managing blood donations with marketingfaculty.washington.edu/orutz/docs/blood_marksci_2014.pdf ·...

This article was downloaded by: [205.175.118.57] On: 13 January 2015, At: 14:41Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Marketing Science

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

Managing Blood Donations with MarketingAshwin Aravindakshan, Olivier Rubel, Oliver Rutz

To cite this article:Ashwin Aravindakshan, Olivier Rubel, Oliver Rutz (2014) Managing Blood Donations with Marketing. Marketing Science

Published online in Articles in Advance 22 Dec 2014

. http://dx.doi.org/10.1287/mksc.2014.0892

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact [email protected].

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2014, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

http://pubsonline.informs.org

http://dx.doi.org/10.1287/mksc.2014.0892

http://pubsonline.informs.org/page/terms-and-conditions

http://www.informs.org

Articles in Advance, pp. 1–12ISSN 0732-2399 (print) � ISSN 1526-548X (online) http://dx.doi.org/10.1287/mksc.2014.0892

© 2014 INFORMS

Managing Blood Donations with Marketing

Ashwin Aravindakshan, Olivier RubelGraduate School of Management, University of California, Davis, Davis, California 95616

{[email protected], [email protected]}

Oliver RutzFoster School of Business, University of Washington, Seattle, Washington 98195, [email protected]

Blood banks rely on marketing to encourage donors to give blood. Many, if not most, blood banks in theUnited States are community-based not-for-profit organizations with limited marketing budgets. As a result,

blood banks increasingly use novel and inexpensive online media, i.e., paid, owned, and earned (POE) media,in their marketing efforts. We propose a dynamic model to help blood bank marketing managers understandhow blood donations can be managed via online POE media. We analytically characterize the optimal forward-looking paid media strategies, taking into account the asymmetric costs related to shortage and excess of blood,as well as the possibility of a cost-free target donation zone. We detail new advertising resource allocationrules for blood banks and show when traditional allocation recommendations do not apply. Additionally, wediscover that under certain circumstances, owned/earned media activities hurt the blood bank’s performance,despite being (predominantly) free. We validate our analytical model by using daily donation data from acommunity-based blood bank and measure the effects of POE media activities on the level of blood donated.

Keywords : blood bank; paid-own-earned media; social/non-profit marketing; optimal control; time series;Bayesian estimation; dynamic linear model

History : Received: April 12, 2013; accepted: September 8, 2014; Preyas Desai served as the editor-in-chief andHarald van Heerde served as associate editor for this article. Published online in Articles in Advance.

1. IntroductionDemand for blood grows faster than blood dona-tions. Because of aging populations, the demand forblood in developed countries rises by 6% to 8% annu-ally, while donations increase by only 2% to 3%.1

Blood collection, a not-for-profit activity conducted byblood banks, involves matching blood donations withblood demand. This task is challenging. First, bloodbanks do not control how hospitals use and orderblood. Second, blood cannot be bought from donors(Hillyer et al. 2007) and no formal “blood market”exists where supply and demand determine a clearingprice (Slonim et al. 2014). Finally, blood is perishable.In the United States 4.3% of collected units are dis-carded because they were not used before their expi-ration dates (Whitaker 2013). Marketing allows bloodbanks to address these challenges and help managedonations to avoid blood shortages, while minimizingcosts due to outdated units.

Focusing on the demand for blood, the operationsresearch literature has shown that a “first in, firstout” inventory policy minimizes the number of out-dated units and blood shortages (e.g., Nahmias 1982,Prastacos 1984, Pierskalla 2005). Meanwhile, research

1 http://www.schlitzpark.com/events/1562-american-red-cross-blood-drive. Last accessed on April 10, 2014.

in information systems has focused on how informa-tion technology can help optimize blood use (Belienand Forcé 2012). To our knowledge, however, thesetwo streams ignore the challenge of managing thesupply of blood, i.e., attracting donors. Behavioralstudies focus on the psychological barriers to blooddonation, but abstract away from the challenge of col-lecting the right amount of blood (e.g., Pessemier et al.1977, Burnett 1981, Nguyen et al. 2008). Thus, theextant literature on blood does not, to our knowledge,address the role of marketing in managing blooddonations.

We aim to address this gap by investigating howblood banks can use marketing to manage blooddonations. While the marketing literature has a longhistory of investigating media allocation rules, it isnot clear that findings from sales or profit maximiz-ing companies apply to the management of blooddonations. Additionally, marketing managers at non-profit blood banks often face advertising budgets thatrestrict the type of marketing instruments at theirdisposal. Specifically, we investigate how inexpen-sive online paid, owned, and earned (POE) media(e.g., Corcoran 2009, Goodall 2009, Stephen and Galak2012) can be leveraged to attract donors in lieu of tra-ditional expensive mass media.

1

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Aravindakshan, Rubel, and Rutz: Managing Blood Donations with Marketing2 Marketing Science, Articles in Advance, pp. 1–12, © 2014 INFORMS

To answer these questions and provide manage-rial guidelines on the management of POE mediaby blood banks, we frame the management of blooddonations as a dynamic advertising resource alloca-tion problem. We propose a dynamic model wherethe blood bank’s POE media strategy drives the vol-ume of blood collected (i.e., donated). Contrary tofor-profit settings, our approach allows for an objec-tive function that captures asymmetric costs betweenshortage and excess of blood in addition to a cost-freeregion.

We contribute to the marketing literature on sev-eral different dimensions. First, we establish that theinverse allocation rule found in the advertising lit-erature (e.g., Erickson 2003) is optimal for bloodbanks only in shortage and excess situations, but notwhen donations are within the cost-free region. Inthat case, the optimal rule is to increase spendingwith donations. Second, we find that an increase ineffectiveness of advertising instruments could causethe blood bank to spend less on these instrumentsand not more as prescribed by the extant integratedmarketing communications (IMC) literature in for-profit settings (e.g., Naik and Raman 2003). Third, weshow that POE media should be treated differentiallyin the blood donation management problem. Morespecifically, we find that despite being predominantlyfree, owned/earned media can hurt the blood bankbecause it cannot adjust how active these platformsare, i.e., these platforms cannot (and should not) beshut down quickly when donation levels are too high.Paid media, on the other hand, can be turned on/offeasily to dampen donations, thus alleviating the riskof spoilage. Fourth, using data from a communityblood bank we empirically measure the impact ofPOE media on blood donations.

We proceed as follows. First, we provide an over-view on the U.S. blood supply and POE market-ing. We then present our analytical model and derivenovel allocation rules for the management of POEmedia for blood banks. Next, we empirically validatethe proposed model in a Bayesian framework using anovel data set from a community blood bank. Finally,we conclude with a discussion of our key insights.

2. Research Background2.1. The U.S. Blood SupplyHundreds of blood banks exist in the United States.At one extreme, large establishments such as the NewYork Blood Center or the Puget Sound Blood Centercollect thousands of daily donations and serve hun-dreds of hospitals. On the other end of the spectrum,small community-based centers such as LifeStream inCalifornia or Suncoast Communities Blood Bank inFlorida collect only a few hundred pints a day and

serve a handful of local hospitals. Collectively, bloodbanks deliver on average 15 million units of blood(supplied by donors) to over 4,000 hospitals across theUnited States (Whitaker et al. 2011).

Before the 1970s, U.S. blood banks paid donorsfor blood. The new U.S. blood policy that aroseunder Nixon’s presidency required U.S. blood banksto label blood units collected from paid donors (Lyons1971, Drake et al. 1982, Hillyer et al. 2007). Bloodbanks then ceased recruiting donors via monetaryincentives. Consequently, marketing communicationsbecame critical in the management of blood dona-tions. The advent of online media enabled bloodbanks to reach donors without having to invest incostly mass media advertising.

2.2. POE MarketingBefore the Internet, marketers were restricted to eitherexpensive advertising media such as TV or print, ormethods that do not scale well such as interpersonalcommunications (e.g., personal selling). The Internetchanged this landscape by allowing inexpensive one-to-many (advertising) communication, as well as scal-able one-to-one interactions. These new media out-lets are referred to as POE media (see Corcoran 2009,Goodall 2009, Stephen and Galak 2012).

Paid media usually refers to paid online media. Thebiggest and most effective of these is paid searchadvertising, which organizations use to address indi-viduals directly during their electronic search. Its pay-per-click model differs from the pay-per-exposuremodel of traditional advertising. A blood bank bene-fits from paid search advertising due to the relativelyinexpensive nature of this instrument and its abilityto intercept a potential donor during her search forblood donation options. Owned media refers to con-tent that belongs to the organization, for example,the own website, a blog, or the presence on socialmedia sites such as Facebook or YouTube. Last, Earnedmedia refers to mentions of the brand not generatedby the organization, e.g., comments on Facebook bydonors. Paid and, in part, owned media are most sim-ilar to the traditional marketing tools such as adver-tising, branding, or corporate identity (e.g., store orcatalogue design). However, earned media are trulynovel in that they allow consumers to broadcast theiropinions effortlessly to a much larger audience thanwas possible at any time before the Internet. Thus,marketing managers must understand how to lever-age earned media to their advantage to circumventthe typically limited marketing budgets of non-profitorganizations.

Substantially, this study contributes to the nascentliterature on POE media. In particular, we build onStephen and Galak’s (2012) work that shows a signif-icant impact of online earned media on the lending

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Aravindakshan, Rubel, and Rutz: Managing Blood Donations with MarketingMarketing Science, Articles in Advance, pp. 1–12, © 2014 INFORMS 3

volume in a microlending marketplace. We general-ize this research by simultaneously considering POEmedia and by deriving POE media allocation rulesfrom our analytical modeling framework. The funda-mental nature of earned media complicates its rolein the modern marketing mix as it is generated byconsumers and not by the organization. As such, thelevel of control exerted by the organization is lowercompared to traditional marketing mix instruments,which inevitably impact the actions taken by the man-ager. The lower level of control on earned and, tosome extent, owned media could have an adverseeffect on the blood donation management problem,especially in the case of sustained high levels of dona-tions that increase the risk of spoilage. In §3, we studyhow blood banks should use relatively inexpensivePOE media to drive blood donations.

3. Managing Blood Donations withPOE Media

We conceptualize POE media as follows. Paid media(uP 5 are “pay for performance” media that can beadjusted easily and quickly, making them flexible,albeit potentially costly. Ease of adjustment makespaid media prime marketing instruments for bloodbanks in terms of balancing blood donations and localblood demand. Owned (uO) and earned (uE5 media,however, are characterized by bigger, often infre-quent, investments. The main advantage of these twotypes of media is that the marginal costs of generatingadditional exposures or clicks are close to zero. Forexample, costs to set up/rebuild a website or estab-lish a presence on Facebook could be high, but thecost for an additional view, comment, or like is neg-ligible. This advantage, however, comes at a price,which is their inherent inflexibility, i.e., they cannotbe increased or reduced at will. Thus, whereas paidmedia can be easily used to manage daily fluctua-tions in blood donations, owned/earned media can-not. Next we link the level of donations to thesemarketing instruments.

Specifically, let D4t5 denote the level of blood dona-tions at time t. We assume that each media type con-tains several instruments, i.e., for paid 4P5, owned4O5, and earned 4E5 media we have the follow-ing set of options, i.e., uP = 8uP11 0 0 0 1uPi1 0 0 0 1uPP 9,uO = 8uO11 0 0 0 1uOj1 0 0 0 1uOO1 9, and uE = 8uE11 0 0 0 1uEk10 0 0 1uEE9. The change in the level of blood donationsat time t is then given by

dD =

( P∑

i=1

�Pi

√uPi4t5+

O∑

j=1

�Oj

√uOj4t5

+

E∑

k=1

�Ek

√uEk4t5− �D4t5

)

dt + dW1 (1)

with D405 = D0 and where �Pi is the effectiveness ofpaid media i, �Oj the effectiveness of owned media j ,and �Ek the effectiveness of earned media k. Theterm � represents the decay rate of past donationsand 1−� the carryover effect; finally, dW is the incre-ment of a Brownian motion. Equation (1) extendsthe Nerlove and Arrow dynamic advertising model(Nerlove and Arrow 1962) to incorporate owned andearned media activities along with paid media. Thismodel is commonly used in advertising research (seeTable 1 in Aravindakshan and Naik 2011) and waspreviously extended to multiple media (e.g., Basset al. 2007).

A blood bank’s objective is to maintain the levelof collected blood within a range that prevents bothshortage and excess. While the cost implications of ashortage are straightforward (e.g., decreased qualityof care, increased hospitalization days due to post-poned surgeries, etc.), the cost implications of anexcess are also substantial. First, collecting and stor-ing blood is expensive and these costs cannot berecouped if the blood is ultimately discarded. Sec-ond, turning donors away when inventory is highcould result in these donors never returning. Beingturned away could serve as a permanent psycho-logical excuse for not giving blood anymore (e.g.,Prastacos 1984, Pillavin 1987, Halperin et al. 1998).Third, blood banks also provide information on howmuch blood is discarded, which can impede donors’willingness to give blood (and/or donate money).Finally, even though both shortage and excess ofblood are costly to the blood bank, a shortage is moreundesirable than excess (e.g., Brodheim and Prastacos1979, Cohen and Pierskalla 1979).

We capture these asymmetric costs using a novelloss function. Let C4D4t55 be the loss function thatpenalizes the shortage or excess of blood

C4D4t55=

qS4D4t5−b52 if D4t5 b3

(2)

where qS and qE capture the different costs associatedwith blood shortage and blood excess, respectively,while the interval 6b3 b7 delineates the cost-free zonedetermined by the demand for blood and its perisha-bility (see Figure 1 for a graphical representation). Thelower bound, i.e., b, is determined by a demand that istoo high such that it exposes the blood bank to therisk of shortage. On the other hand, the upper bound,i.e., b, is determined by a demand that is too lowcompared to the blood bank’s inventory, which meansthat the blood bank will have to discard blood due toits perishability if it exceeds this threshold.

As the forward-looking manager allocates resourcesto paid media to avoid shortage or excess of

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Figure 1 (Color online) Blood Bank’s Cost Function

Cost

Donations

Shortage region Cost free zone Excess region

b– b–

blood, the blood bank’s objective combines (2) withthe amount invested in paid media, i.e.,

∑Pi=1 uPi.

The blood bank thus minimizes the objective func-tion in (3)

J �D�= minup

E

[

∫ �

0e−rt

{

C�D�t��+P∑

i=1

uPi

}

dt

]

� (3)

subject to (1) to obtain its optimal POE media strat-egy, where uP = �uP1� � � � �uPP �. The term r gives thediscount factor and J � · � is the blood bank’s valuefunction. We characterize the optimal strategy in eachregion and derive the following proposition:

Proposition 1. The blood bank’s optimal paid mediastrategy is

uPi�D�∗ =(

max{

0 uPi�D�/2})2

� (4)

with

uPi�D�=

�Pi�A1D+A2� if D b�

(5)

where A1 < 0 and C1 < 0, while B1 > 0.

Proof. See Online Appendix I (available as supple-mental material at http://dx.doi.org/10.1287/mksc.2014.0892). �

Dynamic advertising models prescribe inverse allo-cation rules, i.e., to decrease spending as the tar-get variable increases (e.g., Erickson 2003). We findthat in our setting this recommendation holds onlyfor the shortage and excess regions, but not for thecost-free zone. Indeed, Equation (5) reveals that whenD ∈ �b b�, �upi�D�∗/�D > 0, implying an increase inspending on paid media as the level of donationsincreases, i.e., a proportional spending rule. This out-come is driven by two mechanisms. First, the dynamicshadow price (measuring the sensitivity of the valuefunction to marginal variations in donation levels)

increases as the level of donations (the metric inform-ing the media allocation decision) increases. Second,by implementing this strategy, the level of dona-tions rapidly moves away from the shortage region,because, at the equilibrium, the growth of donationsrapidly increases as dD∗/dt > 0 and d2D∗/dt2 > 0.

To clarify how the existence of a cost-free zoneand asymmetric costs between shortage and excessof blood change the advertising policies, we reportthe optimal paid media strategy for three scenarios inFigure 2. In Panel A, we consider a symmetric lossfunction and no cost-free zone (i.e., qS = qE and b =

b = b�. In this case, the cost of shortage and excessare identical. The optimal strategy is linearly decreas-ing in the level of donations, i.e., spending on paidmedia decreases as the level of donations increases.In Panel B, we consider an asymmetric loss functionand no cost-free zone (i.e., qS > qE and b = b = b�.Note that in this case, the optimal policy is again tospend less on paid media as the level of donationsincreases. However, two differences emerge. First, theoptimal policy becomes piecewise continuous witha jump at b. Second, the slope of the optimal strat-egy is steeper for D b. Intuitively,this is driven by the fact that the cost of shortage ishigher than the cost of excess (i.e., qS > qE�. Finally, inPanel C, we report the optimal policy with an asym-metric loss function incorporating a cost-free zone(i.e., b qE�. We find thatwithin the cost-free zone, a proportional allocationrule is optimal.

Proposition 2. Own (Paid) Media Effect, i.e.,�uPi�D�/��Pi

• In the shortage region,� if �A2/��Pi > 0, then �uPi�D�/��Pi > 0.� If �A2/��Pi < 0, then �uPi�D�/��Pi > 0 when D<

DS and negative otherwise, where DS is a function ofparameters.

• In the excess region,� if �C2/��Pi > 0, then �uPi�D�/��Pi > 0.� If �C2/��Pi < 0, then �uPi�D�/��Pi > 0 when D<

DE and negative otherwise, where DE is a function ofparameters.

• Finally, in the cost-free region,� if �2

Pi <∑

h�=i �2Ph, then �uPi�D�/��Pi > 0 if D >

�/� and negative otherwise.� If �2

Pi >∑

h�=i �2Ph, then �uPi�D�/��Pi > 0 if D <

�∑O

j=1 �Oj

√uOj+

∑Ek=1 �Ek

√uEk�/� and negative otherwise.

Proof. See Online Appendix I.

Proposition 2 shows that in every region, anincrease in the effectiveness of paid media i maylead to decreased spending in this media. Thisinsight contributes to the marketing literature, whichin traditional for-profit settings, advises managersto always spend more when the effectiveness of a

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Figure 2 (Color online) Optimal Paid Media Strategy UnderDifferent Scenarios

Donations

Donations

Donations

Paid media Panel A

Panel B

Panel C

Paid media

Paid media

b = b = bqS > qE

b = b = bqS = qE

Shortage region Cost free zone Excess region

b bb

marketing instrument increases. In all three regions,�uPi�D�/��Pi < 0 holds when two conditions are met.In the shortage and excess regions, the first condi-tion pertains to the effect of increased effectivenesson the intercept of the advertising policy, (�A2/��Pi

and �C2/��Pi, respectively), whereas in the cost-freeregion the first condition pertains to the relativeeffectiveness of media i with respect to other media(�2

Pi >∑

h�=i �2Ph�. In all regions, the second condition is

defined by different thresholds for donations, i.e., DS ,DE , and �

∑Oj=1 �Oj

√uOj +

∑Ek=1 �Ek

√uEk�/�, in the short-

age, excess, and cost-free regions, respectively. Thus,�uPi�D�/��Pi < 0 is driven primarily by the bloodbank’s objective, which is to ensure a target level (orrange) of donations and not to maximize the volumeof blood collected. Next, we investigate cross (paid)media effects.

Proposition 3. Cross (Paid) Media Effect, i.e.,�uPi�D�/��Ph�=i

• In the shortage region,� if �A2/��Ph�=i > 0, then �uPi�D�/��Ph�=i > 0.� If �A2/��Ph�=i < 0, then �uPi�D�/��Ph�=i > 0 when

D<−��A2/��Ph�=i�/��A1/��Ph�=i� and negative otherwise.• In the excess region,� if �C2/��Ph�=i > 0, then �uPi�D�∗/��Ph�=i > 0.� If �C2/��Ph�=i < 0, then �uPi�D�/��Ph�=i > 0 when

D<−��C2/��Ph�=i�/��C1/��Ph�=i� and negative otherwise.• Finally, in the shortage region, �uPi�D�/��Ph �=i > 0

if D < �∑O

j=1 �Oj√uOj +

∑Ek=1 �Ek

√uEk�/� and negative

otherwise.


It follows from Proposition 3 that even in the absenceof direct interactions between media, variations in theeffectiveness of paid media h inform the optimalinvestment in paid media i. This finding departsfrom the extant literature in marketing resource allo-cation, which would have recommended not chang-ing the allocation to media i when �Ph�=i varies (i.e.,�uPi�D�/��Ph�=i = 0). This interdependence in mediaspending, despite the lack of direct interaction, isdriven by the indirect effect of media h on media ithat originates from the shadow price of donationsand not through instantaneous profits as in Naik andRaman (2003). Last, we investigate the effect of ownedand earned media on the value function of the bloodbank or, in other words, their impact on the bloodbank’s long-term performance.

Proposition 4.• In the shortage region, an increase in owned/earned

media increases the blood bank’s value function whenD < −��A3/�v�/��A2/�v� if �A3/�v > 0, where v =

�uOj�uEk�. Otherwise, the value function decreases.• In the excess region, an increase in owned/earned

media increases the blood bank’s value function when D<−��C3/�v�/��C2/�v� if �C3/�v > 0, where v= �uOj�uEk�.Otherwise, the value function decreases.

• In the cost-free region, the sensitivity of the valuefunction with respect to media v = �uOj�uEk� equals�J /�v=−2�r+2��D�−v�/��

∑

i �2Pi��+g�D�v�, which

could be positive or negative.2

2 The function g�D�v� is a nonlinear function of D, owned andearned media, as well as the model’s parameters. We do notreport g�D�v� due to its long mathematical expression. We con-ducted numerical simulations to verify that both �J /�v < 0 and�J /�v > 0 can happen.

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.



Proposition 4 provides conditions under whichowned and earned media negatively affect the valuefunction (i.e., long-term performance) of the bloodbank. Similar to Propositions 2 and 3, the impactof owned and earned media on the value functiondepends not only on the region that donations arein (e.g., excess or shortage) but also on thresholdsspecific to each region. Therefore, Proposition 4 sug-gests that owned and earned media can be costly,despite the fact that the exposures generated arepredominantly free, because the value function candecrease when owned and earned media increase.To understand the intuition behind this result, notethat the blood bank can instantly adjust the levelof paid media. Conversely, changes in owned andearned media are generally less frequent and lessreactive to daily donation levels. For example, theblood bank would be ill-advised to take its website(owned) offline when the level of donations is toohigh. Apart from an immediate (and desired) impacton donations, there might be undesired consequences.For example, donors might assume that the bloodbank is closed as the website is not accessible. As aresult donors may not consider the blood bank forfuture donations. However, if the blood bank stopspaid media, donors are unlikely to conclude that theblood bank is out of business.

To summarize, our results indicate that success-ful POE media strategies for a blood bank gobeyond investing in paid media and capitalizing onowned and earned media. Our proposed allocationrules depart from the insights derived for profit-maximizing firms. In particular, we find that alloca-tion rules in for-profit settings do not directly applyto blood banks. We also find that owned and earnedmedia can have negative effects on the value func-tion, i.e., the long-term performance, of the bloodbank, despite the fact that they are predominantlyfree. Next, we provide an empirical illustration of ourapproach.

4. Data DescriptionUsing donation and POE media data from a commu-nity blood bank we test the implications of the theo-retical model and illustrate its normative results. Theannual operating budget of the blood bank is $48 mil-lion and it collects, on average, 50,000 units of bloodper year to supply nine local hospitals.3 Our dataspans 318 days. During that time, the blood bank col-lected, on average, 157 pints of blood per day, witha minimum of 44 pints and a maximum of 318 pints.

3 The blood bank is part of a larger entity and thus is not requiredto file a Form 990 detailing its marketing budget.

The proportion of pints discarded due to passed expi-ration dates is comparable to the national averagewith the objective to reach zero outdated units. As acommunity-oriented non-profit, the blood bank doesnot have access to a large marketing budget and can-not invest in mass media, e.g., TV.4

Because of budget constraints typical for a smallnon-profit, the blood bank relies on novel and inex-pensive POE media. Specifically, marketing resourcesare allocated to paid search advertising and socialmedia platforms such as Facebook and YouTube. Wefollow the typology of Stephen and Galak (2012)to classify the blood bank’s online media into POEmedia. We measure the blood bank’s paid mediaactivities by its paid search advertising metrics. Onaverage, the blood bank spent $2.38/day on its BloodDonation campaign; the average number of clicks wasfive. This level of spending comports with other stud-ies on paid search advertising (e.g., Table 1 in Chanet al. 2011). The maximum amount spent on a sin-gle day was $12.77 and the minimum was zero dol-lars (for six days). The Blood Donation campaigntargets potential blood donors searching for bloodbanks on Google.5 We measure owned media activ-ity using multiple metrics, i.e., the number of dailyviews on YouTube, the number of unique daily userson YouTube, and the number of daily page views onFacebook from users logged into Facebook.6 Last, wemeasure earned media activity by the number of likesand unlikes on the blood bank’s Facebook page andthe number of daily comments on the blood bank’sFacebook page. Table 1 summarizes the descriptivestatistics for the variables in the data.

5. Empirical Analysis5.1. Empirical ModelTo estimate the parameters in (1) and illustrate itsnormative implications we discretize the donation

4 We collected data from the Form 990s of the top blood banks inthe United States as provided by the Urban Institute based on totalgross receipts. We found that larger blood banks have marketingbudgets that vary from $56,000 for the New York Blood Center atthe low end to $6,358,667 for the Gulf Coast Regional Blood Centerat the high end; the median is $501,206 and the mean is $1,056,236.These figures show the wide variations in marketing budgets acrossblood banks.5 Note that the blood bank also ran a Blood Products and Ser-vices campaign targeting hospitals and pharmaceutical companiessearching for blood products. As expected, we found that the BloodProducts and Services campaign does not have a statistically signif-icant effect on blood donations. Thus, we did not include the BloodProducts and Services campaign data.6 Note that our data on owned media stem from content hosts (suchas YouTube and Facebook) alone. The blood bank did not keepaccurate logs of their own website visits. Because of this constraintwe cannot include data on the blood bank’s website visits.

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 1 Descriptive Statistics

Avg. Std. Min Max

Number of daily donations 157045 55084 44 318Paid media ($)

Blood donation campaign 2038 1068 0 12.77Owned media

YouTube views 11095 16060 0 154YouTube unique users 6075 8009 0 71Facebook page views 12066 9059 0 82

Earned mediaFacebook new likes 2084 3011 0 31Facebook new unlikes 0036 0064 0 3Facebook new comments 3019 5003 0 35

dynamics to the daily level. Our empirical model isgiven by

Dt = �Dt−1 +

Paid media︷︸︸︷

�1√uP1t +

Owned media︷︸︸︷

�2√uO1t +�3

√uO2t +�4

√uO3t

+

Earned media︷︸︸︷

�5√uE1t +�6

√uE2t +�7

√uE3t +�D

t 0 (6)

The level of donations in the current period, i.e.,a day, is driven by the blood bank’s POE mediaactivities. Each donation corresponds to one pint ofblood drawn from a donor, i.e., 450 milliliters of fluidper donation. Paid media metrics include the dollaramounts allocated to the Google Blood Donation cam-paign (uP1t5. Owned media metrics include YouTubeviews (uO1t5, YouTube unique users (uO2t5, and Face-book page views (uO3t5. Finally, earned media met-rics include user activity on the blood banks ownedmedia such as Facebook new likes (uE1t5, Facebooknew unlikes (uE2t5, and Facebook new comments(uE3t5. The effectiveness of the POE media is mea-sured by the parameters �11 0 0 0 1�7. We test whetherPOE media exhibit diminishing marginal effects onthe level of donations by taking the square rootand the logarithmic transformations of the actualobserved value in our estimation procedure. We keptthe square-root transformation as it provided the bestfit.7 From our discussions with the management of theblood bank we learned that the blood bank is a val-ued part of the community and has a group of loyaldonors. To account for this, we include a carry-overeffect in our model specification, i.e., a certain propor-tion (�) of the previous period’s donation level will

7 Note that one could have a potentially different conceptual-ization of diminishing marginal effects. Our approach assumesthat diminishing marginal effects are media specific, i.e., addi-tional paid media exposures do not affect the effectiveness of, e.g.,earned media. We have tested a specification in which diminishingmarginal effects are defined on the sum of the media exposure. Forour data, our proposed specification given in (6) fits the data bet-ter (results are available from the authors on request). We want tothank an anonymous reviewer for this suggestion.

occur in the current period without any marketingeffort in the current period. This carryover (�5 cap-tures the dynamic nature of the blood donation pro-cess. Finally, �D

t is a normally distributed error term.

5.2. Estimation MethodWe use a Dynamic Linear Model (DLM) (West andHarrison 1997) implemented in a Bayesian frameworkto estimate the parameters of Equation (6). In mar-keting, DLMs have been used to address scenariosin which a key component of the data is unobserved(e.g., Bass et al. 2007), as is true in our model. Anappealing feature of the DLM is that it simultane-ously captures the dynamic evolution of the level ofdonations and the effects of POE media on donationdynamics.

Note also that it is potentially important to accountfor seasonality in donation activities. We include 0/1indicator variables to account for day-of-the-weekeffects in blood donations (Mondayt1 0 0 0 1Fridayt). Notethat the blood bank is closed on Sundays and holi-days. The parameters �11 0 0 0 1�5 measure the impactof the day-of-the-week on the level of donations. Toestimate our model given in Equation (6), we definethe required transition and observation equations.The transition equation is given by

Dt = 41 − �5Dt−1 +Ct +�Dt 1 (7)

where Ct = �1√uP1t +�2

√uO1t +�3

√uO2t +�4

√uO3t +

�5√uE1t + �6

√uE2t + �7

√uE3t . Next we link Equa-

tion (7) to the observed volume of blood collected viathe observation equation

Yt = ZD+�1Mondayt +�2Tuesdayt +�3Wednesdayt

+�4Thursdayt +�5Fridayt + �t0 (8)

We relate the estimated level of donations Dt fromEquation (7) to the observed level of donations Yt inEquation (8) through Z (where Z = 1). We assume �t ∼N401�2

� 5.Inference in the model given by Equations (7) and

(8) could be biased. Sources for this bias could besimultaneity in the blood bank’s decision making oran omitted variable bias related to owned and earnedmedia, e.g., these metrics are high or low due tosome unobserved external event. Finally, there couldbe a bias due to measurement error in owned andearned media as these are reported by third partyfirms and not directly measured by the blood bank.To alleviate concerns about these issues, we use aninstrumental variables (IV) approach estimated in aseemingly unrelated regression (SUR) set-up. Com-pared to two-stages estimation procedures typicallyused in a frequentist setting, our Bayesian frameworkallows for simultaneous estimation of the model andthe IV equations (e.g., Rossi et al. 2005).

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


In our case, there is concern about all POE media,thus we use seven IV equations to address thesepotential endogeneity concerns. We describe ourapproach to paid media before we detail our instru-mentation approach for owned and earned media. Weuse two sources of variation to instrument for paidmedia. First, we use the blood bank’s shipment dataas a source of external variation. The shipment datareflect a hospitals’ blood demand that is not underthe control of the blood bank. This is similar to IVapproaches to price endogeneity using input (rawmaterial) prices as instruments for the endogenousprice. For example, Kuksov and Villas-Boas (2008)use tomato prices to instrument for price in a modelof brand choice in the ketchup category. We alsouse lagged paid media to address concerns with thewidely-used management practice of setting the cur-rent level of advertising based on the past level ofadvertising. Last, we use lagged cross media, i.e.,lagged values of owned and earned media. As such,our IV equation for paid media is given by

uP1t = �P1uP1t−1 +�P

2uP1t−2 +�P3 St−1 +�P

4 St−2

+

J∑

j=1

�P4+juOjt−1 +

K∑

k=1

�P4+J+kuEkt−1

+�Pt 1 (9)

where S is shipment of pints of blood, �P1 1 0 0 0 1�

P4+J+K

are parameters to be estimated, and �Pt is a normally

distributed error term.We instrument for owned and earned media using

the following set-up for all owned and earned media.The data allows us to instrument with own laggedvalues as well as cross values for owned and earnedmedia. For example, to instrument for uO1t we uselagged values of uO1t in addition to the lagged valueof uO2t and uO3t as well as the lagged values of allearned media. For exemplary purposes we give theIV equation for the first owned media

uO1t = �O11 uO1t−1 +�O1

2 uO1t−2 +

J∑

j=2

�O11+juOjt−1

+

K∑

k=1

�O11+J+kuEkt−1

+�O1t 1 (10)

where �O11 1 0 0 0 1�O1

1+J+K are parameters to be estimatedand �O1

t a normally distributed error term. The IVequations for the other owned and earned media fol-low the set-up of Equation (10). In addition, we testfor exogeneity of the instrumented advertising met-rics using the approach by Engle et al. (1983). Wefind that the instrumented advertising metrics satisfythe exogeneity requirement (Naik and Raman 2003,Sridhar et al. 2011; please see Online Appendix II).We estimate the model given by (7) and (8) and the

seven IV equations by defining a correlation struc-ture as follows: 4�D

t 1�Pt 1�

O1t 1 0 0 0 1�O3

t 1�E1t 1 0 0 0 1�E3

t 5∼

N401ì5, where ì is a full (8 × 8) covariance matrixto be estimated. See Online Appendix III for a com-plete description of the estimation procedure nestingan SUR IV approach in a DLM framework.

6. Empirical Results6.1. Fit and ForecastsWe investigate model performance based on in-sample fit as well as predictive performance out-of-sample. We find that the proposed dynamic model ofblood donations fits the data well in-sample (marginallog-likelihood = −21670). In comparison, a modelwithout dynamics (only POE variables and days ofthe week estimated in a regression setting) fits thedata worse (marginal log-likelihood = −31177). Com-pared to a model without donation dynamics, theproposed model fits the data significantly better; theBayesian factor differential is 507 points (greater thanthe critical difference of six, Newton and Raftery1994). We conclude that in-sample fit favors a modelaccounting for donations dynamics.8

The proposed model’s prediction accuracy is deter-mined by re-estimating our model using the first 190observations only.9 Based on the set of parameter esti-mates, we calculate k-step10 ahead of daily predictedvalues for observations not included in the estima-tion set. We find that the mean absolute percentageerror (MAPE) of the 10-step ahead daily forecasts is41%. Finally, we compare the out-of-sample perfor-mance of the proposed model to a model that includesthe day of the week and the POE media without thedonation dynamics. We find that when not allow-ing for donation dynamics the ability to correctlyforecast the level of donation is greatly decreased(MAPE of 86% versus MAPE of 41% for our proposedmodel). We present the parameter estimates inTable 2.

6.2. Paid MediaWe find that the Blood Donation paid search cam-paign (targeted toward potential donors) has a sig-nificant and positive effect on the level of donation,

8 Additionally, we investigated the possibility of different carry-over effects per media using an Adstock approach estimated witha three-stage-least-squares procedure (e.g., Dinner et al. 2014). Wefind that, for our data, our proposed model provides a better fit(details are available from the authors on request). We want tothank the review team for this suggestion.9 We split the data 60/40. We have also tried different splits yieldingsimilar results. Results are available from the authors on request.10 We present the results of a 10-step head forecast. Our resultsare robust to the number of steps, i.e., the proposed model withdonations dynamics outperforms the model without dynamics fora wide variety of step sizes. Results are available from the authorson request.

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 2 Empirical Results

Variable Parameter estimates

Carryover Effect 0082(0.71, 0.90)

Blood Donation Campaign 4P5 7065(1.17, 15.57)

YouTube Views (O1) 2093(−17010, 21.74)

YouTube Unique Users (O2) −0040(−9040, 9.32)

Facebook Page Views (O3) −3035(−12069, 6.31)

Facebook New Likes (E1) 16092(1.63, 33.07)

Facebook Unlikes (E2) −4056(−15025, 7.02)

Facebook Comments (E3) 3099(−2054, 11.31)

Monday 51057(32.29, 70.90)

Tuesday 52059(33.18, 71.82)

Wednesday 60020(40.75, 79.36)

Thursday 40092(20.98, 61.12)

Friday 30065(10.64, 50.40)

Note. Cell entries are posterior mean and 95% coverage intervals (inparentheses).

�1 = 7065. The implied elasticity (at the mean) is 0.04,which is consistent with previous findings on adver-tising effectiveness (e.g., Ataman et al. 2008). Recentresearch on paid search advertising also supports itseffectiveness as an instrument to drive clicks andsales (for example, see the chapter on paid search inCoussement et al. 2013). In the case of blood bankswe speculate that paid traffic is the closest instru-ment to donation. Specifically, a consumer who hasmade up her mind to donate blood is looking foran outlet in her area. Paid (search) media allows theblood bank to directly address this potential donorand guide her to a collection site. In other words,paid (search) media provide a solution to the donor’swants and needs and are highly effective acquisitiontools. Finally, there could also be a yellow-page effectin that a donor who knows about the blood bank willuse a search engine to find it and responds to paid(search) media to gather the information needed todonate blood.

6.3. Owned MediaOwned media includes the blood bank’s YouTube(uO11uO2) and Facebook page views (uO3). We firstdiscuss the effect of YouTube page views. The bloodbank has a YouTube channel and has posted videoson topics such as blood donations, the activities ofthe blood bank, and how donating blood helps to

improve other people’s lives. We summarize viewsacross these types of videos in one metric, YouTubeviews. Additionally, we include a second measurecapturing the number of unique users who viewvideos. We find that owned media activities on theblood bank’s YouTube channel have no significanteffect on donation levels (see Table 2; �2 and �3 arenot significant). The same holds for Facebook pageviews. We speculate that simply viewing a Facebookpage does not create enough engagement to donateblood immediately. For example, a casual visitor tothe Facebook page might be simply collecting infor-mation. However, the Facebook page allows visitorsto interact with the blood bank on a more activeengagement level, i.e., by providing Facebook likes orleaving comments (which we classify as earned mediaand discuss next).

6.4. Earned MediaFacebook allows users to comment and expressexplicit preferences via a “like” or “unlike.” We findthat earned media activities measured by these Face-book metrics have a positive and significant effecton donation levels. First, the Facebook “like” metrichas a positive and significant effect, �5 = 16092 (elas-ticity of 0.05). “Likes” not only signal the intrinsicinterest of potential donors towards the blood bankbut also reveals their preferences to their social net-works on Facebook. From the blood bank’s perspec-tive, a new “like” increases the blood bank’s (brand)presence on Facebook, e.g., via the newsfeed. Thiscould result in creating trust in the organization andits mission, potentially reducing the barriers to dona-tions. We also find that the number of “unlikes” ona Facebook page (i.e., an individual removes a pre-vious “like”) has no impact on blood donation lev-els. This result might occur because “unlikes” are notcommunicated across the network, i.e., friends are notnotified (e.g., via newsfeed) if a previously liked itemis “unliked.” Finally, we test for the impact of Face-book comments on blood donation levels. We foundno significant effect on donations (�7).

6.5. Donation Carry-Over andDay-of-Week Effects

Table 2 reports the parameter estimates of the pro-posed model. We find a significant donation carry-over effect, � = 0082, which indicates that marketingactions not only affect current levels of donations butalso future donations. Moreover, we find that dur-ing the week (Monday–Friday), on average, the levelof donations is higher compared to a Saturday (see

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Table 2). The average level of donations for a week-day versus a weekend is about 47 pints higher.

6.6. Impacts of Owned and Earned Media onthe Blood Bank

Based on these empirical results we investigate whenowned and earned media affect the performance ofthe blood bank, as measured by its value function,J �D�. We compute �J /��, the sensitivity of the valuefunction with respect to marginal variations of �, thecollective effect of owned and earned media on thenumber of blood donations, i.e., � =

∑Oj=1 �Oj

√uOj +

∑Ek=1 �Ek

√uEk.

Differentiating the value function with respect to �allows us to derive when owned and earned mediapositively (or negatively) affect the blood bank. Basedon our discussions with the blood bank, we learnedthat it aims at an average target number of donations(b) of 140 pints. For exposition, we assume that thebounds of the cost-free zone are b = 0�90b and b =

1�2b, that the discount rate is 5% per annum, and thatthe cost of shortage is three times the cost of excess.Normalizing the excess cost to one, we compute thesensitivity of the value function to marginal varia-tions of � over time. This results in novel insights.First we find that owned/earned media improve theblood bank’s performance (in terms of the value func-tion) on 199 days (out of 318), but are detrimentalon the remaining days. This split (63% versus 37%)is obtained by counting the number of days where�J /�� > 0 and where �J /�� < 0. We also find thatthere are differences between regions. Indeed, we dis-cover that, on average, the elasticity of the valuefunction with respect to owned and earned mediaequals 1.55 in the shortage region, 0.99 in the cost-free region, and −0�19 in the excess in region. Fig-ure 3 plots the sensitivity of the blood bank’s valuefunction with respect to owned/earned media overtime.

Figure 3 (Color online) Sensitivity of the Value Function of the BloodBank with Respect to O/E Media

Days

�J��

7. Investigating POE Synergies7.1. Paid Media Strategy in the

Presence of SynergyWe complement our analytical and empirical findingsto investigate synergistic effects between media.11 Forease of exposition, we first investigate the role of syn-ergy in a parsimonious normative model, where thereis only one paid media, one owned media, and oneearned media. In that case, Equation (1) becomes

dD =(

�P

√

uP �t�+�O

√

uO�t�+�E

√

uE�t�

+�PO

√

uP �t�√

uO�t�+�PE

√

uP �t�√

uE�t�

+�OE

√

uO�t�√

uE�t�− �D)

dt+ dW� (11)

where �PO is the synergistic effect of paid with ownedmedia, �PE is the synergistic effect of paid with earnedmedia, and �OE is the synergistic effect betweenowned with earned media. We obtain the followingproposition.

Proposition 5. The blood bank’s optimal paid mediastrategy when synergistic effects are present is

uP �D�∗ =(

max{

0 uP �D�/2})2

� (12)

with

uP �D�=

��P +�PO

√uO+�PE

√uE�

·�G1D+G2� if Db�

(13)

where G1 < 0 and I1 < 0, while H1 > 0.


The optimal paid media strategy, when synergisticeffects are present, is qualitatively similar to Propo-sition 1, i.e., decreasing in the level of donations inboth the shortage and excess regions and increasingin the level of donations in the cost-free zone. We alsoconducted sensitivity analyses to investigate whetherthe results from the IMC literature generalize in oursetting. First, as synergies increase, should advertis-ing spending increase? Second, if owned or earnedmedia become more effective, should the advertisingallocation change?

We find that, contrary to traditional recommen-dations, paid media spending should not alwaysincrease when synergistic effects increase since��uP �D�/��PO��uP �D�/��PE� could be both positive

11 We thank the review team for this suggestion.

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


and negative, while ¡uP 4D5/¡�OE is always nega-tive. At the same time, if owned and earned mediaincrease, any increase or decrease in paid mediaspending depends on the respective levels of thesemedia and model parameters. Finally, we find thatwhen the main effect of owned and earned mediaincreases, spending in paid media should decrease.

7.2. Estimating a Synergy ModelWe also investigate empirically the existence of syn-ergies in two different ways. First, we add all pos-sible two-way media interactions to the model givenby Equations (7) and (8). This results in seven maineffects and 21 interaction effects. We call this theFull Synergy Model. In our setting we find that theinteraction effects are highly correlated, raising con-cern about multicollinearity hampering the estimationresults. We address these concerns using a variableselection approach. We use a stochastic search vari-able selection procedure ((SSVS), e.g., Trusov et al.2010). The basic idea of SSVS is to leverage Bayesianmethods to efficiently search across all possible sub-sets of predictors in terms of effects (in our case,∑10

n=0

(21n

)

subsets could exist); we call this the Selec-tion Synergy Model. We find that both models fitthe data significantly worse than our proposed modelwith no synergy (Proposed Model, marginal log-likelihood = −21670; Full Synergy Model, marginallog-likelihood = −21696; Selection Synergy Model,marginal log-likelihood = −21714). This implies aBayesian factor of 26 for the Full Synergy Model(compared to the Proposed Model) and a Bayesianfactor of 44 for the Selection Synergy Model (com-pared to the Proposed Model). Both Bayesian fac-tors are larger than six, implying that both synergymodels fit the data significantly worse than the pro-posed model without synergies (Newton and Raftery1994).12 We conclude that, in the case of our data, syn-ergies between POE media do not increase donationlevels.

8. ConclusionBlood banks improve people’s lives with marketing.The challenging aspect of this marketing problem isdriven by the perishability of blood, which preventsblood banks from simply over-collecting blood to con-tinuously ensure that enough blood is stocked to meetdemand. Thus, blood collections must remain close tocertain target bounds. We investigate how marketingmanagers can use POE media to achieve this goal.Our results indicate that a successful POE media strat-egy for a blood bank goes beyond investing in paidmedia and capitalizing on owned and earned media.

12 This result could also emerge due to the low levels of the dailymedia activity. We thank an anonymous reviewer for this intuition.

Specifically, we first find that blood banks shouldfollow inverse allocation rules for ad spending onlyin shortage and excess regions, but not in the cost-free region. In the cost-free region, the blood bankshould increase spending even when blood donationsincrease to move the level of donations away fromthe shortage region on the fastest possible trajectory.Second, we find that an increase in the effectivenessof advertising instruments can lead to lower spend-ing on these instruments and not more, as the extantliterature suggests in for-profit settings. This result isdriven primarily by the objective of the blood bank,which is not to maximize profit but to maintain dona-tions within a target range. Third, we show that thevalue function (i.e., long-term performance) of theblood bank can decrease with owned and earnedmedia activities, as the blood bank cannot adjust themwith respect to donations, contrary to paid media.We illustrate the normative implications of the the-oretical model using donation and POE media datafrom a community blood bank. This allows us todocument that, consistent with Proposition 4, ownedand earned media decreases the blood bank’s valuefunction about 37% of the time. Thus, POE medianeed to be carefully managed and can be criticalmarketing instruments in helping to optimize blooddonations. Consequently, blood banks with limitedadvertising budgets can use inexpensive online paidmedia to efficiently manage blood donations. More-over, this approach also applies to blood banks withlarger operations, as neither the analytical results northe estimation strategy makes any assumption aboutthe size of the blood bank.

We propose two avenues for future research. First,in our study we distinguish between POE mediabased on their impact on the cost structure of the firm.There could however be other distinct characteristicsof each media type. As more detailed data becomeavailable on how donors and customers interact withPOE media (e.g., typical length of exposures, con-tributions to content, reactions to positive/negative/neutral content, etc.) a hierarchical set-up could beused to inform the estimates and to provide a betterunderstanding of the inner workings of the underly-ing consumer processes. Finally, most of the empiricaland strategic work in marketing academia is drivenby managerial problems faced by for-profit firms.Yet, next to this sector exists an ever-growing sectorof non-profit organizations. The U.S. nonprofit sec-tor grew by 25% between 2001 and 2011, and con-tributes about 5.4% to the GDP (Roeger et al. 2012).The majority of non-profits are small, often local,organizations. According to the Urban Institute, ofthe one million public charities in the United States,about 75% reported less than $100,000 in annual grossreceipts. The issues faced by these small local charities

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


are very different from the typical for-profit setting,e.g., limited marketing budgets, constrained growthopportunities, strong dependency on outside fund-ing, etc. However, despite their important roles inmodern economies, very few marketing studies havefocused on non-profits (e.g., Krishna and Rajan 2009,Arora and Henderson 2007, Ansari et al. 1996). Weadd to this literature by showing how state-of-the-artmarketing science, especially novel online marketingmethods, can be used by these small local non-profitorganizations in their pursuit of charitable goals.

Supplemental MaterialSupplemental material to this paper is available at http://dx.doi.org/10.1287/mksc.2014.0892.

AcknowledgmentsThe authors thank seminar participants at the University ofMannheim, Université de Cergy-Pontoise, Theory and Prac-tice Marketing Conference at Northwestern University, theMarketing Science Conference at Emory University, and theMarketing Dynamics Conference in Las Vegas. The authorsare grateful to the collaborating blood bank for sharing dataand in particular to its marketing director for support dur-ing this project. The authors also thank the area editor, thetwo anonymous reviewers, and the editor-in-chief for con-structive comments that helped to enrich this paper. Theauthors are listed alphabetically. The authors believe thateveryone should donate whenever possible.

ReferencesAnsari S, Siddarth S, Weinberg CB (1996) Pricing a bundle of

products or services: The case of nonprofits. J. Marketing Res.33(1):86–93.

Aravindakshan A, Naik PA (2011) How does awareness evolvewhen advertising stops. Marketing Lett. 22(3):315–326.

Arora N, Henderson T (2007) Embedded premium promotion: Whyit works and how to make it more effective. Marketing Sci.26(4):514–531.

Ataman B, Mela CF, van Heerde HJ (2008) Building brands. Mar-keting Sci. 27(6):1036–1054.

Bass F, Bruce N, Majundar S, Murthi BPS (2007) A dynamicbayesian model of the advertising-sales relationship. MarketingSci. 26(2):179–195.

Belien J, Forcé H (2012) Supply chain management of blood prod-ucts: A literature review. Eur. J. Oper. Res. 217(1):1–16.

Brodheim E, Prastacos GP (1979) The long island blood distributionsystem as a prototype for regional blood management. Inter-faces 9(5):3–20.

Burnett JJ (1981) Psychographic and demographic characteristics ofblood donors. J. Consumer Res. 8(1):62–66.

Chan TY, Wu C, Xie Y (2011) Measuring the lifetime value of cus-tomers acquired from google search advertising. Marketing Sci.30(5):837–850.

Cohen MA, Pierskalla WM (1979) Target inventory levels for a hos-pital blood bank or a decentralized regional blood bankingsystem. Transfusion 19(4):444–454.

Corcoran S (2009) Defining earned, owned and paid media.Accessed April 2, 2013, http://blogs.forrester.com.

Coussement K, De Bock KW, Neslin SA, eds. (2013) AdvancedDatabase Marketing: Innovative Methodologies and Applica-tions for Managing Customer Relationships (Gower Publishing,Surrey, UK).

Dinner IM, van Heerde HJ, Neslin SA (2014) Driving online andoffline sales: The cross-channel effects of traditional, online

display, and paid search advertising. J. Marketing Res. 51(5):527–545.

Drake AW, Finkelstein SN, Sapolsky HM (1982) The American BloodSupply (MIT Press, Cambridge, MA).

Engle RF, Hendry DF, Richard JG (1983) Exogeneity. Econometrica51(2):277–304.

Erickson G (2003) Dynamic Models of Advertising Comptition (KluwerAcademic Publishers, Boston).

Goodall D (2009) Owned, bought and earned media. AccessedMarch 28, 2013, http://danielgoodall.com/2009/03/02/owned-bought-and-earned-media/.

Halperin D, Baetens J, Newman B (1998) The effect of short-term temporary deferral on future donation. Transfusion 38(2):161–169.

Hillyer CD, Silberstein LE, Ness PM, Anderson KC, Roback JD, eds.(2007) Blood Banking and Transfusion Medicine: Basic Principlesand Practice, 2nd ed. (Elsevier Churchill Livingstone, London).

Krishna A, Rajan U (2009) Cause marketing: Spillover effects ofcause-related products in a product portfolio. Management Sci.55(9):1469–1485.

Kuksov D, Villas-Boas JM (2008) Endogeneity and individual con-sumer choice. J. Marketing Res. 45(4):702–714.

Lyons RD (1971) Blood bank bill sponsored. New York Times(December 12), 44.

Nahmias S (1982) Perishable inventory theory: An overview. Oper.Res. 30(4):680–708.

Naik PA, Raman K (2003) Understanding the impact of synergy inmultimedia communication. J. Marketing Res. 40(3):375–388.

Nerlove M, Arrow KJ (1962) Optimal advertising policy underdynamic conditions. Economica 29(114):129–142.

Newton N, Raftery A (1994) Approximate bayesian inference bythe weighted likelihood bootstrap. J. Royal Statist. Soc. Series B56(1):3–48.

Nguyen DD, DeVita DA, Hirscher NV, Murphy EL (2008) Blooddonor satisfaction and intention of future donation. Transfusion48(4):742–748.

Pessemier E, Bemmaor A, Hanssens D (1977) Willingness to supplyhuman body parts: Some empirical results. J. Consumer Res.4(3):131–138.

Pierskalla WP (2005) Supply chain management of blood banks.Brandeau ML, Sainfort F, Pierskalla WP, eds. OperationsResearch and Health Care: A Handbook of Methods and Applications(Kluwer Academic Publishers, Boston), 103–145.

Pillavin JA (1987) Temporary deferral and donor return. Transfusion2(2):199–200.

Prastacos G (1984) Blood inventory management: An overview oftheory and practice. Management Sci. 30(7):777–800.

Roeger KL, Blackwood AS, Pettijohn SL (2012) The NonprofitAlmanac 2012 (Urban Institute Press, Baltimore).

Rossi PE, Allenby GM, McCulloch R (2005) Bayesian Statistics andMarketing (Wiley, Hoboken, NJ).

Slonim R, Wang C, Garbarino E (2014) The market for blood.J. Econom. Perspect. 28(2):177–196.

Sridhar S, Mantrala M, Naik PA, Thorston E (2011) Dynamic mar-keting budgeting for platform firms: Theory, evidence andapplication. J. Marketing Res. 48(6):929–943.

Stephen AT, Galak J (2012) The effects of traditional and socialearned media on sales: A study of a microlending marketplace.J. Marketing Res. 49(5):624–639.

Trusov M, Bodapati AV, Bucklin RE (2010) Determining influen-tial users in Internet social networks. J. Marketing Res. 47(4):643–658.

West M, Harrison J (1997) Bayesian Forecasting and Dynamic Models(Springer, New York).

Whitaker BI (2013) The 2011 national blood collection andutilization survey report. U.S. Department of Health andHuman Services, Office of the Assistant Secretary for Health,Washington, DC. http://www.hhs.gov/ash/bloodsafety/2011-nbcus.pdf.

Whitaker BI, Schlumpf K, Schulman J (2011) The 2009 NationalBlood Collection and Utilization Survey Report. U.S. Depart-ment of Health and Human Services, Office of the AssistantSecretary for Health, Washington, DC.

Dow

nloa

ded

from

info

rms.

org

by [

205.

175.

118.

57]

on 1

3 Ja

nuar

y 20

15, a

t 14:

41 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

managing blood donations with marketingfaculty.washington.edu/orutz/docs/blood_marksci_2014.pdf ·...

Documents