Eur J Health Econ DOI 10.1007/s10198-014-0625-1

ORIGINAL PAPER

Healthy donor effect and satisfaction with health The role of selection effects related to blood donation behavior Edlira Shehu • Annette Hofmann • Michel Clement Ann-Christin Langmaack



Received: 15 September 2013 / Accepted: 5 August 2014  Springer-Verlag Berlin Heidelberg 2014

Abstract The objective of this paper is to quantify selection effects related to blood donation behavior and their impact on donors’ perceived health status. We rely on data from the 2009 and 2010 survey waves of the German socio-economic panel (N = 12,000), including information on health-related, demographic and psychographic factors as well as monetary donation behavior and volunteer work. We propose a propensity score matching approach to control for the healthy donor effect related to the health requirements for active blood donations. We estimate two separate models and quantify selection biases between (1) active and inactive blood donors and (2) active donors and non-donors. Our results reveal that active donors are more satisfied with their health status; after controlling for selection effects, however, the differences become non-significant, revealing selection biases of up to 82 % compared with non-donors. These

Electronic supplementary material The online version of this article (doi:10.1007/s10198-014-0625-1) contains supplementary material, which is available to authorized users. E. Shehu (&)  M. Clement  A.-C. Langmaack Institute for Marketing and Research Center for Health Marketing, University of Hamburg, Welckerstr. 8, 20354 Hamburg, Germany e-mail: [email protected] M. Clement e-mail: [email protected] A.-C. Langmaack e-mail: [email protected] A. Hofmann Hamburg Institute for Risk and Insurance, HSBA Hamburg School of Business Administration, Alter Wall 38, 20457 Hamburg, Germany e-mail: [email protected]

differences also exist between active and inactive donors, but the differences are less distinct. Our methodological approach reveals and quantifies selection biases attributable to the healthy donor effect. These biases are substantial enough to lead to erroneous statistical artifacts, implying that researchers should rigorously control for selection biases when comparing the health outcomes of different blood donor groups. Keywords Propensity score matching  Blood donation services  Socio-economic panel  Donor management JEL Classification

I12  C83  C21  I14

Introduction Individuals willing to donate blood are required to be healthy, according to the criteria developed by blood donation organizations. Thus, the causal relationship between blood donation and health status should be in line with the fact that only healthy persons are allowed to become donors.1 However, the reverse causal effect has also been widely discussed. For example, donors believe that donating blood improves their true and perceived health status [1, 2]. The positive effect of blood donation on donors’ health has been shown in past studies related to cardiovascular diseases [3, 4] or cancer [5]. In addition, donors argue in web forums (e.g., askwille.com) that they became healthier after donating blood, and some blood donation organizations

1

http://www.redcrossblood.org/donating-blood/eligibility-requirements. Accessed August 2014.

123

E. Shehu et al.

even actively promote this causal effect (e.g., the American Red Cross).2 These two lines of arguments are based on contrary causality. However, the positive effect of blood donation behavior on donors’ health in particular is potentially based on selection effects. The methodological pitfall underlying this selection effect is called the ‘‘healthy donor effect’’, suggesting an overestimation of the positive post-donation health effects [6]. Interestingly, only a few studies have noted the need to control for selection effects related to the healthy donor effect when investigating the health differences between donors and non-donors [6, 7]. Specifically, Atsma and de Vegt [6] recommended avoiding betweengroup (donors and non-donors) comparisons and instead focusing on within-group analyses, thereby underlining the need for more research about how to control for selection effects related to donation behavior. In addition, Atsma et al. [7] showed initial support for the fact that donors and non-donors differ systematically with regard to several relevant characteristics. However, they did not quantify the overall effect of selection bias on health outcomes. Thus, the effect size of the healthy donor effect remains an open and important issue because the positive health-related effects associated with donating blood may be simply driven by selection effects. We address this problem by applying propensity score matching (PSM) to a large sample of donors and nondonors. This methodological approach allows us to uniquely measure the magnitude of the healthy donor effect. We measure individual health as the participants’ self-assessed satisfaction with health, given that past research has shown that self-assessed health information is a valid indicator of real health outcomes [8]. We provide a comparison of the self-assessed satisfaction with individual health status between active and inactive blood donors as well as non-donors by controlling for selection effects. By capturing the selection bias related to the healthy donor effect, we are able to avoid overestimating the perceived health status of the active donors. For our econometric analyses, we use data from the German socio-economic panel (SOEP) which comprises more than 12,000 active and inactive donors and non-donors in Germany.

Previous research Self-assessed health status Previous research on the drivers of individual satisfaction with health is voluminous [8, 9]. ‘‘Good’’ self-assessed 2

http://www.americanredcrossblood.org/faq.html. Accessed on 10 June 2014.

123

health status is a relevant factor for explaining future mortality and morbidity. An early study by Mossey and Shapiro [10] tested the hypothesis that self-assessed health is a predictor of mortality independently of ‘‘objective health status’’. The increased risk of death associated with poor self-rated health was greater than that associated with poor objective health status, poor life satisfaction, low income and being male. These findings provide empirical support for the belief that the way a person views his or her health has an important relationship with subsequent health outcomes. More recently, two related studies offered a review of 27 community studies from different countries and analyzed the effect of self-assessed health status on future mortality [8, 11]. Almost all of the investigated studies showed that individuals with a lower self-assessed health status tend to have a higher mortality risk. An additional meta-analysis confirmed the significant relationship between self-assessed health status and mortality even after controlling for depression or co-morbidities [9]. The findings described above have been confirmed for different countries, including Denmark [12], the United States [13], Canada [14] and the UK [15]. Overall, these findings underline the validity of selfassessed health as an indicator of the objective health of respondents. Selection effects of blood donation on health outcomes Previous studies have investigated the influence of blood donation behavior on the health of donors, finding mostly beneficial effects [3, 16]. Nonetheless, recent articles argue that these findings may be biased because of the systematic differences between donors and non-donors which lead to selection effects [6, 7]. These selection effects may have different causes besides the requirements for donors to fulfill specific health criteria, e.g., self-selection on health leading potential donors to feel able to donate. They may be compared to a membership bias like the healthy worker effect, which is a well-known methodological problem in occupational studies. It refers to the fact that mortality rates are usually lower in the working population because a working individual must be healthy to be able to work. Similarly, donors have to fulfill specific health criteria to become blood donors, so the donor population is generally healthier than the nondonor population. Until now, the healthy donor effect has been recognized as a methodological problem in health research related to blood donation [4, 16]. While single studies have indicated that the healthy donor effect exists [7], a methodological framework that controls for selection effects while comparing health-related outcomes between donors and non-

Healthy donor effect and satisfaction with health

donors is missing. We address this gap and control for selection biases related to the healthy donor effect while comparing the self-assessed health status of active blood donors with that of inactive donors and non-donors.

Method Donors and non-donors are characterized by systematic differences [17, 18]. The decision to become a donor does not occur independently of these characteristics. Because of the systematic differences with respect to their health status [19] and demographic [20] or psychographic characteristics [21], populations of donors and non-donors show systematic selection effects which may be related to different causes. One of the most relevant sources of these selection effects relates to the health criteria that are required from potential donors. Before a person may donate blood, a series of medical checks must be conducted to prove that the person’s health status is sufficient [22]. Hence, group comparisons of satisfaction with health between donors and non-donors are biased if the systematic differences between the groups are ignored [23]. PSM helps to reduce these systematic differences [24]. It separates the selection effect caused by the health requirements for donating blood from the unbiased effect of health outcomes attributable to donation behavior. Considering blood donation behavior as an experimental treatment, a comparison between donors and non-donors would be unbiased only if each person were assigned randomly to one of the groups [23]. As outlined above, becoming a blood donor is not a random process; it is related to systematic health-related differences because only healthy individuals are allowed to donate blood. PSM captures these systematic group differences, separates selection effects, and enables corrected estimates of group differences [23, 25].3 The average effect of blood donation on health is captured by the group-dependent differences in health outcomes, which are unobservable because a person either donates blood or not and can only be part of one group (i.e., if a person is a blood donor, we cannot observe this person’s health status as a non-donor; [26]). PSM approximates the non-observable health outcome for donors that would have occurred if they had not donated blood. This approximation is realized by identifying individual pairs of donors and non-donors that are similar with regard to relevant characteristics [27]. The estimated propensity score — indicating (if normalized) the probability that a person becomes blood donor given his/her characteristics—is used 3

For a mathematical description of PSM see Mithas and Krishnan [29].

to reduce selection bias by matching donors and nondonors based on their characteristics. The propensity score is estimated via a logit (probit) regression in which blood donation behavior (group membership) is used as the dependent variable. In contrast to ordinary least-squares regression, PSM enables a comparison only for those individuals whose propensity scores overlap to a certain extent, indicating common support [28]. After the matching process, ‘‘twin’’ pairs of individuals are identified that (optimally) differ only in their blood donation behavior. Consequently, blood donation behavior (i.e., being a donor or not) can be considered equivalent to a random experimental treatment, and the comparison of health outcomes within the matched sample is corrected for the selection biases caused by the healthy donor effect. Furthermore, by comparing the health status between the matched and non-matched samples, we can quantify the healthy donor effect.

Empirical analysis Data and sample The German SOEP4 is a representative longitudinal survey of private households and individuals that has been conducted annually since 1984. More than 20,000 persons are sampled yearly to participate in the paper-and-pencil survey conducted by the market research institute TNS Infratest Sozialforschung [29]. SOEP has a high degree of stability: 53 % of the households who participated in the first SOEP wave were still participating 25 years later in 2010. The data provide rich information about health, satisfaction, demographic and psychographic factors, and the survey is constantly adapted to current social developments. Consequently, new questions on specific topics are included at regular intervals for 2–10 years. In 2010, questions about blood donation behavior were integrated for the first time (see Appendix 1). This survey wave serves as the baseline for our empirical analyses. Because blood donors are more prone to ‘‘volunteer’’ in general [30], it is important to control for alternative donation forms. The SOEP wave from 2009 includes information on voluntary work. Hence, to integrate this information from 2009 into our model, we considered participants from both the 2009 and 2010 waves in our analyses. A total of 18,024 individuals participated in both survey waves (in 2009 and 2010 there were 18,587 and 4

Detailed documentation for the SOEP data is open to the public via the project’s homepage; (www.diw.de/soep).

123

E. Shehu et al.

19,101 participants, respectively). We excluded individuals who are not permitted to donate blood according to German legislation (3,184 individuals who are younger than 18 or older than 71 and are therefore not allowed to donate due to age limits5) and persons who are not allowed to donate blood for self-reported medical reasons (2,353 individuals). Our final data set consisted of 12,487 (potential) donors. Variables This section presents the measures and descriptive statistics for the variables used in the propensity score estimations that are necessary for the subsequent matching procedure as well as for the outcome variable. Treatment variables We divided the respondents into three groups based on their stated blood donation behavior in the SOEP wave of 2010: (1) active donors (8.11 %), respondents who donated blood in 2009; (2) inactive donors (9.26 %), respondents who donated blood in the 9-year period before 2009 but not in 2009; and (3) non-donors (82.63 %). 6Our objective was to compare the health of the active donors with the health of the other two groups. Hence, we derived two dummy variables comparing (1) active and inactive donors and (2) active donors and non-donors. Both variables were coded as 1 for active donors and 0 otherwise. Both variables served as treatment identifiers for our PSM models and were therefore the dependent variables in the PSM logit estimations. Matching variables used in PSM Our explanatory variables can be divided into four groups that have been shown to characterize blood donors in previous research [31, 32]: health-related, socio-demographic, psychographic, and donation behavior-related variables. Table 1 gives an overview of descriptive statistics and displays the pre-matching mean differences that provided initial evidence of substantial selection effects. For example, in the non-matched sample, fewer active donors were chronically ill compared with inactive and non-donors. After the matching process, however, the significance disappeared, indicating that the differences were caused by the healthy donor effect.

5 See e.g., http://www.blutspende-ost.de/infos-zur-blutspende/spen derinformationen/blutspender-gesucht.php. 6 See http://www.diw.de/en/diw_02.c.222729.en/questionnaires. htmlfor the complete questionnaires.

123

Health-related factors This group comprises characteristics related to health status and health insurance. In general, active donors seem to pay more attention to their health status [7]. Like Boulware et al. [33], we included the variables ‘‘health status in the previous year’’, ‘‘alcoholic beverages’’ and ‘‘smoking’’ as predictors in the PSM logistic models. Furthermore, we controlled for systematic differences in healthy nutrition (‘‘health-conscious diet’’) and overall hours of sleep (‘‘hours of sleep’’). The tests of significant differences in Table 1 reveals that active donors live more healthfully than inactive donors and especially non-donors. Next, we included the body mass index (BMI), defined as the individual’s body mass (in kg) divided by the square of his or her height (in m). We derived two indicator variables that show whether an individual had a BMI that fell into the categories of overweight/underweight and find that there were fewer overweight persons (19.29 %) among the active donors than among the inactive donors (22.70 %; p \ 0.10) and the non-donors (22.24 %; p \ 0.01, see Table 1). An important additional factor is an individual’s health insurance status. Boulware et al. [33] showed that donors are more likely to purchase health insurance than nondonors. Therefore, we included three variables indicating whether a person had a general social health insurance policy (‘‘compulsory health insurance’’), an additional private insurance policy (‘‘additional private insurance’’) or an additional private insurance policy for hospital treatments (‘‘additional private insurance hospital’’). In Germany, about 10 % of the population are privately insured. These individuals are mostly civil servants and self-employed people. It should be noted that individuals with private health insurance coverage in Germany tend to be healthier because of the risk-related premium structure [34]. Interestingly, the proportion of individuals who had an additional private insurance contract for hospital stays was highest in the inactive donor segment (10.17 %) in comparison with the active donors (8.04 %; p \ 0.10) and non-donors (7.99 %; p \ 0.01). Moreover, we took into account whether a person qualified for additional allowances, which are related to employment status (‘‘qualifying for additional allowances’’). Socio-demographic factors The relevance of socio-demographic variables for donor profiling has been demonstrated in numerous studies [35]. Donors are likely to work more, be male [17, 37–39] and have more education [17, 39, 40]. We thus included hours of work per weekday as well as qualification for additional allowances. It is remarkable that active donors worked the most hours (p \ 0.10 versus inactive donors; p \ 0.01

* p \ 0.10

** p \ 0.05

*** p \ 0.01

Donation behavior

Psychographic factors

Socio-demographic factors

Health-related factors

38.94 % 32.07 %

6.428

Return favors

Money donation

2.630

Little control over my life

Voluntary work

2.316

Political interests

3.241 -0.044

Mood (negative)

Not achieved what I deserved

Age category 5 (56–71) 4.507

30.42 %

Age category 4 (46–55)

Risk affinity

24.87 %

Age category 3 (36–45)

7.175

19.97 %

Age category 2 (26–35)

Satisfaction with life

9.05 % 15.69 %

Age category 1 (18–25)

1.53 % 21.42 %

Children

New federal state

31.25 %

Married

Maternity benefits

49.40 % 61.05 %

Gender (male)

11.13 %

Qualify for additional allowance

6.082

8.17 %

Additional private insurance (hospital)

Hours of job (per weekday)

95.57 % 22.06 %

6.05 %

BMI underweight

Additional private insurance (general)

22.01 %

Compulsory health insurance (normal tariff)

28.93 %

BMI overweight

1.652

Alcoholic beverages (spirits)

Chronically

2.488

Health conscious diet 30.32 %

6.945

Hours of sleep

Smoke

3.557

82.63 %

Health status (t-1)

Non-donor

9.26 % 8.11 %

Active donor

Inactive donor

Treatment

0.880

1.455

0.788

0.978

1.773

2.253

1.693

4.368

0.712

0.720

1.085

0.883

12,412

11,000

12,459

12,427

12,439

12,444

12,362

12,456

12,463

12,487

12,487

12,487

12,487

12,487

12,487

12,487

12,487

12,487

12,487

12,439

11,964

12,487

10,472

10,425

12,487

12,487

12,455

12,475

12,356

12,459

12,456

12,467

12,487

12,487

12,487

35.65 %

45.74 %

6.494

2.513

2.360

-0.051

2.989

4.606

7.193

22.90 %

21.03 %

25.77 %

25.57 %

4.74 %

26.46 %

3.65 %

30.70 %

57.65 %

47.68 %

12.20 %

6.481

10.17 %

28.93 %

95.62 %

4.54 %

22.70 %

32.57 %

28.06 %

1.697

2.533

6.96

3.579

Mean

N

Mean

SD

Inactive [1]

Total sample

Explanatory variables

Factor group

Table 1 Descriptive statistics

0.785

1.307

0.732

0.955

1.650

2.117

1.695

4.306

0.672

0.678

1.002

0.881

SD

1,007

1,010

1,011

1,011

1,012

1,009

1,003

1,013

1,012

1,013

1,013

1,013

1,013

1,013

1,013

1,013

1,013

1,013

1,013

1,008

974

10,138

847

845

1,013

1,013

1,010

1,012

1,002

1,009

1,012

1,012

N

29.75 %

37.03 %

6.415

2.654

2.306

-0.038

3.275

4.469

7.160

32.21 %

25.32 %

19.50 %

14.14 %

8.83 %

20.10 %

1.42 %

31.47 %

62.21 %

49.56 %

10.67 %

5.962

7.99 %

20.85 %

95.76 %

6.30 %

22.24 %

29.33 %

31.02 %

1.630

2.480

6.951

3.532

Mean

0.895

1.478

0.796

0.980

1.785

2.271

1.697

4.394

0.714

0.727

1.098

0.889

SD

Non-donors [2]

10,252

8,841

10,292

10,265

10,272

10,284

10,208

10,288

10,297

10318

10,318

10,318

10,318

10,318

10,318

10,318

10,318

10,318

10,318

10,277

9,885

10,318

8,648

8,608

10,318

10,318

10,291

10,308

10,205

10,296

10,291

10,302

N

49.52 %

47.61 %

6.484

2.520

2.368

-0.087

3.156

4.758

7.295

21.11 %

24.13 %

19.12 %

20.85 %

14.79 %

28.63 %

0.69 %

29.67 %

53.63 %

49.39 %

14.30 %

6.813

8.04 %

26.82 %

93.83 %

5.19 %

19.29 %

22.18 %

25.97 %

1.811

2.518

6.880

3.763

Mean

Active [3]

0.818

1.356

0.768

0.982

1.754

2.188

1.651

4.086

0.702

0.690

1.036

0.803

SD

1,153

1,149

1,156

1,151

1,155

1,151

1,151

1,155

1,154

1,156

1,156

1,156

1,156

1,156

1,156

1,156

1,156

1,156

1,156

1,154

1,105

1,156

977

972

1,156

1,156

1,154

1,155

1,149

1,154

1,153

1,153

N

***

**

*

***

*

***

***

*

*

*

*

*

***

***

*

***

cf [1] pre post

***

***

***

***

**

**

***

***

***

***

***

***

**

***

***

***

***

***

**

***

***

***

***

**

***

cf. [2] pre post

Healthy donor effect and satisfaction with health

123

E. Shehu et al.

versus non-donors) and qualified for additional allowances (14.30 %) more often than inactive donors (12.20 %) and non-donors (10.67 %), implying that the percentage of civil servants is higher in the active donor segment. We also included gender, education and marital status in our study. With 49.40 % male participants, the whole sample has a balanced gender structure. Active donors are relatively less likely to be married (53.63 %) than inactive donors (57.65 %; p \ 0.10) and non-donors (62.21 %; p \ 0.01). We also controlled for maternity benefits because donating blood is not allowed for breastfeeding women. Indeed, the percentage of persons who received a maternity benefit is highest in the inactive donor segment (3.65 vs 0.69 % and 1.42 % for active donors and nondonors, respectively). Previous studies also found significant effects of age on blood donation behavior [31, 37]. Due to the relevance of age as a discriminating characteristic between donors and non-donors, we grouped the individuals into five age categories (18–25, 26–35, 36–45, 46–55 and 56–71 years). For all of the categories, we derived dummy variables set at 1 if the age of a respondent fell within one age category and 0 otherwise. The first category served as the reference.

donations cf. to non-donors and p \ 0.01 for voluntary work for both inactive and non-donors; see Table 1).

Psychographic factors

We estimated two PSM models to control for the systematic differences between active versus inactive donors and active donors versus non-donors. The dependent variables are the treatment dummy variables outlined above. The demographic, psychographic and health-related characteristics and donation behavior served as independent variables to estimate the propensity scores using logit regression. The propensity scores were then used in a kernel-matching algorithm with common support (Epanechnikov kernel function and a bandwidth parameter of 0.06; [42]). Table 2 displays the results of the logit models.

We included a measure of overall satisfaction (‘‘general satisfaction’’), which was shown to be a relevant discriminating characteristic between donors and non-donors [19]. Indeed, the active donors are significantly more satisfied (p \ 0.05 cf. to inactive and p \ 0.01 cf. to non-donors). Blood donation is related to psychological barriers such as the fear of needles [40]. Therefore, in line with past studies [31], we included risk aversion in our study (‘‘risk affinity’’) and the general mood of the individuals was calculated as a factor comprising four emotional moods (angry, worried, happy and sad). The factor was computed by means of a principal component analysis (a = 0.69, explained variance = 51 %) with high values indicating an unhappy mood. In addition, we included satisfaction with achievements in life as a measure of self-esteem ([41];‘‘have not received what I deserve’’), a variable that shows the degree of control over one’s life, reciprocity (‘‘return favors’’) and political interest. Donation behavior We added information on donation behavior for other donation forms because it has been proven that blood donors are also more likely to donate time or money [30]. Interestingly, we find significantly more donations of money and time within the groups of active and passive donors than in the non-donor group (p \ 0.01 for money

123

Outcome variable The outcome variable of our analyses is satisfaction with health. In the SOEP survey, the respondents were asked directly about their degree of satisfaction with their own health on an 11-point scale (0 = ‘‘not satisfied at all’’ to 10 = ‘‘very satisfied’’). Active donors are on average significantly more satisfied with their health (mean 7.23) than inactive donors (mean 6.77) and non-donors (mean 6.86). However, while at first glance the statistics indicate that active donors are more satisfied than inactive or nondonors, it is unclear whether these differences may be due to the healthy donor effect. We investigated whether this difference is still significant after controlling for selection effects.

Results Propensity score matching

Active donors versus non-donors Active donors have significantly different health profiles. Generally, they live more healthfully: they smoke less, prefer healthy nutrition and suffer less from chronic illnesses.7 In addition, they are healthier in the previous year. 7

The positive effect of drinking more alcohol on the probability of being an active donor as compared to being a non-donor is somehow inconsistent with the healthy donor effect. While we can only speculate regarding the causes for this effect, we assume that the effect is inverted U-shaped; i.e., that very high levels of drinking alcohol influence the probability of being an active blood donor negatively. A robustness check shows indeed that the squared effect for drinking alcohol is significantly negative (p \ 0.05) in the model comparing active donors and non-donors (and not significant in the model comparing active and inactive donors). The squared effect does not influence the PSM results and the estimates can be provided by the authors upon request.

Healthy donor effect and satisfaction with health Table 2 Results of propensity score estimation (logit model) Factor group

Health-related factors

Socio-demographic factors

Psychographic factors

Donation behavior

Goodness of fit

Explanatory variables

Active versus non-donors

Active versus inactive donors

Coef.

SE

Coef.

Health status (t-1)

0.187

0.054

***

0.144

0.073

**

Hours of sleep

-0.084

0.036

**

-0.142

0.053

***

Health conscious diet

0.130

0.057

**

0.024

0.080

SE

Alcoholic beverages (spirits)

0.237

0.053

***

0.196

0.079

Smoke

-0.199

0.086

**

-0.100

0.120

Chronically

-0.118

0.094

-0.462

0.125

BMI overweight BMI underweight

0.078 -0.385

0.093 0.164

**

-0.052 0.315

0.130 0.256

Compulsory health insurance (normal tariff)

-0.393

0.151

***

-0.427

0.229

Additional private insurance (general)

-0.444

0.148

***

-0.210

0.200

***

-0.003

0.139

Additional private insurance (hospital)

0.428

0.101

Qualify for additional allowance

0.264

0.166

Gender (male)

-0.254

0.085

*** *

Hours of job (per weekday)

0.016

0.010

Married

-0.089

0.097

0.278

0.244

-0.037

0.120

0.008

0.014

0.040

0.134

** ***

*

Children

-0.205

0.094

**

0.147

0.137

Maternity benefits

-0.971

0.473

**

-1.761

0.509

New federal state

0.594

0.083

***

0.111

0.116

Age category 2 (26–35)

-0.221

0.132

*

-1.291

0.221

***

Age category 3 (36–45)

-0.500

0.150

***

-1.428

0.240

***

Age category 4 (46–55)

-0.577

0.149

***

-0.927

0.245

***

Age category 5 (56–71) Satisfaction with life in general

-0.983 -0.020

0.163 0.028

***

-1.084 -0.013

0.257 0.040

***

Risk affinity

0.038

0.018

**

0.006

0.026

Have not achieved what I deserved

-0.001

0.023

0.055

0.034

Mood (negative)

0.007

0.046

-0.032

0.068

Political interests

0.052

0.052

-0.126

0.075

Little control over my life

-0.001

0.029

-0.003

0.043

Return favors

0.136

0.046

0.032

0.065

***

***

*

Money donation

0.387

0.081

***

0.141

0.112

Voluntary work

0.729

0.077

***

0.565

0.108

***

Constant

-3.359

0.557

***

1.429

0.821

*

Observations

7,544

1,696

Pseudo-R2 before

0.079 (v2 = 435.40; p \ 0.000)

0.075 (v2 = 175.13; p \ 0.000)

Pseudo-R2 after

0.001 (v2 = 3.35; p \ 1.000)

0.003 (v2 = 7.13; p \ 1.000)

Mean bias before

12.360

8.710

Mean bias after

1.236

1.970

Propensity score estimation (logit model) within the kernel matching procedure (Epanchenikov kernel function with common support condition) Standard errors based on 1,000 bootstrap replications *** p \ 0.01, ** p \ 0.05, * p \ 0.1

The model also reveals differences in health insurance status: active donors are mostly privately insured individuals (i.e., individuals who are healthier in general). This result aligns with the finding that active donors have less compulsory health insurance (the statutory German health insurance system with community rating and no risk

assessments). The active donors tend to have less additional private insurance in general but more additional coverage for hospital stays. These results indicate that active donors may make use of the family insurance packages that are often included in private policies in Germany.

123

E. Shehu et al. Table 3 Comparison of satisfaction with health in the matched sample Mean treated Active donors (treated) vs non-donors (control) Active donors (treated) vs inactive donors (control)

Mean control

Difference

SE

t value

Before matching

7.237

6.839

0.398

0.071

5.59

After matching

7.237

7.143

0.094

0.070

1.35

Before matching

7.237

6.774

0.463

0.098

4.73

After matching

7.236

7.069

0.167

0.114

1.46

*** ***

Kernel matching (Epanchenikov kernel function, bandwidth 0.06, with common support condition) *** p \ 0.01

With regard to socio-demographic factors, the results reveal that active donors tend to be female, with a lower probability of having children. Consequently, they are rather unlikely to receive maternity benefits. We find that younger individuals are overrepresented within the active donors group [31, 37]. Finally, active donors live in the new federal states of Germany, have higher risk affinity [31] and demonstrate higher reciprocity. Compared to nondonors, blood donors are also prone to giving other types of donations [43, 44]. Active versus inactive donors When compared to inactive donors, active donors seem to have had better health in the recent past. However, the differences with respect to other health-related factors are less prominent, which is probably because there are fewer systematic differences between the groups in general [7]. Comparison of satisfaction with health After controlling for selection biases, we compared satisfaction with health in the matched samples. Table 3 suggests significant pre-matching differences between the groups (differences: 0.39 and 0.46; p \ 0.01), indicating that blood donation leads to greater health satisfaction. However, after applying PSM, the difference in satisfaction with health between active donors and non-donors becomes non-significant (the t value declines from 5.59 before matching to 1.35 in the matched sample). Furthermore, the comparison between active and inactive donors shows a similar picture: the differences in satisfaction with health are non-significant after matching (the t value declines from 4.73 before matching to 1.46 in the matched sample). Both findings indicate that the difference in satisfaction with health without using PSM is substantially overestimated: the difference is reduced by 82 % for the comparison between active donors and non-donors and 26 % when comparing active and inactive donors. These results are plausible because active and inactive donors are very likely to have similar profiles, especially because we excluded respondents who were not allowed to donate blood for medical reasons to avoid related biases.

123

In summary, we find substantial selection biases that provide empirical evidence for the healthy donor effect. Goodness of fit and robustness checks Both models reduce the selection bias substantially (Table 2). In the matched samples, the mean bias for all included variables is substantially lower, and within the advised range of 4 %, with 1.23 and 1.97 for the model comparing active donors to inactive donors and nondonors, respectively [45]. We conducted several robustness tests. First, we tested the robustness of the result using the alternative nearestneighbor matching algorithm [46, 47]. Despite differences in the search methodology—the nearest-neighbor approach seeks only one match for each donor, whereas the kernel algorithm uses all persons as matching partners—both algorithms lead to similar results. Second, we conducted PSM estimations with alternate bandwidth sizes (0.02, 0.04). The results are consistent with our estimation using a bandwidth size equal to 0.06. Furthermore, we tested the assumption of conditional independence, which is essential for the validity of PSM results. This assumption implies that all of the variables that simultaneously influence the propensity of a person to be a donor/non-donor must be observable [24]. We investigated whether our results are robust regarding the unobserved heterogeneity of donors by applying the bounding approach [48]. In a sensitivity analysis, we determined the level of the effect of unobserved components on blood donation behavior that could affect the validity of our results. We find that our results can only become insignificant because of unobserved individual heterogeneity if the effect of the unobserved heterogeneity is approximately 30 % as strong as that of all our observed donors’ characteristics (Gamma values of 1.28 and 1.33, respectively, in model specifications I and II). Given the depth and the richness of the SOEP data, this seems rather unlikely. Last, we replicated our PSM estimation using an alternative outcome variable for the measurement of the health status (the number of nights spent in hospital). The results were consistent with Table 3: before matching, active donors spent on average significantly fewer nights in

Healthy donor effect and satisfaction with health

hospitals compared with both inactive donors and nondonors. This difference, however, becomes non-significant after controlling for selection effects (see Appendix 2).

Discussion

non-significant. These findings indicate that the differences without using PSM were overestimated up to 82 % when comparing donors and non-donors. In summary, we show that selection biases due to the healthy donor effect exist and that these biases lead to an overestimation of the positive influences of blood donation on the donors’ health.

Summary Implications for research Many blood donors seem to believe that blood donation is beneficial for their health [1], and this idea is even actively promoted by the World Health Organization [2]. However, it is unclear whether blood donation leads to better health status, as several studies have shown [3], or whether donors are healthier because of the health requirements for donating blood [7]. Selection effects caused by the healthy donor effect would lead to erroneous results and implications because of the substantial overestimation of the beneficial health effects of blood donation. Past research has underlined the necessity of taking these selection effects into account, recognizing the methodological problem, and a single study even reports the first empirical indications by comparing different donor groups [6]. However, no studies have quantified the healthy donor effect and controlled for selection biases while comparing health outcomes between donors and nondonors. Previous research studies recommend avoiding comparisons between donors and non-donors and focusing on within-group studies [7], indicating the need for alternative methodological approaches that also allow for comparisons between groups. Our study fills this gap and proposes a PSM approach for capturing selection biases related to the healthy donor effect by comparing satisfaction with health between active donors and inactive/nondonors. In particular, we investigated whether active blood donation behavior actually leads to higher satisfaction with health. Using data from the German SOEP comprising more than 12,000 individuals, we included health-related, demographic and psychographic factors and past donation behavior (i.e., monetary donations and volunteering) as matching variables. We estimated two PSM models comparing active and inactive donors as well as active and nondonors. The results reveal significant substantial effects of numerous variables, indicating selection biases related to the healthy donor effect. The differences between active donors and non-donors are greater than those between active and inactive donors. After controlling for these selection effects, we compared the satisfaction with health in both matched samples. Interestingly, while active donors are significantly more satisfied with their health in the unmatched samples, after matching the differences become

Two important implications for academic research can be derived from our results. First, the findings quantify substantial selection biases related to the healthy donor effect and expand the stream of research on the healthy donor effect. This result is interesting because it indicates that the beneficial effects of blood donations that have been reported in health research studies are overestimated. Consequently, future research on health comparisons between blood donor groups should control for selection biases to increase estimation accuracy. Second, we propose a methodological approach for measuring selection biases related to blood donation behavior. Our approach is highly flexible and can be easily adapted to other research questions that require separating selection effects before comparing health outcomes of groups. Our method enables comparison of the effects between donor groups and vastly expands the existing recommendations to focus on withingroup designs. Implications for management We have demonstrated that donors’ better health status is due to selection effects related to blood donation behavior. This reverse causality implies that blood donation organizations and health organizations should be cautious when they proactively communicate the beneficial effects of blood donation on health status. Rather, blood donation organizations should communicate the beneficial health status effects caused by regular health checks that are made before a blood donation, for example. Despite the selection effects, the better health status of donors has implications for health insurance companies: insurance companies might offer reduced premiums for donors because donors are significantly less often sick and tend to behave in a more health-conscious manner. Limitations of our study Our study was limited by the available data: the SOEP survey contains self-stated measures, and we relied upon self-reported satisfaction with health to measure the health status. Hard information e.g., mortality, hospitalization, or

123

E. Shehu et al.

specific diagnoses which could be more objective measures of the health status, were not available in the data. Despite substantial evidence that self-rated health is a valid measure due to its relationship to mortality and morbidity, future studies could try to replicate our findings using observed health outcomes. The PSM techniques allow to capture of selection biases related to the observed variables that are included as predictors into the estimation. Because of this, they do not allow for selection biases due to unobservable factors, e.g., preferences or motives. Future studies could complement our findings by investigating other sources of selection bias and expanding the set of predictor variables that can lead to selection effects of health outcomes between different donor groups.

Outlook for future research There are a number of important ways this research could be extended. First, as stated above, we plan to investigate ‘‘harder’’ measures, such as mortality and morbidity measures, in future. Second, we would like to include motives and investigate related potential biases due to e.g., motivational factors. Finally, it seems interesting to expand the scope to other forms of donation, e.g., marrow or organ donation, and to compare the resulting effects.

Appendix 1 See Table 4.

Table 4 Operationalization of variables Category

Variable

Description

Treatment

1. Active donor

Did you donate blood in 2009?

Operationalization Dummy variable 1 = yes; 0 = no

2. Inactive donor

Have you donated blood in the last 10 years? (donation in the last 10 years but not in 2009)

Dummy variable 1 = yes; 0 = no

3. Non-donor

(No donation in the last 10 years.)

Dummy variable

Outcome

Satisfaction with health

How satisfied are you with your health?

0 = totally unhappy to 10 = totally happy

Health factors

Health status (t-1)

How would you describe your current health? (2009)

1 = bad

1 = yes; 0 = no

2 = poor 3 = satisfactory 4 = good 5 = very good Hours of sleep

How many hours of sleep do you average on a normal day during the working week?

Health conscious diet

To what extent do you follow a health-conscious diet?

0–24 h 1 = not at all 2 = not so much 3 = much 4 = very much

Alcoholic beverages (spirits)

How often do you drink alcoholic beverages?

1 = never 2 = seldom 3 = occasionally 4 = regularly

Smoke

Do you currently smoke, be it cigarettes, a pipe or cigars?

Dummy variable 1 = yes; 0 = no

Chronically

Are you suffering for at least 1 year, or as a chronic condition, from certain complaints or illnesses

Dummy variable

BMI overweight

Overweight = 1 if BMI [ 29 (BMI = weight/height2)

Dummy variable 1 = yes; 0 = no

BMI underweight

Underweight = 1 if BMI \ 20 (BMI = weight/height2)

1 = yes; 0 = no

Dummy variable 1 = yes; 0 = no

Compulsory health insurance (normal tariff)

123

In compulsory health insurance at the normal rate

Dummy variable 1 = yes; 0 = no

Healthy donor effect and satisfaction with health Table 4 continued Category

Sociodemographic factors

Variable

Description

Operationalization

Additional private insurance (general)

Do you have additional private health insurance?

Additional private insurance (hospital)

Which of the following are covered by your additional health insurance? (Hospital stay)

Dummy variable

Hours of job (per weekday)

How many hours per day do you spend on the following activities? Job, apprenticeship, second job (including travel time to and from work)

0–24 h

Qualify for additional allowance

Do you qualify for additional allowances?

Dummy variable 1 = yes; 0 = no 1 = yes; 0 = no

Dummy variable 1 = yes; 0 = no

Gender (male)

Your sex:

1 = male; 0 = female

Married

What is your marital status?

Dummy variable

Children

Are there children in your household?

1 = married; 0 = not married Dummy variable 1 = yes; 0 = no Maternity benefit

Sociodemographic factors

From which of the following income sources do you personally receive money at this time? Maternity benefit during maternity leave

Dummy variable

New federal state

Former East Germany

Dummy variable

Age category 1

Age 18–25 years

Dummy variable

Age category 2

Age 26–35 years

Dummy variable

1 = yes; 0 = no

1 = yes; 0 = no 1 = yes; 0 = no 1 = yes; 0 = no Age category 3

Age 36–45 years

Dummy variable

Age category 4

Age 46–55 years

Dummy variable

Age category 5

Age 56–71 years

Dummy variable

Satisfaction with life in general

How satisfied are you with your life, all things considered?

1 = yes; 0 = no 0 = completely dissatisfied to 10 = completely satisfied

Risk affinity

How do you see yourself: are you generally a person who is fully prepared to take risks or do you try to avoid taking risks?

0 = risk averse to 10 = fully prepared to take risks

Have not achieved what I deserved

Compared to other people, I have not achieved what I deserve.

1 = disagree completely to 7 = agree completely

Mood (negative) (a = 0.69; explained variance = 50 %)

I will now read to you a number of feelings. Please indicate for each feeling how often or rarely did you experience this feeling in the last four weeks. How often have you felt…–angry/worried/happy/sad?

1 = very rarely

1 = yes; 0 = no 1 = yes; 0 = no

Psychographic factors

2 = rarely 3 = Occasionally 4 = often 5= very often

Political interest

Generally speaking, how interested are you in politics?

1 = not at all interested 2 = not so interested 3 = interested 4 = very interested

Little control over my life

I have little control over the things that happen in my life

1 = disagree completely to 7 = agree completely

Return favors

If someone does me a favor, I am prepared to return it

1 = does not apply to me at all to 7 = applies to me perfectly

123

E. Shehu et al. Table 4 continued Category

Variable

Description

Operationalization

Donation behavior

Money donation

And now a question about your donations. We understand donations here as giving money for social, church, cultural, community, and charitable aims, without receiving any direct compensation in return. These donations can be large sums of money but also smaller sums, for example, the change one puts into a collection box. We also count church offerings. Did you donate money last year, in 2009—not counting membership fees?

Dummy variable

Which of the following activities do you take part in during your free time? Volunteer work in clubs or social services

Dummy variable

Voluntary work

1 = yes; 0 = no

1 = yes; 0 = no

Table 5 Comparison of the hospitalized nights as indication of the health status

Active donors (treated) vs non-donors (control) Active donors (treated) vs inactive donors (control)

Mean treated

Mean control

Difference

SE

t value

Before matching

0.449

0.778

-0.328

0.145

-2.26

After matching

0.449

0.575

-0.125

0.098

-1.27

Before matching

0.429

0.952

-0.502

0.116

-3.22

After matching

0.450

0.671

-0.221

0.196

-1.13

*** ***

Kernel matching (Epanchenikov kernel function, bandwidth 0.06, with common support condition) *** p \ 0.01

Appendix 2 The SOEP survey contained self-stated information related to hospital admissions of the respondents. Specifically, respondents were asked to state the number of nights spent in hospitals assessed by the open question ‘‘How many nights altogether did you spend in hospital last year?’’ We used the variable ‘‘number of nights spent in hospitals’’ as an alternative outcome of our PSM models for investigating the robustness of our results (Table 5). The results are consistent with those displayed in Table 3 (see Table 5). We find significant differences between active and inactive donors and between active donors and non-donors regarding the number of nights spent in hospitals before matching. Active donors spent significantly fewer nights in hospitals, indicating a better health status. However, after considering the selection biases related to the healthy donor effect, the differences become non-significant (see Table 5).

References 1. Glynn, S.A., Kleinman, S.H., Schreiber, G.B., Zuck, T., Mc Combs, S., Bethel, J., Garratty, G., Williams, A.E.: Motivations to donate blood: demographic comparisons. Transfusion 42(2), 216–225 (2002)

123

2. World Health Organization (2013). http://www.who.int/media centre/factsheets/fs279/en/. Accessed 22 Aug 2013 3. Ascherio, A., Rimm, E.B., Giovannucci, E., Willett, W.C., Stampfer, M.J.: Blood donations and risk of coronary heart disease in men. Circulation 103, 52–57 (2001) 4. Salonen, J.T., Tuomainen, T.P., Salonen, R., Lakka, T.A., Nyyssonen, K.: Donation of blood is associated with reduced risk of myocardial infarction. The Kuopio Ischaemic Heart Disease Risk Factor Study. Am J Epidemiol 148, 445–451 (1998) 5. Zacharski, L.R., Chow, B.K., Howes, P.S., Shamayeva, G., Baron, J.A., Dalman, R.L., Malenka, D.J., Ozaki, C.K., Lavori, P.W.: Decreased cancer risk after iron reduction in patients with peripheral arterial disease: results from a randomized trial. J Natl Cancer Inst 100(14), 996–1002 (2008) 6. Atsma, F., de Vegt, F.: The healthy donor effect: a matter of selection bias and confounding. Transfusion 51(9), 1883–1885 (2011) 7. Atsma, F., Veldhuizen, I., Verbeek, A., de Kort, W., de Vegt, F.: Healthy donor effect: its magnitude in health research among blood donors. Transfusion 51(8), 1820–1828 (2011) 8. Benyamini, Y., Idler, E.: Community studies reporting association between self-rated health and mortality: additional studies 1995 to 1998. Res Aging 21(3), 392–401 (1999) 9. DeSalvo, K., Bloser, N., Reynolds, K., He, J., Muntner, P.: Mortality prediction with a single general self-rated health question: a meta-analysis. J Gen Intern Med 21(3), 267–275 (2006) 10. Mossey, J.N., Shapiro, E.: Self-rated health: a predictor of mortality among the elderly. Am J Public Health 72(8), 800–808 (1982) 11. Idler, E., Benyamini, Y.: Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav 38(1), 21–37 (1997) 12. Nielsen, S., Brit, A., Siersma, V., Hiort, C., et al.: Self-rated general health among 40-year-old Danes and its association with

Healthy donor effect and satisfaction with health

13.

14.

15.

16.

17. 18.

19.

20.

21.

22.

23.

24.

25. 26.

27. 28.

29.

30.

all-cause mortality at 10-, 20-, and 29 years’ follow-up. Scand J Public Health 36(1), 3–11 (2008) Jylha¨, M., Volpato, S., Guralnik, J.: Self-rated health showed a graded association with frequently used biomarkers in a large population sample. J Clin Epidemiol 59(5), 465–471 (2006) Banks, J., Crossley, T., Goshev, S.: Looking for private information in self-assessed health. In: HEDG Working Paper. University of York (2007) Bond, J., Dickinson, H., Matthews, F., Jagger, C., Brayne, C.: Self-rated health status as a predictor of death, functional and cognitive impairment: a longitudinal cohort study. Eur J Ageing 3(4), 193–206 (2006) Edgren, G., Reilly, M., Hjalgrim, H., Tran, T.N., Rostgaard, K., Adami, J., Titlestad, K., Shanwell, A., Melbye, M., Nyren, O.: Donation frequency, iron loss, and risk of cancer among blood donors. J Natl Cancer Inst 100, 572–579 (2008) Burnett, J.J.: Psychological and demographic characteristics of blood donors. J Consum Res 8(1), 62–66 (1981) Veldhuizen, I.J.T., Doggen, C.J.M., Atsma, F., De Kort, W.L.A.M.: Donor profiles: demographic factors and their influence on the donor career. Vox Sang 97(2), 129–138 (2009) Nguyen, D.D., DeVita, D.A., Hirschler, N.V., Murphy, E.L.: Blood donor satisfaction and intention of future donation. Transfusion 48(4), 742–748 (2008) Condie, S., Maxwell, N.: Comparative demographic profiles: voluntary and paid blood donors. Transfusion 10(2), 84–88 (1970) Ibrahim, N.A., Mobley, M.F.: Recruitment and retention of blood donors: a strategic linkage approach. Health Care Manage Rev 18(3), 67–73 (1993) Ringwald, J., Zimmermann, R., Eckstein, R.: Keys to open the door for blood donors to return. Transfus Med Rev 24(4), 295–304 (2010) Rubin, D., Thomas, N.: Matching using estimated propensity scores: relating theory to practice. Biometrics 52(1), 249–264 (1966) Caliendo, M., Scheel-Copeinig, S.: Some practical guidance for the implementation of propensity score matching. J Econ Surv 22(1), 31–72 (2008) Roy, A.: Some thoughts on the distribution of earnings. Oxf Econ Pap 3(2), 135–145 (1951) Imbens, G.W.: Nonparametric estimation of average treatment effects under exogeneity: a review. Rev Econ Stat 86(1), 4–29 (2004) Mithas, S., Krishnan, M.S.: From association to causation via a potential outcomes approach. Inf Syst Res 20(2), 295–313 (2009) Heckman, J.J., Ichimura, H., Todd, P.: Matching as an econometric evaluation estimator. Rev Econ Stud 65(2), 261–294 (1998) Krupp, H.J.: Das Sozio-oekonomische Panel (SOEP)–Genese und Implementation. In: SOEP-papers on multidisciplinary panel data research, pp. 2–16. DIW, Berlin (2007) Lee, L., Piliavin, J.A., Call, V.R.A.: Giving time, money, and blood: similarities and differences. Soc Psychol Q 62(3), 276–290 (1999)

31. Andaleeb, S.S., Basu, A.K.: Explaining blood donation: the trust factor. J Health Care Mark 15(1), 42–48 (1995) 32. Bekkers, R.: Traditional and health-related philanthropy: the role of resources and personality. Soc Psychol Q 69(4), 349–366 (2006) 33. Boulware, L.E., Ratner, L.E., Ness, P.M., Cooper, L.A., Campbell-Lee, S., LaVeist, T.A., Powe, N.R.: The contribution of sociodemographic, medical, and attitudinal factors to blood donation among the general public. Transfusion 42(6), 669–678 (2002) 34. Hofmann, A., Browne, M.: One-sided commitment in dynamic insurance contracts: the market for private health insurance in Germany. J Risk Uncertain 46(1), 81–112 (2013) 35. Burnett, J.J.: Examining the profiles of the donor and nondonor through a multiple discriminant approach. Transfusion 22(2), 138–142 (1982) 36. Newman, K., Pyne, T.: Service quality and blood donors–a marketing perspective. J Mark Managt 13(6), 579–599 (1997) 37. Nonis, S.A., Ford, C.W., Logan, L., Hudson, G.: College student’s blood donation behavior: relationships to demographics, perceived risk, and incentives. Health Mark. Q 13(4), 33–46 (1996) 38. Tscheulin, D.K., Lindenmeier, J.: The willingness to donate blood: an empirical analysis of socio-demographic and motivation-related determinants. Health Serv Manage Res: an official journal of the Association of University Programs in Health Administration/HSMC, AUPHA 18(3), 165–174 (2005) 39. Duboz, P., Cuneo, B.: How barriers to blood donation differ between lapsed donors and non-donors in France. Transfus Med 20(4), 227–236 (2010) 40. Oswalt, R.M., Hoff, T.E.: The motivations of blood donors and nondonors: a community survey. Transfusion 15(1), 68–72 (1975) 41. Burnett, J.J., Leigh, J.H.: Distinguishing characteristics of blood donor segments defined in terms of donation frequency. J Health Care Mark 6(2), 38–48 (1986) 42. Silverman, B.W.: Density estimation for statistics and data analysis. Chapman & Hall, London (1986) 43. Priller, E., Schupp, J.: Soziale und o¨konomische Merkmale von Geld- und Blutspendern in Deutschland. DIW Wochenbericht 78(29), 3–10 (2011) 44. Shehu, E., Langmaack, A.-C., Felchle, E. and Clement, M.: Profiling donors of blood, money and time: a simultaneous comparison of the German population, Nonprofit Management & Leadership, forthcoming 45. Sianesi, B.: An evaluation of Swedish system of active labour market programmes in the 1990s. Rev Econ Stat 86(1), 133–155 (2004) 46. Black, D., Smith, J.: How robust is the evidence on the effects of college quality? Evidence from matching. J Econom 121, 99–124 (2004) 47. Lechner, M.: Some practical issues in the evaluation of heterogeneous labour market programmes by matching methods. J R Stat Soc A 165, 59–82 (2002) 48. Rosenbaum, P.R.: Observational studies. Springer, New York (2002)

123

Healthy donor effect and satisfaction with health: The role of selection effects related to blood donation behavior.

The objective of this paper is to quantify selection effects related to blood donation behavior and their impact on donors' perceived health status. W...
255KB Sizes 0 Downloads 6 Views