Research Article Power to the Peers: Authority of Source ... · evant sources of knowledge...

81

Neil [email protected] Ground Floor,CP ChambersNr. Navrang Circle, Opp.Navrang High SchoolAhmedabad, GujaratIndia

Krishna [email protected] ProfessorNational University ofSingaporeMochtar Riady Building, Biz 1,07-4015 Kent Ridge DriveSingapore

Paresh [email protected] LeadDevelopment Support CentreNear Government Tubewell,BopalAhmedabad, GujaratIndia

Kapil [email protected] TrusteeJatan TrustVinoba Ashram, GotriVadodara, GujaratIndia

Power to the Peers PATEL, SAVANI, DAVE, SHAH, KLEMMER, PARIKH

Research Article

Power to the Peers: Authority ofSource Effects for a Voice-BasedAgricultural Information Servicein Rural India

Neil PatelKrishna SavaniParesh DaveKapil ShahScott R. KlemmerTapan S. Parikh

Abstract

Online communities enable people to easily connect and share knowledgeacross geographies. Mobile phones can enable billions of new users in emerg-ing countries to participate in these online communities. In India, where socialhierarchy is important, users may overvalue institutionally recognized authori-ties relative to peer-sourced content. We tested this hypothesis through a con-trolled experiment of source authority effects on a voice-based agricultural in-formation service for farmers in Gujarat, India. Over a two-week period,305 farmers were sent seven agricultural tips via automated phone calls. Thesame seven tips were each voice-recorded by two university scientists and twopeer farmers. Participants received a preview of the tip from a randomly as-signed source via the automated call, and then they played the remainder ofthe tip by calling a dedicated phone number. Participants called the follow-upnumber signiªcantly more often when the tip preview was recorded by a peerthan a scientist. On the other hand, in interviews conducted both before andafter the experiment, a majority of farmers maintained that they preferred re-ceiving information from scientists. This stated preference may have been ex-pressing the more socially acceptable response. We interpret our experimentalresults as a demonstration of the demand for peer-based agricultural informa-tion dissemination. We conclude by identifying design implications for peer-to-peer information services for rural communities in India.

1. IntroductionIndian society has been noted for the prominent role that hierarchy playsin it (Dumont, 1981), leading to a tendency to defer to authorities(Savani, Morris, & Naidu, 2012). This deference effect has been demon-strated in a range of scenarios, from the workplace (Storti, 2007) to familylife (Levine, Sato, Hashimoto, & Verma, 1995). As broader segments ofthe population come online, many of them via mobile phones, this socialdynamic could also play out online. In contexts that include informationsources from all social strata, norms that place pressure on individuals to

© 2013 USC Annenberg School for Communication & Journalism. Published under Creative Commons Attribution-Non Commercial-Share Alike 3.0 Unported

license. All rights not granted thereunder to the public are reserved to the publisher and may not be exercised without its express written permission.

Volume 9, Number 2, ICTD2012 Special Issue, 81–93

82 Information Technologies & International Development

POWER TO THE PEERS

Scott R. [email protected] ProfessorStanford UniversityGates Computer Science #384Palo Alto, CA 94305-9035USA

Tapan S. [email protected] ProfessorUniversity of California,Berkeley102 South HallBerkeley, CA 94720-4600USA

defer to authority ªgures may lead people to overvalue authority sourcesrelative to peer-sourced content.

India has also been characterized as a collectivist culture (Verma &Triandis, 1999) with a rich legacy of cooperation and sharing through peernetworks. These values are also found in many online communities. Peershave been demonstrated to be scalable, accessible, trusted, and locally rel-evant sources of knowledge (Mamykina, Manoim, Mittal, Hripcsak, &Hartmann, 2011). Earlier work demonstrates that farmers who were pro-vided access to a voice-based information forum for agriculture engagedin rich exchange and found the information provided to be highly relevant(Patel, Chittamuru, Jain, Dave, & Parikh, 2010). However, while farmersenjoyed hearing the questions and experiences of other farmers, mostgave a stated preference for receiving advice directly from authorities.

This article investigates how the authority of an information sourceaffects the likelihood that farmers will follow up on the information. In acontrolled experiment (see Figure 1), 305 users of the Avaaj Otalo forumwere called with seven farming tips recorded by two types of sources:peer farmers and scientists from local agricultural universities. To isolatethe effect of the source’s authority on participants’ subsequent actions,the tip content itself was held identical across the two sources. After abrief introduction from the source, they heard a preview of the agricul-tural tip, and then they were told that they would be able to hear theconclusion of the tip recording if they hung up and dialed another num-ber. Participants chose to call back and listen signiªcantly more frequentlywhen the tip was recorded by a peer farmer. Still, participant farmers con-tinued to state in interviews before and after the experiment that theypreferred receiving information from authorities. The stated preferencesmay have been biased by the fact that the interviewers were perceived asauthorities, leading participants to provide a more socially desirableanswer. The results indicate the demand for peer-based informationdissemination.

1.1 Authority in Indian SocietySome have described social hierarchy as a deep-rooted feature of Indiansociety (Appadurai, 1998; Dumont, 1981; Mines, 2009). Researchers haveobserved a “deference syndrome” in the Indian work environment, inwhich subordinates act against their better judgment and struggle toexpress views independent of their bosses’ views (Storti, 2007). Whilethese observations could have come from any work environment, deferen-tial behavior in India may be especially strong. One study of Indian andAmerican college-age individuals found that Indians adjusted their choicesin deference to authority, even when the decisions went against personalpreference, and even when the subject was told that the authority wouldnever know about the decision; Americans, by contrast, did not (Savani etal., 2012). In another context, researchers found that videos featuringlocal high-status or authoritative individuals can be highly effective in per-suading healthy practices in villages (Parmar, Keyson, & de Bont, 2009;Ramachandran, Canny, Das, & Cutrell, 2010).

While hierarchy is inºuential, Indian society also has a strong culture ofpeer-to-peer exchange, rooted in a group orientation (Sinha, Sinha,Verma, & Sinha, 2001; Verma & Triandis, 1999). The Honey Bee Network

has demonstrated that there is both a signiªcantsupply of and demand for local knowledge andinformation to be shared among rural farmers.1 Dig-ital Green found that including peer farmers in vid-eos of new practices led to increased likelihood ofadoption (Gandhi, Veeraraghavan, Toyama, &Ramprasad, 2007). When compared to authorities,peers more easily establish common ground,because they “speak their language.” A 2005nationwide survey by the International Food PolicyResearch Institute found that “other progressivefarmers” were the most popular source of informa-tion on agricultural technology. Traditional authoritysources (agencies, technicians, NGOs) were at ornear the bottom of the list (Birner & Anderson,2007).

1.2 Information Processing and CultureSome information processing practices have beenshown to vary by culture. For example, studies havefound that people in different cultures pay attention(Maddux & Yuki, 2006) and incorporate (Kitayama,Duffy, Kawamura, & Larsen, 2003) different contex-tual information. The elaboration likelihood model(ELM) was developed by social psychologists toexplain how people process various cues while pro-cessing information (Petty & Cacioppo, 1986). TheELM differentiates between systematic informationprocessing (where people form attitudes based onthe intrinsic strength, quality, or persuasiveness of

the message) and heuristic processing (where theyrely on heuristics like “authorities should betrusted,” “long messages are valid messages,” or“majority opinions are usually true”; Chaiken &Maheswaran, 1994). The ELM predicts that peopleresort to heuristic processing in “low-involvement”situations, or instances where they are not highlypersonally vested in the outcome.

The applicability of ELM can be inºuenced by cul-tural norms. An ELM experiment investigating theeffects of the race of information sources found thatwhite American subjects were systematically pro-cessing messages in a low-involvement situationwhen the source of the information was black. Inother words, where the ELM would predict thatwhite participants would not pay attention to thecontent of the message in forming an opinion, theywere doing so if and only if the source was black(White & Harkins, 1994). A follow-up experimentconcluded that white participants were stronglymotivated to attend to the black source to avoidbeing perceived as racist (White & Harkins, 1994).

2. Experiment Design and Method2.1 BackgroundAn earlier ªeld study showed that 65% of AvaajOtalo users expressed a preference for receivinganswers exclusively from staff and scientists workingwith our implementing partner NGO, Development

Volume 9, Number 2, ICTD2012 Special Issue 83

PATEL, SAVANI, DAVE, SHAH, KLEMMER, PARIKH

1. Honeybee.org

Figure 1. In this experiment, tips from farmers (left) and scientists (right) were alternately sent to rural Indiansthrough automated phone calls. After hearing the tip, subjects were given the option to hear more informationby calling a phone number. An experiment captured the number of follow-up calls that were made in response totips from farmers versus tips from scientists.

Support Centre (DSC). The remaining 35% ofrespondents wanted both authority and peerresponses; none said they preferred information onlyfrom peers (Patel et al., 2010; Savani et al., 2012).Participants stated that DSC’s experts had a greaterbreadth and depth of knowledge than peers, thatexperts were more articulate, and that “scientiªc”knowledge was more reliable than “experiential”knowledge. The prevailing sentiment seemed to bethat farmers were unreliable, and even incapable ofcontributing high-quality responses:

[Only] when these other farmer’s questions will beanswered by an expert, then I will get to learnfrom [answers]. Farmers don’t know everything,right? What most of what the farmers talk aboutis common knowledge to us. So I am interested inlistening to what the experts say about the ques-tions on Avaaj Otalo.

After the pilot, DSC recruited staff members andscientists from local agricultural universities to partic-ipate as “expert” responders for the service. Nofarmers were targeted in this recruitment. In discus-sions with DSC staff, the farmers indicated that staffand scientists would be best suited to provide high-quality, accurate advice. DSC’s weekly radio programand quarterly newsletter already routinely proªledfarmers, highlighting their innovations. DSC’s reluc-tance to include expert farmers as experts waslargely based on logistical concerns, including thecomplexity of managing a larger and more distrib-uted group of experts. But many DSC staff alsoshared the farmers’ lack of faith in peer-providedadvice.

2.2 Research Question and HypothesisThe farmers’ stated preference for information fromauthorities may be a reºection of underlying socialnorms favoring authorities. On the other hand,many farmers may also not have had prior access toa consistent, high-quality source of peer informa-tion. We wanted to determine whether rural Indianfarmers would engage equally with informationfrom their peers if it could be provided with thesame quality and consistency as information fromexperts. To do this, we designed a controlled experi-ment to answer the following research question:

Given the same informational message, are ruralIndians more inºuenced by the information if itcomes from an institutional authority ªgure, com-pared to a peer?

Prior ªeld and experimental research (Patel et al.,2010) suggested the following hypothesis:

Rural Indian farmers are more likely to act uponinformation presented by an authority than thatpresented by a peer.

2.3 ParticipantsParticipants were recruited from a pool of 1,014phone numbers that had called Avaaj Otalo at leastonce during the prior nine months. Two paid assis-tants ºuent in Gujarati and familiar with Avaaj Otalorecruited participants over the phone over a two-week period. Participation in the experiment wasintroduced as an opportunity to participate in a trialof a new service, Avaaj Otalo Margdharshan Seva(literally, Avaaj Otalo’s Direct Information Service).

Farmers were told that AO Margdharshan wouldprovide them with recorded agricultural tips deliv-ered via automated voice phone calls from the AvaajOtalo phone number. Participants were told that thetips would come from farmers and scientists acrossthe state who were associated with DSC. After hear-ing the description, farmers were asked if theywanted to subscribe, at no cost to them. If theyagreed, basic demographic information was col-lected, and their number was included in the trial.All farmers who agreed to participate were acceptedinto the study.

Basic information for these participants is shownin Table 1. Most participants were small-scale ormarginal farmers. All were male, since the originalpool led to only male callers. Most of the districtsand crops grown in the state were represented.Twenty-eight users participated in a pilot designedto validate our scripts, the usability of the voiceinterface, and the relevance of the information.Our analysis is based on data from the remaining277 users. After the study, as a thank-you gift, DSCmailed all participants a booklet with all of the tipsin full, along with supplemental farming-related arti-cles and DVDs.

2.4 Study DesignThe experiment was conducted entirely over thephone. Each participant received seven tips in thesame order, as well as an even spread of tips fromeach of the four sources (two farmers and two sci-entists). Participants were randomly assigned to oneof four tip schedules (see Table 2), counterbalancingtips and sources to achieve an equal number ofevery combination.


POWER TO THE PEERS

2.5 Study MaterialsThe phone calls for the experiment were executedover an ISDN primary rate interface (PRI) line con-nected to a commodity UNIX server. PRI lines sup-port up to 30 simultaneous calls, and a single linecan map 90 distinct phone numbers. We recordedand assigned a distinct phone number to each tip-source combination (7 x 4 ! 28), logging the iden-tity of each inbound call to count the number offollow-ups. Phone numbers were not assigned ran-domly; each source was assigned a continuousseries of four phone numbers for each tip. While wedo not believe any speciªc numbers were easier torecall than others, it is possible that this could haveintroduced some bias into the experiment.

The tips and the previews themselves were devel-oped by agricultural staff members at DSC, andthen they were reviewed for accuracy by outsidescientists. The tips were designed to be factuallyaccurate and clearly articulated, offering practicalinformation that was relevant for a wide rangeof farmers. It was important that the tip contentwas equally plausible coming from either a scientistor a farmer. To achieve this, DSC staff membersrecommended using “farmer-friendly language,”which was colloquial and playful, avoiding technical

jargon. Two tips dealt with cotton, which is grownby a large portion of Gujarati farmers. Two othertips dealt with animal husbandry, which is relevantto nearly all farmers, as most keep animals forhome dairy consumption, manure, or labor. Theother four tips discussed disease management,orchard promotion, drip irrigation, and soil testing.See the following text for a sample tip preview andthe associated tip. The technical terms in Englishwere replaced by colloquial Gujarati words. A localfarming expert who had signiªcant experienceworking with farmers veriªed the appropriatenessof the language.

Preview: Your animals are very much prone toseveral serious diseases like hemorrhagic septicemiaand foot and mouth disease with varying frequency.Once the animal gets sick, there are so many trou-bles. You need to call the vet, you need to spendmoney on visits and medicines, it is very much timeconsuming for you, and sometimes your agriculturalactivities get delayed. If the sickness is more serious,and the animal gets weakened, then it is a long-term damage. For milking animals like cow and buf-falo, milk production will go down. If you want tosave your animals from all these troubles and wantto ensure health of animals, the very simple and



Table 1. Subjects by Demographics.

305

20 (of 26 in Gujarat)

33 (mean), 30 (median)

10 acres (mean), 7 acres (median)

8th grade (median)

60%

Peanuts, millet, lentils, sesame, beans, corn, castor seed, cumin, mustard, tobacco,wheat, rice (of 26 crops grown in the state)

96%

Table 2. Tip Schedules.*

Tip1 Tip2 Tip3 Tip4 Tip5 Tip6 Tip7

Grp1 S1 S2 P1 P2 S1 P1 S2

Grp2 S2 S1 P2 P1 S2 P2 S1

Grp3 P1 P2 S1 S2 P1 S1 P2

Grp4 P2 P1 S2 S1 P2 S2 P1

*Subjects were randomly assigned one of the four tip schedules. The tips were assigned to all sources equally.The tips’ sources alternated between peer (P1, P2) and scientist (S1, S2) sources.

cheap way is timely vaccinations. To receive informa-tion on which vaccinations should be done forwhich disease, when, and where the service is avail-able, listen to the following instructions.

Tip: In Gujarat, we need to vaccinate the ani-mals, especially for foot and mouth disease and HS.The germs of HS may cause the disease while theanimal grazes on moist grasses, especially in Julyand August. The animal should be vaccinated forthis disease in the months of April and May. Butdon’t worry if you have missed it, you can do it evenin this month. If your area has experienced this dis-ease in previous years, better to vaccinate the ani-mal every six months. While the foot and mouthdisease generally occurs in summer, and the vaccina-tion should be done between October and Decem-ber, better would be to vaccinate the animals at six-month intervals. To protect the animal frombrucellosis, heifers with four to nine months of ageshould be vaccinated once in a lifetime. Vaccinationservice is freely available from the state government.Please contact the nearest animal dispensary.

We recorded two speakers for each source typeto mitigate individual effects. The scientists wereboth retired professors; one from soil science, theother from agronomy. Both were in their 60s andhad prior experience recording scripted agriculturalmessages for radio programs. The farmers werefrom two districts in Gujarat. Both had attendedDSC-organized events in the past. One was in his50s, farmed three acres of land, and had been for-mally schooled through the 10th grade. The otherwas in his mid-20s, farmed one acre, and was alsoschooled through the 10th grade.

The four selected individuals had no prior ofªcialdesignation with either DSC or the Avaaj Otalo ser-vice. The tips were recorded in quiet ofªce spacesusing a Macbook Pro’s built-in microphone. Weasked the sources to study and practice each tipcarefully before recording to ensure a smooth deliv-ery. We also asked them to internalize the messageas if they had generated the tip themselves. The tipswere re-recorded when a speaker misspoke, stut-tered, or was otherwise unnatural in his delivery.

2.6 ProcedureThe original automated call provided backgroundand motivation for a topic, but it was limited to aproblem statement or high-level description of aprescribed practice. To learn the full solution, includ-

ing implementation details, participants could learnmore information by calling the provided phonenumber. The AO Margdharshan “system” voiceinterface was similar to the Avaaj Otalo service par-ticipants had previously used. If the participantplaced a return phone call at their own expense, thecall provided a real-world measure of the partici-pant’s assessment of the original message’s value.While adoption of the advice is the theoretical goldstandard for inºuence, our approach allowed us totest our hypothesis within a reasonable timeframeand budget.

Figure 2 shows the structure of the automatedphone calls used for the experiment. Each callbegan with a welcome prompt which reminded theuser about the service and emphasized that the tipscame from scientists and farmers from across thestate of Gujarat. The tip source then introducedhimself. The farmers spoke their names and loca-tions: village, block, and district. The scientists spoketheir name (preceded by the title Doctor) and uni-versity afªliation, introducing themselves as retiredprofessors. Next, the speakers recited the tip, endingwith instructions on how the listener could obtainmore information by calling the provided phonenumber. We marked the initial call as complete ifthe listener stayed connected to this point. Afterthat, the source re-stated his name to sign off. Thisrepetition, along with limiting farmer introductionsto simply name and location, was intended to createa strong authority manipulation. Finally, the systemrepeated the follow-up phone number and providedthe option to listen to this message again. Thisprompt repeated automatically three times beforethe call self-terminated.

The seven tips were sent to subjects over thecourse of two weeks, with a new tip every twodays. Twenty-eight participants were randomlyselected to pilot the experiment. The pilot conªrmedthat most of the phone calls were, indeed, beingreceived and completed, and that the follow-up ratewas within an acceptable range for data analysis.Pilot participants also responded that that the tipswere useful and credible, and that the callback pro-cedure was convenient and affordable. Based onthis satisfactory feedback, calls for the remaining277 participants were scheduled. We began with aninitial reminder call about AO Margdharshan, urgingsubjects to pick up the following calls from thisnumber and listen to the tips carefully. The seven


POWER TO THE PEERS

tips were delivered over a two-week period accord-ing to the assigned tip schedules.

3. ResultsOf 1,883 total attempts to contact the 277 partici-pants, 1,316 (70%) calls were successful, with theperson who picked up listening to the full tip pre-view and instructions at least one time through. Of667 successful calls from a peer farmer source,72 (10.8%) resulted in a follow-up. For the scientist-recorded tips, 53 of 649 (8.2%) successful callsresulted in a follow-up. We analyzed the data usinglogistic hierarchical linear models (HLMs), treating tipcalls as nested within participants. This analysisaccounts for dependencies in response likelihoodwithin each farmer, since some farmers might bemore likely to respond to any given tip than otherfarmers. At the same time, this analysis approachassessed the impact of the experimental manipula-tion on response likelihood (Raudenbush & Bryk,2001). A dummy variable indicating whether partici-pants called back in response to the tip was thetrial-level dependent measure; the source of themessage was the trial-level predictor variable. Therewas a signiªcant effect of source, indicating thatfarmers were signiªcantly more likely to call backafter hearing a message from a peer than from ascientist (log odds ! 0.47, odds ratio ! 0.64,

z ! 2.08, p " 0.05; see Figure 3). Follow-up logisticHLMs conªrmed that the two peers elicited a similarrate of response (log odds ! –0.10, odds ratio !0.90, z ! 0.35, p ! 0.73), as did the two scientists(log odds ! 0.34, odds ratio ! 1.40, z ! 1.04,p ! 0.30).

3.1 Follow-Ups by Age, Farm Size, andEducationLogistic HLMs showed that participants’ ages didnot predict their likelihood of calling back, nor didage inºuence the difference between response ratesin the peer and expert conditions. The size of theparticipants’ farmland also did not predict their like-lihood of calling back, nor did it inºuence the differ-ence between response rates in the peer andscientist conditions. Farmers with more education(eighth-grade education or higher) were signiªcantlymore likely to call back in response to the tip(log odds ! 0.122, odds ratio ! 1.13, z ! 2.26,p " 0.05), and they were marginally more respon-sive to peers than to scientists (log odds ! –0.115,odds ratio ! 0.89, z ! 1.75, p ! 0.080). To explorethis interaction further, we split the data by medianeducation and found that, whereas farmers withless than eight years of education were equally likelyto respond to peers and scientists, farmers withmore education were signiªcantly more likely to



Figure 2. The prompt ºows for the inbound tip (a) and outbound follow-up information (b) phone calls. Thesolid boxes contain prompts spoken by a voice representing the AO Margdharshan Seva tip service. The dottedboxes are the voices of either the peer or authority source. The voice on a tip would be the same voice heard onthe corresponding follow-up call.

respond to peers than to scientists (log odds !–0.99, odds ratio ! 0.37, z ! 3.32, p ! 0.001; seeFigure 4).

3.2 Post-Study InterviewsStarting one week from the end of the study, 34randomly selected participants, including both call-ers and noncallers, were interviewed over the phoneusing a semistructured protocol. The interview wasconducted in Gujarati by a native speaker. At twopoints in this protocol, participants were asked tostate whether they preferred to receive informationfrom scientists or from peers. Of the responses,42% explicitly stated a preference for scientists,19% stated a preference for farmers, and 39% saidthat either they had no preference, or that bothwere preferable. On the other hand, 26% of inter-viewees were able to recall some detail about theidentity of at least one of the farmer sources (suchas name or where they were from), compared toonly a 13% recollection rate for the scientistsources. The sample was too small for these differ-ences to be signiªcant.

Those in favor of farmer information cited theirpractical knowledge and ability to speak from expe-rience. Take one response, for example:

I usually go by my experience and when farmerstalk about their experiences I like that better. Wehave spent most our lives farming so naturally Iwould like information from farmers.

Advice from farmers is important, as they havelocal information. Different areas have differentcrops, so local experience is important. Scientistshave to discover or invent new things in order togive advice. Farmers have experiences every

10–15 days which they can talk about. Scientiststake longer to do their experiments and get theirresults.

[I prefer information from] farmers, becausethey are experienced. I can give you any informa-tion because I am experienced. . . . [W]ithoutexperience, how can I give you advice? This isfarming, anything can happen, whether it rains orºoods is in the hands of God. Such situations canonly be handled by an experienced person.

Several respondents said they preferred informa-tion from peer farmers, because they spoke in amore understandable language (despite the tipsbeing provided in the same language for both). Takeanother interviewee’s response, for example:

Information given out by farmers is more clear.Scientists will not be able to explain clearly like or-dinary farmers. Farmers talk in our language.

When farmers give the message I feel that I canunderstand, but when scientists speak it isdifªcult as they speak differently. I like the farmersas they talk in a simple language. Maybe theinformation from the scientists is better, but Ican’t understand their high-level language, sowhat’s the point of listening to them?

At the same time, farmers appreciated that theinformation coming from scientists was backed bythe latest facts and more rigorous experiments. Onerespondent had this to say on that account:

I trust scientists and authorized people more, asthey are dependable. Farmers do trial and error,which is not very dependable.

I think scientists give better information. These


POWER TO THE PEERS

Figure 3. Aggregate follow-up rates by source for alltips.

Figure 4. Follow-up rates for each source, split by levelof education. Better educated participants followed upsigniªcantly more when the tips came from peers.

days agriculture and farming have become a veryscientiªc process.

A notable number of interviewees found infor-mation from both sources to be valuable. These par-ticipants added how the theoretical knowledge ofscientists and practical, experience-based knowledgeof farmers were complementary. One intervieweeoffered the following on that point:

Both [provide good information], as scientists giveinformation which they get from their lab experi-ments and farmers speak of their actual experi-ence.

[I value] both, as a farmer is also a type ofscientist, as he has real life experiences.

I would prefer messages from those peoplewho have tried it and done things practically. Sci-entists conduct experiments and get results, andfarmers also have actual experiences. So informa-tion from both of them will prove to be useful.

3.3 Enthusiasm for the ServiceInterviews also provided other feedback about whatparticipants liked and did not like about the service,whether the tips were useful, and any other issuesor concerns they faced. The service was generallyreceived enthusiastically, with many reporting thatthe quality and practical usefulness of informationprovided was its best aspect. One interviewee addedthe following:

The information is very useful and was deliveredin a timely manner. Animal rearing informationwas especially useful. When I got the ªrst call Ithought the service wouldn’t be [very] useful, butI changed my opinion as more information camethrough the subsequent calls.

For one illiterate participant, the service was use-ful enough to go to signiªcant lengths to keep trackof the various callback numbers:

Yes, I had no problems listening to the message.In fact I have been waiting eagerly for thesephone calls for many days. The service seems tohave stopped since few days, why is that? I usedto write the number on the phone and ask some-one to type in the numbers as I am illiterate andcannot recognize letters. I sometimes assign acharacter to every phone number so that I recog-nize that it is from that particular person. In fact Isaved [AO’s] number that way when you hadcalled me previously, so this time when you calledI knew it was you. I store very few numbers sothis system works.

The most common complaint from participantswas that the full informational message was notprovided in a single call, requiring them to use air-time for the follow-up call. One interviewee sharedhis perspective:

The information in the message is not complete,and we have to call the number which we getcharged for. I have made several calls, and I havelost 50 to 60 rupees’ credit in getting this infor-mation.

Many interviewees (44%) mentioned that thecost of the outbound phone call factored into theirdecision to follow up. Several participants reportedthat they wanted to call back, but were either con-cerned about their airtime balance or didn’t keepany balance at all, using their phone only forinbound calls. Few reported difªculty in recordingthe callback phone numbers, which was done eitherwith pen and paper, or by entering the numberdirectly on the phone.

Some callers not included in the original recruit-ment also called the follow-up numbers (these call-ers are not included in the data analysis). Thesefarmers had gotten the numbers from a friend orrelative who was a participant. Interviews alsorevealed that participants were using call-recordingcapabilities built into their phones to store the tips,later replaying the tips for friends, family, orthemselves.

The enthusiastic response to AO MargdharshanSeva prompted DSC to retain it as a regular serviceafter the study, with tips recorded mostly by staffmembers and farmers who are permitted to recordresponses.

4. DiscussionThis study’s main ªnding is that the informationsource did, indeed, matter for farmers, albeit not inthe expected manner. Farmers followed up sig-niªcantly more frequently when presented the sameinformation by peer farmers compared to authori-ties. In this section, we discuss our results and pro-vide some explanations for the discrepancy betweenthe farmers’ behavior and their stated preferences ascollected from interviews.

4.1 Authorities in Word, Not in DeedFarmer responses during the interviews may havereºected some social desirability bias (Fisher, 1993).Farmers may have been answering based on what



they believed to be the most socially acceptableanswer or that which reºected most positively onthem. There also could have been a response bias—answering questions based on what the intervieweethought the interviewer wanted to hear (Paulhus,1991). Subjects likely viewed the researchers, whowere conducting the interviews, as scientiªc authori-ties, as well. On the other hand, the decision to fol-low up on a tip was made without socialsanctioning from authorities. Researchers havenoted that social norms are situationally activated,particularly those injunctive norms that guide behav-ior based on how one thinks others perceive theiractions (Cialdini, Reno, & Kallgren, 1990).

4.2 The Power of PeersAgricultural extension programs in India focus ontraining agricultural scientists from universities todisseminate technologies and practices. This experi-ment showed that farmers followed up on informa-tion provided by peers more than on the sameinformation when it came from scientists. This studycorroborates prior work (Gandhi et al., 2007; Patelet al., 2010) suggesting that farmers should bemore deeply integrated into the knowledge diffu-sion process for effective knowledge transfer in agri-culture. A common sentiment expressed duringinterviews was that experience-based knowledgefrom other farmers is a necessary complement tothe hard evidence-based recommendations of scien-tists. In recent years, the Indian government hasexperimented with more participatory approaches toextension, including working through local farmergroups and NGOs, and even enlisting local govern-ment (panchayat) ofªcers as para-extension workers(Sulaiman, 2003).

While farmers commonly exchange adviceinformally with friends and neighbors (Birner &Anderson, 2007), word-of-mouth can lead to misin-formation. Relying on one’s immediate friends, rela-tives, and acquaintances limits the potential qualityand breadth of information that can be obtained.We have directly observed farmers unacquaintedwith knowledgeable and innovative farmers livingjust a few kilometers away, often farming the samecrop. This study shows that receiving informationfrom peers can have higher demand than receivingthe same information from scientists. Combining

crowdsourced ratings and moderation to these richpeer-to-peer exchanges represents a “best of bothworlds” scenario, ensuring quality while maintainingconsistency, scale, diversity, and breadth.

Most ICT4D projects are coordinated with localpartners embedded in the target communities. It iscommon for ICT4D researchers to defer to theexpertise of these local partners, particularly in mat-ters related to local practices or culture. Throughoutour partnership, DSC relied on its well-trained staff,whom it trusted to answer questions and providecontent for Avaaj Otalo. While this approach hasbeen successful in providing a useful, efªcient ser-vice to farmers, our results indicate that, together,we may have underestimated the demand for peerinformation exchange. We are working with DSC todesign ways for farmers to participate more effec-tively in responding to questions and content. Thisincludes providing incentives and recognition, andlowering the costs and other barriers for farmers toparticipate.

4.3 Did the Tip Content Inherently Favora Source?If the tips’ content or linguistic structures were notbelievable for the speaker, then a participant mayhave been motivated to call out of curiosity or incre-dulity (“Does this farmer know what he’s talkingabout?”). There was no evidence in the post-studyinterviews that the credulity of the tips’ sources wasin doubt. As an additional check, the tip contentwas independently rated by 20 Gujarati readers onAmazon’s Mechanical Turk.2 The Turkers were pre-sented with each tip’s introduction in Gujarati script.The task ªrst asked for a summary of the tip as acheck to ensure it was understood and the Turkerswere putting sufªcient effort into the task. TheTurkers were then asked to answer two questionsfor each tip:

1. Who is most likely to have given this tip: ascientist or a farmer?

2. Who is more appropriate to provide the res-olution information to this tip: a scientist ora farmer?

For each question, seven options were given. Theªrst option was, “A farmer is very likely/very muchmore appropriate to give this tip/resolution.” Theseventh option was, “A scientist was very likely/very

2. See mturk.com


POWER TO THE PEERS

much more appropriate.” The intermediate optionssubstituted “moderately,” “slightly,” and “equally”as descriptors for likelihood and appropriateness. Forthe 15 surveys that provided correct summaries forthe tips, no signiªcant deviation was observed foreither question when t-tests were applied compar-ing the mean and variance to the midpoint of thescale. We caution that these results are only sugges-tive, given the small sample size and ambiguityabout how qualiªed the participants were to judgethe content.

4.4 Limitations of the StudyA future study will investigate what aspects of peer-sourced information yielded a higher follow-up rate.Farmers may have been more attracted by the famil-iarity of the accent, the novelty of the source, orsome sense of camaraderie with fellow farmers. Par-ticipants may have been curious to hear advice froma farmer that they typically would receive from anoutside expert.

Participants may also have been unclear aboutwhat would happen in the follow-up call, especiallythe ªrst time they decided to follow up. The initialcall did not explicitly state that the follow-up callwould deliver the conclusion, and that it would beanother recorded message. Participants may havecalled back with the expectation that the tip wouldbe delivered by a different person, or perhaps thatthey would speak with a live person. On the otherhand, these uncertainties would have been resolvedfor any farmer who called back the ªrst time. Inpost-study interviews, no participant indicated thatsuch a confusion existed at any time, which wasasked explicitly in several interviews. Still, the overallfollow-up rate was also low (9.5%), perhapsreºecting farmers’ frustration in having to pay toaccess a service that was originally advertised as toll-free.

To avoid disclosure of the design in advance ofthe experiment, study participants were invited toopt into a trial of a new information service, ratherthan a research study. After the study, DSC sent allparticipants a booklet with the full content of all thetips, along with supplemental articles and DVDs, asa thank-you gift. DSC had used scripted content inother media projects without explicitly disclosing thisscripting to people; this study elected to do thesame. The design was approved through a standarduniversity IRB (institutional review board) process.However, it is important to note that the researchers

considered the steps taken above to be appropriate,given the beneªcial nature of the content and thetips provided, and given our experience workingwith the partner organization and participant com-munity. Using subjective judgment for a study’sappropriateness relieves some of the incompatibili-ties between the nature of ICT4D research and theIRB process (Sterling & Rangaswamy, 2010). How-ever, going this route puts the onus on researchersto vet their choices with local partners to employethically appropriate procedures.

Future research is required to generalize theseresults, as the Gujarati farmers may not be represen-tative of all farmers. In particular, their perceptionsof authorities and their willingness to seek informa-tion may differ from other farmers in India. Thesespeciªc farmers, who were all connected to DSC insome manner as early adopters of Avaaj Otalo, maynot even be representative of farmers in Gujarat.The way in which users interact with the messageboard is also likely to change and evolve over time,reºecting their experiences and learnings within andoutside the system.

5. ConclusionThis article presents a controlled experiment testingthe inºuence of authority on agricultural informa-tion dissemination to rural Indian farmers via avoice-based phone information service. Contrary tostated preferences, farmers followed up signiªcantlymore to agricultural tips when they were deliveredby peer farmers, as compared to when the sameinformation was presented by agricultural scientists.This result demonstrates that there is a signiªcantunmet demand for high-quality peer-provided infor-mation for farmers in rural India, and that in somesense, this demand is greater than that for informa-tion from established authorities. !

AcknowledgmentsThe authors thank Development Support Centre,Jigar Patel, and Hiral Shah for their assistance indeveloping and implementing this research. Thisresearch was funded by Nokia Research and theNational Science Foundation under Grant No. IIS-1054332. This is a revised version of an earlier arti-cle that appeared in the Proceedings of the FifthInternational Conference on Information and Com-munication Technologies and Development.



ReferencesAppadurai, A. (1988). Putting hierarchy in its place.

Cultural Anthropology, 3(1), 36–49.

Birner, R., & Anderson, J. R. (2007). How to makeagricultural extension demand driven? The caseof India’s agricultural extension policy (vol. 729).Washington, DC: International Food Policy Re-search Institute.

Chaiken, S., & Maheswaran, D. (1994). Heuristicprocessing can bias systematic processing: Effectsof source credibility, argument ambiguity, andtask importance on attitude judgment. Journal ofPersonality and Social Psychology, 66(3), 460–473.

Cialdini, R. B., Reno, R. R., & Kallgren, C. A. (1990).A focus theory of normative conduct: Recyclingthe concept of norms to reduce littering in publicplaces. Journal of Personality and Social Psychol-ogy, 58(6), 1015–1026.

Dumont, L. (1981). Homo hierarchicus: The castesystem and its implications. Chicago, IL: Univer-sity of Chicago Press.

Fisher, R. J. (1993). Social desirability bias and thevalidity of indirect questioning. Journal of Con-sumer Research, 20(2), 303–315.

Gandhi, R., Veeraraghavan, R., Toyama, K., &Ramprasad, V. (2007, December). Digital green:Participatory video for agricultural extension. InInternational Conference on Information andCommunication Technologies and Development,2007 (pp. 1–10).

Kitayama, S., Duffy, S., Kawamura, T., & Larsen, J. T.(2003). Perceiving an object and its context indifferent cultures: A cultural look at new look.Psychological Science, 14(3), 201–206.

Levine, R., Sato, S., Hashimoto, T., & Verma, J.(1995). Love and marriage in eleven cultures.Journal of Cross-Cultural Psychology, 26(5), 554–571.

Maddux, W. W., & Yuki, M. (2006). The “ripple ef-fect”: Cultural differences in perceptions of theconsequences of events. Personality and SocialPsychology Bulletin, 32(5), 669–683.

Mamykina, L., Manoim, B., Mittal, M., Hripcsak, G.,& Hartmann, B. (2011, May). Design lessons fromthe fastest Q&A site in the west. In Proceedingsof the 2011 Annual Conference on Human Fac-tors in Computing Systems (pp. 2857–2866).

Mines, M. (2009). Conceptualizing the person: Hier-archical society and individual autonomy in India.American Anthropologist, 90(3), 568–579.

Parmar, V., Keyson, D., & de Bont, C. (2009). Persua-sive technology to shape social beliefs: A case ofpersuasive health information systems for ruralwomen in India. Communications of the Associa-tion for Information Systems, 24(1), 25.

Patel, N., Chittamuru, D., Jain, A., Dave, P., & Parikh,T. S. (2010, April). Avaaj Otalo: A ªeld study ofan interactive voice forum for small farmers in ru-ral India. In Proceedings of the 28th InternationalConference on Human Factors in Computing Sys-tems (pp. 733–742).

Paulhus, D. L. (1991). Measurement and control ofresponse bias. In J. P. Robinson, P. R. Shaver, &L. S. Wrightsman (Eds.), Measures of personalityand social psychological attitudes (pp. 17–59).New York, NY: Academic Press.

Petty, R. E., & Cacioppo, J. T. (1986). The elaborationlikelihood model of persuasion. Advances in Ex-perimental Social Psychology, 19(1), 123–205.

Ramachandran, D., Canny, J., Das, P. D., & Cutrell, E.(2010, April). Mobile-izing health workers in ruralIndia. In Proceedings of the 28th InternationalConference on Human Factors in Computing Sys-tems (pp. 1889–1898).

Raudenbush, S. W., & Bryk, A. S. (2001). Hierarchi-cal linear models: Applications and data analysismethods (vol. 1). Thousand Oaks, CA: SAGE Pub-lications.

Savani, K., Morris, M. W., & Naidu, N. V. R. (2012).Deference in Indians’ decision making: Introjectedgoals or injunctive norms? Journal of Personalityand Social Psychology, 102(4), 685–699.

Sinha, J. B., Sinha, T. N., Verma, J., & Sinha, R. B. N.(2001). Collectivism coexisting with individualism:An Indian scenario. Asian Journal of Social Psy-chology, 4(2), 133–145.


POWER TO THE PEERS

Sterling, S., & Rangaswamy, N. (2010, December).Constructing informed consent in ICT4D re-search. In Proceedings of the 4th ACM/IEEE Inter-national Conference on Information andCommunication Technologies and Development(pp. 46–55).

Storti, C. (2007). Speaking of India: Bridging thecommunication gap when working with Indians.Boston, MA: Intercultural Press.

Sulaiman, R. V. (2003). Innovations in agriculturalextension in India. Rome, Italy: Sustainable Devel-opment Department, Food and Agriculture Orga-

nization of the United Nations. Retrieved fromhttp://www.fao.org/sd/2003/KN0603a_en.htm

Verma, J., & Triandis, H. C. (1999). The measure-ment of collectivism in India. In InternationalCongress of the International Association forCross-Cultural Psychology. Lisse, The Netherlands:Swets & Zeitlinger Publishers.

White, P. H., & Harkins, S. G. (1994). Race of sourceeffects in the elaboration likelihood model. Jour-nal of Personality and Social Psychology, 67(5),790–807.



Research Article Power to the Peers: Authority of Source ... · evant sources of knowledge...

Documents

Transcript of Research Article Power to the Peers: Authority of Source ... · evant sources of knowledge...