordinates on the 2nd Axis are less than ‑0.0498 (BOTTOM part)
We found that 11.8% of the alumni belong to Group A, 75.4% to group B, and 12.8% to Group C.
Now, we will attempt to find out which motives are the most and the least popular in each of the above three groups of respondents. Up to now, we were analyzing the set of respondents in the 8-dimensional space. The above mentioned multivariate methods such as MCA, PCM, or APCM allow us to analyze the data from the opposite point of view: the set of 8 motives in the 289-dimensional space. Our new watermelon has now only 8 invisible black seeds, each seed described by 289 numbers – ratings given to it by all respondents. Figure 3 shows projections of the 8 motives onto a plane defined by the first two principal axes of the APCM.
This time, the projections are not stretched along the second principal axis only, as we observed in the case of respondents. Kaciak (1987) proved that projections of motives (or any other variables) along the first principal axis are always arranged from the most important ones on the right to the least important on the left. He also proved that the normalized (the sum is 1) coordinates of projections on the first axis match the average scores allocated by the group of respondents to the variables (in our case – motives). In this case, the normalized coordinates are:
LTBA OBLG NOST PAYB TAXC SYMP JOYG ALTR
0.1125 0.1184 0.1346 0.1156 0.1125 0.1298 0.1280 0.1485
Notice a remarkable similarity between these coordinates and the normalized scores in Table 1, which confirms the Kaciak’s finding. One may therefore consider that the first principal axis in the 289-dimensional motive space represents an "UNPOPULAR/POPULAR" dimension. The motives on the left side of the axis (LTBA, TAXC, PAYB, OBLG) are
less "liked" by the 289 alumni, and so less motivating, than the motives on the right side, ALTR, NOST, and SYMP. In other words, as a motive for philanthropy to this particular cause, LTBA and TAXC are the weakest of the eight investigated, while ALTR and NOST are the strongest motivators.
We can also see from Figure 3 that the second principal axis clearly opposes OBLG on the top to ALTR on the bottom. We label this an "INDEBTEDNESS/SELFLESSNESS" dimension.
Figure 3: APCM Projections of Motives from the 289-dimensional Space of Respondents
(Axes 1 and 2)
Segmentation Procedure (Step 2)
In order to better examine our data we will project the 8 motives onto another plane, this time defined by the first and the third principal axes. Using our watermelon analogy, it as we have turned the watermelon around, say 90 degrees, and then forced the seeds to drop on the table. This exercise obviously allows us to view the watermelon from a different perspective (the more perspectives the better, but for the sake of brevity we limit our analysis to these two views only). The new projections are presented in Figure 4.
Figure 4: APCM Projections of Motives from the 289-dimensional Space of Respondents
(Axes 1 and 3)
This time we observe a new intriguing feature: the third principal axis clearly opposes TAXC on the top to NOST on the bottom. We label this a HEAD/ HEART dimension. Because of such a pronounced difference in the location of TAXC and NOST along the third axis, we will further divide the central group B of respondents into three subgroups using the Kaciak-Sheahan segmentation procedure.
First, we calculate the arithmetic mean and the standard deviation of the coordinates of projections of respondents on the third principal axis: 0.0002 and 0.0458, respectively. (In the interest of saving space we do not present the relevant APCM figure of projections of the 289 respondents on to the plane generated by the 1st and the 3rd principal axes.)
Second, we split the CENTRAL group B of respondents into three subgroups:
B1) alumni B whose coordinates on the 3rd Axis are greater than 0.0460 (CENTRAL-TOP)
B2) alumni B whose coordinates on the 3rd Axis are in the range [-0.0456, 0.0460] (CENTRAL-
CENTRAL)
B3) alumni whose coordinates on the 3rd Axis are less than ‑0.0456 (CENTRAL-BOTTOM)
We found that 8.0% of alumni belong to Group B1, 60.2% to group B2, and 7.3% to Group B3.
Segmentation Procedure (Final Results)
In summary, the average scores allocated to the 8 motives by each of the above five subgroups of alumni are as follows:
Group A (11.8%):
Group B1 (8.0%):
Group B2 (60.2%):
Group B3 (7.3%):
Group C (12.8%):
Arranging the motives in each group from the most to the least important one, we obtain:
Notice that the CENTRAL-CENTRAL Group B2 has motives arranged in the same way as the entire group of 289 respondents (Table 1). However, this group accounts for only 60.2% of the total number of respondents. In other words, we managed by employing the Kaciak-Sheahan procedure to peel off almost 40% (39.8%) of respondents from the entire sample. Additional analysis of this 40% gives us additional insight into other possible arrangements of motives governing donorship choices – besides those which govern the “average” 60% majority.
We draw the following conclusions from Figures 3 and 4:
1. There are pronounced differences in the 289 respondents' attitudes towards the motives.
2. There is a group of respondents who are guided simultaneously by OBLG, PAYB, LTBA, and - to some extent - JOYG, in their choice whether to donate or not. We notice that Group A matches this description very closely. This group is located on the left (L) of Axis 1, on the top (T) of Axis 2, and in the middle (M) of Axis 3.
3. The second group of respondents clearly focuses on TAXC with some links to ALTR and SYMP. Group B1 seems
to be the best representative for these motives. This group is located in the middle (M) of Axis 1, on the bottom (B) of Axis 2, and towards the top (T) of Axis 3.
4. The third group of respondents is most likely to be guided by NOST, followed by ALTR and SYMP. We notice that Group B3 defined above matches this description to a marked degree. This group is located on the right (R) of Axis 1, in the middle (M) of Axis 2, and towards the bottom (B) of Axis 3.
5. Group C of respondents matches basically the average profile of the entire group of 289 respondents (as well as that of B2) with one important exception: they are quite strongly motivated by TAXC.
Discussion
With respect to the proposed approach to segmentation, the results may be summarized as follows:
i) the subjects are continuously distributed in the preference space, and thus clusters of subjects cannot be isolated (this is a technical way of saying that the group of subjects does not appear upon initial scrutiny to contain actionable sub-groupings).
ii) despite an apparent lack of clusters, there is nonetheless a method which can differentiate subjects according to their attitudes towards objects (i.e. in this illustration differentiated in their responses towards various "motivators").
Recall that it would be highly useful if we could segment donors into particular sociodemographic segments that link to particular motives. We have included in the study the following sociodemographic measures: gender, age, year of graduation, and income. Analysis of groups A, B1, B2, B3, and C based on these measures revealed that group A consisted mainly of middle age men (ratio of men to women was 2:1), who graduated 5 to 15 years ago. Group B1 consisted mainly of older (51+ years old) respondents (evenly split on gender), who graduated more than 15 years ago and were showing income above $80,000. Group B3 consisted mainly of young respondents (20-30 year old), who have graduated very recently (less than 5 years), and show low income (below $20,000). When analyzing the entire group of subjects, one may not detect these differences. They are overshadowed by the average majority which comprises the largest subgroup.
In this paper, we have presented a method whereby a seemingly non-differentiated mass of subjects can be divided into a number of mutually exclusive and exhaustive subgroups. We have also shown how analysis of these same subgroups of subjects can reveal actionable attitudinal differences. Using the suggested procedure, these subgroups which differ from one another can be defined in terms of background variables, such as sociodemographic characteristics. Obviously, the above observations can guide practitioners in fine tuning their advertising and marketing campaigns.
References
Andreoni, J. (1989). Giving with Impure Altruism: Applications to Charity and Ricardian Equivalence. Journal of Political Economy, 97, 1447-1458.
Brittingham, B.E., Pezzullo, T.R. (1990). The Campus Green: Fund Raising in Higher Education. Washington D.C.: School of Education and Human Development, George Washington University.
Greenacre, M. (1984). Theory and Applications of Correspondence Analysis. London: Academic Press, Inc.
Kaciak, E. (1987). A Descriptive Principal Components Method of Testing Feasibility of a Set of Cardinal Rankings. ASAC 1987 Conference Proceedings, Toronto, Canada, 8, 3, 58-67.
Kaciak E., Sheahan, J.N. (1988). Market Segmentation: An Alternative Principal Components Approach, ASAC 1988 Conference Proceedings, Halifax, Canada, 9, 3, 139-148.
Kaciak E., Louviere, J. (1990). Multiple Correspondence Analysis of Multiple Choice Experiment Data. Journal of Marketing Research, November, 455-465.
Kaciak E. (2000). Multiple Correspondence Analysis: Theory and Practice. The Current State of Business Disciplines (ed. S.B. Dahiya), Spellbound Publications PVT. LTD., Rohtak, Volume 2, 465-502.
Lebart, L., Morineau, A., Warwick, K.M. (1984), Multivariate Descriptive Statistical Analysis, John Wiley & Sons.
Leslie, L.L., Ramey, G. (1988). Donor Behavior and Voluntary Support for Higher Education Institutions. Journal of Higher Education, 59, 115-132.
Leslie, L.L. and others (1983). Factors Accounting for Variations Over Time in Voluntary Support for Colleges and Universities. Journal of Education Finance, 9, 213-225.
Lindahl, W.E., Winship, C. (1992). Predicting Models for Annual Fundraising and Major Gift Fundraising. Nonprofit Management and Leadership, 3, 1, 43-64.
Mount, J. and F. Quirion (1988), "A Study of Philanthropy in the Educational Sphere," Philanthropist, 8, 34‑46.
Mount J. (1996). Why Donors Give? Nonprofit Management and Leadership. 7, 1, 3-14.
Prince, R.A., File, K.M., Gillespie, J.E. (1993). Philanthropic Styles. Nonprofit Management and Leadership, 3, 3, 255-267.
Rosenblatt, J.A., Cusson, A.J., McGown, L. (1986). A Model to Explain Charitable Donation – Health Consumer Behaviour. In R. J. Lutz (ed.) Advances in Consumer Research, 12, 235-239.
Kaciak, E., Sheahan, J.N. (1987). An Alternative Method of Principal Components Analysis", Research Report No.87.06.18, Dept. of Stat. and Appl. Prob., Univ. of Alberta.