THE IMPACT OF AUDIO PHONE REMINDERS ON KENYA FARMERS' KNOWLEDGE AND UPTAKE OF DROUGHT TOLERANT (DT) MAIZE By Tian Cai A THESIS Submitted to Michigan State University in partial f ulfillment of the requirements for the degree o f Agricultural, Food and Resource Economics Master of Science 201 8 ABSTRACT THE IMPACT OF AUDIO PHONE REMINDERS ON KENYA FARMERS' KNOWLEDGE AND UPTAKE OF DROUGHT TOLERANT (DT) MAIZE By Tian Cai Changing climate, such as drought stress, can lead to har vest failures and losses in Eastern and Southern African (ESA) countries . Scientists have been breeding improved seeds that are resistant to drought . However, the demand for these drought tolerant (DT) maize varieties is still low. Farmers also lack knowle dge about modern practices that can increase their productivities in growing DT maize. This study design ed and test ed two strategies of using information and Th e first o ne is a locally - made The second one is a multichannel method that integrates the same video with timely mobile phone - based audio messages. We conduct ed a field experiment in Machakos and Makueni count ies in south centra l Kenya. The experiment randomly allocated the two strategies to farmers in the study areas. We found that farmers in the multichannel group retained significantly higher knowledge about DT maize and its accompanying management practices than farmers in th e control group . Though the difference is not statistically significant, farmers in the video - only group also had a higher score in knowledge of these practices than those in the control group. Further, farmers in the multichannel group were significantly more likely to be willing to plant DT maize in the next primary growing season. This study contributes to a greater understanding of farmers' learning and uptake of DT maize. The results suggest that ICT strategies that integrate contextualized knowledge a nd tim e ly reminders could help farmers gain knowledge about DT maize and encourage them to try new seed varieties . iii I dedicate this work to my parents and grandparents iv ACKNOWLEDGEMENTS I would like to extend my deepest gratitude to many individuals who in one way or another have contributed to making this study possible. I would like to thank my advisor, Prof. Eduardo Nakasone, and committee members, Prof. Nicole Mason and Prof. Charles Steinfield for their unremitting guidance and encouragements throug hout my program. I would also like to thank the late Prof. Mark Levy who gave me the impetus to pursue the topic of this thesis and the hope that our work can somehow help the poor to improve their livelihood. This thesis would not have succeeded in ha ving this depth of research without the contribution of scientists in International Maize and Wheat Improvement Center and field staffs in Farm Input Promotion Africa. Furthermore, I would like to thank my colleagues in Department of Media and Informatio n and Department of Agriculture, Food and Resource Economics for their suggestions which improved the quality of the thesis. I want to especially acknowledge the United States Agency for International Development who generously provided funding for this study and supported my studies through the Global Center for Food Systems Innovation at Michigan State University. And, finally, I would like to thank farmers who participated in this study for their time and valuable input. v TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ................................ ............................ vi LIST OF FIGURES ................................ ................................ ................................ ................................ ......................... vii CHAPTER 1. INTRODUCTION ................................ ................................ ................................ ................................ .. 1 CHAPTER 2. LITERATURE REVIEW ................................ ................................ ................................ ....................... 5 2.1. Drought tolerant (DT) maize in sub - Saharan Africa (SSA) ................................ ................................ ....................... 5 ................................ ................................ ................................ ........................... 7 2.3. Videos and Phone Messages in Agricultural Extension ................................ ................................ ............................... 8 CHAPTER 3. RESEARCH METHOD ................................ ................................ ................................ ...................... 11 3.1. The Interventions ................................ ................................ ................................ ................................ ....................... 11 3.2. Study Area ................................ ................................ ................................ ................................ ............................... 13 3.3. Experi mental Design ................................ ................................ ................................ ................................ ................. 14 3.4. Baseline Comparisons ................................ ................................ ................................ ................................ ................ 15 3.5. Empirical Estimation Strategy ................................ ................................ ................................ ................................ .. 17 CHAPTER 4. RESULTS ................................ ................................ ................................ ................................ ................ 22 4.1. The Effect of Treatments on Knowledge about DT Maize and Management Practices ................................ ................ 22 4.2. The Effect of Treatment on Willingness to Grow DT maize ................................ ................................ ...................... 24 4.3. Differential Impacts of the Interventions ................................ ................................ ................................ ..................... 25 CHAPTER 5. DISCUSSION AND CONCLUSION ................................ ................................ .............................. 27 APPENDICES ................................ ................................ ................................ ................................ ................................ . 31 MAIZE SEED AND MANAGEMENT PRACTI CES ................................ ................................ ....................... 3 2 APPENDIX 2. FIGURES AND TABLES ................................ ................................ ................................ ............ 34 REFERENCES ................................ ................................ ................................ ................................ ................................ . 42 vi LIST OF TABLES Table 1: Experimental Design ................................ ................................ ................................ ........................ 36 Table 2: Household Characteristics in Baseline by Treatment Groups ................................ ................... 37 Table 3: Household Characteristics in Baseline by Compliance with Treatments ................................ . 38 Table 4: Knowledge Scores Regression by Treatment Groups 1 ................................ ............................... 39 3 ....... 40 Table 6: Differential Impacts of the Treatment on K nowledge about DT maize by NG O Status 1 .. 41 vii LIST OF FIGURES Figure 1. Timeline of the S tudy ................................ ................................ ................................ ..................... 34 Figure 2. The S tudy A rea ................................ ................................ ................................ ................................ 35 1 CHAPTER 1. INTRODUCTION Maize comprises forty - five percent of calories consumed by people in Eastern and Southern Africa (ESA) and is a particularly important income source to the poor in this region 1 (Shiferaw, 2011). st pressing concerns in the region are harvest failures and losses attributable to changing climate conditions , such as drought stress . Such changes affect food production both directly , through changes in agricultural productivity, and indirectly , through reducing agricultural incomes (Fisher et al., 2015). In respon se to changing rain patterns and climate , scientists have been breeding improved seed s , some of which are drought tolerant (DT) 2 . Hendrix & Glaser (2007) suggest that planting these new varieti es , in conjunction with improved management practices can reduce yield loss by up to 40%. Although yields remain low and, as a result of changing climate conditions in ESA, variable, demand for DT maize seed still falls far short of the what it could be (Fi sher et al., 2015). Moreover, while tests conducted in experimental stations suggest sizeable productivity increases from adoption of DT maize , actual profitability of doing so is unknown . The successful diffusion of improved seed in Asia and othe r regions indicate s that technology adoption and impact at scale is a combination of innovative technologies and institutional and policy shifts, such as improvements to and credit market s (Shiferaw, 2011). In addition to posters, radio, television , and newspapers, the existing methods for communicating DT maize to farmers include fiel d demonstrations and field days organized by 1 Maize currently covers 25 million ha in sub - Saharan African , and an estimated 38 million metric tons are produced annually( Smale et al., 2011) 2 The Drought Tolerant Maize for Africa (DTMA) project , led by the International Improved Maize and Wheat Center, released 160 drought tolerant maize varieties between 2007 and 2013 and disseminated them to farmers in 13 African countries. In addition to drought tolerance, the varieties have other attractive traits, such as resistance to major disease s and high protein content. Cost of DT maize seeds is similar to other non - DT commercial varieties. 2 extension officers and NGO experts (Fisher et al., 2015). However, due to a seve re shortage of extension officers and NGO experts, farmers struggle to access relevant and up - to - date information (Lovo, 2013). Scholars believe that information communication technologies (ICTs) can be a potential solution (e.g. Nakasone & Torero, 2014). This study assess es how different ICT - based extension approaches can improve knowledge about DT maize and its complementary practices. In particular, we test two approaches. The first approach includes , s. Th is approach i s utilized by agricultural extension s workers and NGO staff who shoot videos featur ing farmers from the same or a similar community . Our video integrate s both contextualized social and cultural cues and technical information about DT maize and the modern management practices. The video also use s audience s about potential risks of growing DT and its ris k management practices . The second approach is a multichannel method. In addition to the participatory video farmers in this group receive d four audio messages on their mobile phone over a maize grow ing season . An audio system complements information in the video by " push ing " audio message s before the key decision points in the maize growing season . T hese audio messages are reminders t hat reinforce the knowledge farmers learned in the video and encourage adoption . The first message reminded farmers to purchase DT maize seed. The second and the third messages include d content that remind ed farmers of proper tim ing , practice , and quantity used in fertilizer micro - dosing. The last message inform ed farmers about post - harvesting practices about storage and moisture testing that can be used to manage DT maize, traditional varieties , or other improved varieties. Gi ven that improved seed like DT maize is more sensitive to factors like soil fertility, these management practices are more critical to farmers who grow DT maize than to those who grow traditional varieties (Byerlee & Heisey, 1996 ; Foster & Rosenzweig, 201 0) . 3 In total, 615 households in 27 villages were included in the study. W e partnered with Farm Input Promotions Africa (FIPS), a local NGO , to design ICT strategies and to conduct a field experiment on the intervention in Machakos and Makueni Counties, K enya. FIPS has worked in the study area for more than 10 years and operates in some of the villages in the sample. A detailed description of its activities in the study area is presented at the end of method s section. We randomly selected villages from a l ist of villages provided by FIPS. This list includes villages where FIPS is currently working and villages where FIPS planned to work but had no activity prior to and during this study. We assigned these villages to one of three groups. Farmers in the fir st group watched a video, and farmers in the second group watched the same video followed by four audio messages. The third one is a control group, so no video was screened and no audio messages were sent to these villages. We study three research questi ons . First, we study whether the participatory video or the knowledge about DT maize and about the management practices relative to those in the video - only group and to those in the control group (RQ1) . Second, we i nvestigate whet her there was any difference in willingness to try DT maize between farmers in the multichannel group relative to those who only watched the participatory videos and to those who in the control group (RQ2). Moreover, FIPS provide s services r elated to DT maize in some villages sampled in the study. Therefore, this c urrent study aim s to understand the differential treatment impacts on farmers' knowledge and uptake according to whether they live in villages with FIPS services (RQ 3 ) . W e found t hat farmers in the multichannel group demonstrated higher knowledge about both DT maize and its accompanying management practices than farmers in the video - only group 4 and in the control group . They also were more w i lling to plant DT maize in the next prima ry growing season. T hese effects were strongest in communities supported by FIPS . This study has four main contributions. First, i t provides new empirical evidence on the impact of using ICTs on agricultural knowledge provision. We also test new design p rinciples that create a synergy among various ICTs to improve , building on previous literature that mainly discusses the effect of a single ICT (e.g. Gandhi, et al., 2009, Cole & Fernando 2012 ). Third , this stud y suggests the importance of design ing ICT strategies base d ; the characteristics of agricultural innovation s , such as whether farmers need to practice multiple steps over a long cycle ; and functions of various ICTs. Lastly, t he findings contribute to a greater understanding of farmers' learning and uptake of DT maize. The rest of the paper is organized as follow s . The second section presents a literature review s like DT maiz e. The same section discusses existing studies that use participatory videos and mobile phones in agricultural knowledge provision. The third section discusses the intervention, the study area, the experimental design, the sampling strategies and sample ch aracteristics, and the estimation framework . R esults are presented in the fourth section , followed by a discussion of contribution and implication of the research findings . 5 CHAPTER 2. LITERATURE REVIEW 2.1. Drought tolerant (DT) maize in s ub - Saharan Afr ica ( SSA ) Studies have found that farmers use different strategies to reduce the negative impacts of climate variability and change 3 to their maize production . Among these strategies, the adoption of d rought tolerant crops can play an increasingly importa nt role in coping with climate change and variability (Kassie et al., 201 4 ). Research institutes like the International Wheat and Maize Improvement Center (CIMMYT) and seed companies have been researching and introducing Drought Tolerant varieties (DT) 4 to the ESA market for about a decade . 5 Between 2007 and 2013, 160 DT maize varieties were released to ESA market . 6 Field trials in ESA found that DT maize yield were higher than popular commercial varieties by 82 - 127% in controlled drought conditions, 26 - 47 % in random drought conditions, and 25 - 56% in optimal rainfall condition 7 (unpublished data from T s edeke Abate, DTMA project leader, March 2015, cited in Fisher et al., 2015 pp. 284 - 285). La Rovere et al. (2014) found that DT maize could increase average y ield and lower yield variability. The authors also speculate that the yield advantage between DT and local varieties could be greater when there is a drought. 3 These methods include crop rotation, changing planting date s , switching crop species, c rop diversification, and soil and water conservation techniques. 4 S cientists use modern conventional method s to breed DT. C urrent DT varieties in ESA market are not genetically modified. 5 Global efforts aiming to develop drought tolerant maize germplasm include drought - tolerant maize for Africa (DTMA) , which was implemented by CIMMY, the International Institute for Tropical Agriculture (IITA) , and the national research/extension institutions of 13 African countries. 6 This is the total number of DT variet ies released to all ESA countries. Each countr y ha s a smaller number of varieties available to local farmers. 7 Maize under random drought condition s receive s approximately less than 600 mm rainfall per year , and is plagued by pests and disease under rain fed conditions. Maize under optimal rainfall condition s receive more than 750mm per year in a temperature range of 24 - 33 °C. Under managed drought condition s , maize is grown in the off season and the irrigation interval is calculated based on the crop wate r balance (Setimela et al, 2017 ). 6 W e still know little about how farmers adopt DT maize and whether they can profit from growing it , especially because farmers commonly apply insufficient amounts of input s like fertilizer and labor and use inadequate management practices (e.g. Tambo & Abdoulaye, 2011). Scientists have suggested that growing improved varieties like DT maize requires farmers to use more inputs, especially fertilizer, in comparison to traditional varieties (e.g. Smale, Byerlee & Jayne, 2011), so underutilization of these inputs is potentially a problem for DT maize adoption. Fisher et al. (2015) found that labor constra it was because farmers currently only grow DT maize at small scales, and the labor shortage may become a salient problem if they decide to expand DT maize production. Seed markets are not well established in ESA, and farmers have limited access to DT seed s in areas where there is no government intervention or development project directly distributing the seeds to farmers (Fisher et al., 2015). Further, Fisher et al. (2015) found in six ESA countries 8 that farmers who were older, owned less land, and had less exposure to information about DT maize we re less likely to try DT seed. Langyintuo et al. (2010) and Fisher et al. (2015) argue that an effective strategy to increase the adopt ion of improved maize, like DT maize , is to enhance awareness and knowledge about DT maize. However, most efforts of knowledge provision ha ve mainly targeted farmers who already knew about improved seed s prior to the intervention . Thus, more tailo red communication strategies are needed to increase awareness and demand among farmers who are unfamiliar with improved seed . There is also a need to better inform farmers about input requirements and modern cultivation practices associated with improved s eed. Access to extension services is vital to raising farmers the existence, benefits , and usage of the technology ( Kabunga et al. , 8 This six countries are Ethiopia, Tanzania, Uganda, Malawi, Zambia, and Zimbabwe 7 2012). However, extension services in ESA, whether publicly or privately funded, are less efficient in boosti ng adoption, especially where demand for improved seed is low (Muyanga & Jayne, 2008). Seed companies also disseminate information about improved seed through their market network s and other information channels , including seed packs and radio and TV progr ams. However, there is no evaluation of the effectiveness of these efforts . 2.2. Improved Seed Khonje et al. ( 2015 ) found that households who gr ew improved maize saw gains in crop income, consumption expenditure, and food security. However, a substantial portion of maize farmers still plant traditional seed s . In the 2006 - 07 season, only 44% of total maize area in ESA was planted with improved seed 9 (Abate et al., 2017) . An earlier study show ed that around 60% of Kenyan farmers used f ertilizer and hybrid seed in 2004 (Suri, 2011). Further, f armers who adopted improved maize preferred old er improved varieties or jointly planted new and old hybrids in one farm - saved impure seed that farme rs kept from previous production using improved seed . 10 Many farmers switch ed back and forth between traditional and improved seed from season to season (Duflo et al., 2008). There are several reasons behind the low levels of adoption of improved seeds in ESA , includ ing lack of information about improved varieties and their benefits ( e.g. Fisher et al., 2015 ) , lack of knowledge about how to manage new varieties ( e.g. Diagne & Demont, 2007 ), behavioral biases (Duflo et al., 2011) , and lack of credit to purc hase improved varieties and input s that complement the technology . Moreover, supply of quality improved seed varieties, such as drought tolerant maize, are unreliable. S eed varieties that are supplied to farmers are often not suitable to 9 Th ese data exclude South Africa. Improved varieties include open - pollinated varieties (OPVs) and hybrids. 10 Impure improved seed could lo s e some improved attributes, such as high yield. 8 agro - ecol ogical conditions ( personal communication, 2016 ) . Further, i n area s like Machakos and Makuen i, agro - ecological conditions can vary within a village because of the changing elevations . However, f armers receive little information about how to choose among va rious improved varieties because that few sources can provide credible recommendation s based on their needs and conditions . Fisher et al. believe s attributes of improved seeds, such as early maturation or drought tolerance, can reluctance to choose yield - enhancing inputs like improved seeds and intensifi cation cultivation practices (Fisher et al. , 2015) . Other studies provide various arguments to explain varieties. One ar gument claims that farmers are knowledgeable about the technical components of improved seed. However, farmers with a high fixed cost of growing hybrid maize delay their adoption because the return is low (Suri, 2011). Suri (2011) suggests that this high c ost is caused by supply and infrastructure constraints. A third argument emphasized the importance of knowledge about complementary practice s to effectively cultivate improved seeds . I n this case, i ncomplete knowledge about how to most effectively grow DT maize such as how to manage maize crop s to avoid potential production risks can lead to low adoption rates (e.g. Diagne & Demont, 2007) . 2.3. Videos and Phone Messages in Agricultural Extension Information communication technologies (ICT), including ra dio, video s , mobile phone s , and computers are increasingly used in provision of agricultural extension services. For example, agricultural training videos have been created to complement and supplement conventional extension training s . These training video s are lauded for a number of reasons , such as their potential role in improving farmers' knowledge and adoption of agricultural technology (Gandhi et al., 2009; Cai et al., 2015; Nakasone & Torero 2016). One of the most well - known video - based training appr oaches is developed by Digital Green, an Indian NGO (Gandhi et al., 2009). Digital Green 9 trains grassroots development workers to use low - cost digital cameras and laptops to produce videos. A local moderator screens these videos in their communities. Other efforts that investigate video approach in extension training focus on the video production process (e.g. Van Mele, 2006 ), dissemination approach es , such as combining the video screening with farmers field school ( Ongachi, Onwonga, Nyanganga & Okry, 2017 ) , and video content ( e.g. Medhi & Toyama, 2007 ). The video approach was found to be effective as it can reach populations with lower levels of literacy and education , is cost - effective, can induce behavioral change (Zossou et al., 2009) , and help s to bridge gender divides by making information access more equitable across male and female farmers (Bery, 2003; Cai et al., 2015). More recent studies suggest potential for this approach beyond an information transfer tool , such as influenc ing the psychological we ll - being of the impoverished by creating aspirations (Bernard et al, 2015) and forward thinking (Bernard & Taffesse, 2014). However, local communities commonly lack video playing devices, and training videos are normally only shown to a community once. The c omplexity of some agricultural technologies requires intensive training programs that provide relevant information at specific times in order to refresh and remind farmers e right moment . The m obile phone is another ICT that is used to provide agricultural extension services to farmers. Cole & Fernando (2012) found that voic e messages can be effective in help ing farmers learn to safely handle pesticides and to improve yiel d. Text messages (or SMS ) is another widely used function to deliver agricultural extension information due to its simplicity and low cost . Whereas studies investigating impacts of the SMS approach show mixed results, the impact of SMS information on knowl edge learning and adoption vary depending on the context, content, technology , and delivery strategies . Fafchamps and Minten ( 2012) investigated a program that provided crop advisory tips and local weather forecasts to farmers through SMS. They found that 10 SMS had no impact on cultivation practices or harvest losses . The advice provided by the program were Casaburi et al (2014) designed an intervention in Kenya which sent out SMS to sugarcane grower s at individualized moments to remind them to perform certain task. Timing of the messages was based on their harvest cycle and age of their cane . Casaburi et al. (2014) found t he program increased yield by 11.5%. A study conducted in Ecuador (Larochelle e t al. , 2016) tested the impact of SMS as reminders of practices that extension trainer s had taught to farmers in formal trainings earlier in the season? . The study found that post - training SMS rated pest management (IPM) practices. The authors suggest ed that impacts of the messages may have be en associated with their timing and content , as well as y also argue d that the messages had a positive impact because the information provided through SMS did not provide new information to farmers , but acted as a reminder of the contents that had provided by formal trainings. To conclude, r to a certain topic. In our study, using mobile phone messages to remind farmers of the information they learned in the video can be effective to help farmers recall important complementary practices that they need to implement several weeks or months after viewing the video. These management practices include proper ways of planting, fertilizer applicat ion , and fertilizer micro - dosing, and they need to be implemented at key decision making points during the maize growing cycle. 11 CHAPTER 3. RESEARCH METHOD We conduct ed a field experiment in Machakos and Makueni C ount ies , Kenya and collected baseline information of 615 households in 27 villages in the area . We randomly assigned these villages to one of three groups (see Table 1) . Farmers in the first group watched a video that explained the benefits of DT maize and the practice s farmers need to use to cultivate DT maize. The technical content of the video wa s provided by the scientists at CIMMYT who developed g erm plants of these DT varieties . L ocal experts working f or FIPS also offer ed recommendations of modern practices that accompany the growing of DT maize . These practices were tested locally by farmers and were found to be effective. We consulted with experts in both organizations in order to compose re commendations that are suitable for the local agro - ecological and social conditions. Farmers in the second group watched the same video and also received four audio messages before key decision points of a maize growing cycle. The third one was a control group, so no video was screened and no audio message was sent to these villages. In addition to the random variation in the treatment group, we also associated with the impact of treatm ents. The object ive is to investigate whether the ICT strategies are a ssociate d with greater knowledge about DT seed and modern practices , as well as increased willingness to plant DT maize in the next growing season . In this section , we explain the inter vention, study area, experimental design, sampling strategies and sample characteristics , and estimation framework . 3.1. The Intervention s A team of researchers, staff , and the local farmers produced a 30 minute long video . In the video , f armer actors explain the benefits of DT maize, potential risks relating to the 12 type of seed, how to identify and purchase seed, modern cultivation practices , a nd their personal experiences with cultivating DT maize . We expect the video to provide information abo ut DT seed and the modern practices and to motivate farmers to test the seed. Video screenings were moderated by FIPS . In the video - only group and multichannel group, sampled households were invited to attend a training that included the video screenings a nd group discussions. We screened the video once in each treatment village around one month prior to the 2016 first maize growing season. Further, the team created four audio messages with advice and reminders related to seed purchasing and cultivation ac tivities (such as land preparation, planting , and fertilizer application). In the multichannel treatment group, in addition to the same video screening, we sent out four audio messages before the key decision points of a maize growing cycle. The first mess age was sent a few days after the screening, and it encouraged farmers to purchase DT maize from certified sources. The second and third message s emphasized proper quantity for fertilizer micro - dosing and the application practices. These two messages were sent out a week prior to each fertilizer application. The last message was about post - harvesting practices and was sent during harvesting. The control group received no intervention during the study period. Figure 1 provides a calendar of intervention s and data collection activities along with an agricultural calendar. A pre - screening survey was conducted to obtain production information, such as the maize varieties farmers grew in the previous growing season (2015 second season) and other demographic data about the study area. Then a follow - up survey was implemented after the 2016 first growing season (March interventions. The timeline allows us to estimate the impacts of our tre atments over the first growing season in 2016 (March to August). This enables us to observe how the treatments associate 13 the treatment impacts in the 2016 seco nd growing season (primary season). Farmers commonly grow substantial ly more maize in the second growing season , as rainfall is more consistent and reliable than in the first growing season. 3.2. Study A rea The intervention s took pl ace in two counties i n the s outh eastern region of Kenya (Figure 2 ). These two count ies are locate d in the dry transitional 11 a gro - ecological zone and are considered a medium drought - risk zone (20 40% PFS) target area s 12 , so there are potential benefits to DT seed (La Rovere et al . , 2010). Maize is the primary food and cash crop in both counties. The majority of the agricultural activities in these two areas are rain - fed and in the small - scale semi - subsistence sector. The soil generally is of low fertility : 97% of the farms had in sufficient nitrogen and 40% of the farms had insufficient phosphorus (Barber et al. 19 79 ; NAAIAP, 2014 ). In the NAAIAP survey , conducted in 201 4 , in Machakos Sub - County, all the sampled farms had low soil organic matter content, which lead s to low water ho lding capacity and low water infiltration rate. This results in highly erodible soil in the region (NAAIAP, 2014). Poor soil fertility and high erosion has also been found in Makueni County. M aize yield is low compar ed to the national average . The average production of this zone is generally low , ranging from 300 - 1,200kg/ha, while the national average is 1,600kg/ha (Muhammad et al., 2010). Most farmers in this area still grow traditional seed s. However, some of them grow both traditional and improved seed s in the same season . The Duma and Pioneer H ybrid seeds are the most common improved maize variet ies grown in the study areas. Few farmers grew newer improved varieties . F arming households experien ce food shortage s regularly due t o low yield s . Farmers sell l ittle, if any output, as most of the harvest is used for home - 11 This zone includes the lower midland zone (800 - 1300 MASL) and upper midland zone (1300 - 1800 MALS). The two - risk zone (20 - 40% probability of failed season) target area. 12 14 consumption . The higher variation in rainfall is associated with the fluctuation in maize production in this area (Omoyo et al. seed , including drought toleran t seed , is low, while the dis - adoption rates of the improved seed are high (Muhammad et al., 20 10 ). Machakos County has a population of 442,930 and a poverty rate 13 of 57% . T he population in Makueni County is 253, 316, and the poverty rate is 63.8% (Government of Kenya, 2013) . These two counties encompass 368.57 thousand hectares of land and mainly consist of hills and small plateaus rising between 700 and 1700 meters above sea level (MASL). The mean elevation is 1,357 MASL in Machakos and 1,047 MASL in Makueni. The capital towns of both counties are less than 1 00km from Nairobi. The western part of the Machakos area is considered part of the great er Nairobi area. Off - farm employment is prevalent among men, leaving women to take the main responsibilit y of crop management. The two production seasons are from November to February, and from March to August. Total annual precipitation ranges from 500 to 1 , 300 mm , depending on altitude and other factors. The m ajority of farmers produce maize in both seasons , and the primary growing season in this area is associate d with the November rains (Hassan, 1998). Jointly with FIPS, we identified two wards for our study. The first one is Kola, located in the southern part of Machakos County. The second one is Kee, loc ated in the northern part of Makueni County. Kola ward has a population of 24,264; and the population of Kee is 20,926 (Government of Kenya, 2013). 3.3. Experiment al Design The local NGO, FIPS has been operating in Kenya for more than 10 years. They are interested in testing different ICT approaches in agricultural knowledge provision in the villages 13 Defined as households subsisting on under $1.25 per day 15 where they already work and in villages where they want to expand their operations. They also seek to understand adoption of DT seeds . FIPS provided us with a list of 50 villages located at agro - ecological zones suitable to DT seed adoption and whose yields were greatly affected by recently drought , and which thus had potential demand for DT maize . FIPS was already working in 41 of these villages and planned t o cover the remaining 9 villages in the near future , though, in these 9 villages, they had no activities before or during the time of our study . We sampled all 9 villages where they did not operate and randomly selected 18 out of 41 villages where they alr eady operated. Then t he 27 villages in our sampling framework were randomly assigned to one of three treatment conditions: eight villages were assigned to the video - only group, eight villages were assigned to the multichannel group, and 11 villages were as signed to the control group (see Table 1) . FIPS also provided a list of households in each village. We randomly selected 30 households in each non - FIPS village and 17 households in each FIPS village to include in our study . As the non - FIPS villages are big ger than the FIPS villages , a higher number of households were selected in the non - FIPS village . Our final sample included 615 households in the baseline , and we were able to survey 581 households in the endline (see Table1 ). 3.4. Baseline Comparisons I n this section, I show that the randomization process delivered three similar groups: households in the video - only treatment, those in the multichannel treatment, and those in the control group . The baseline characteristics of those in the two treatment gr oups are compared with those in the control group with the following Ordinary Least Squares (OLS) Regression: Y i = 0 + 1 Video i 2 Multichannel i + µ i (1) Y i is a characteristic of the i th household at baseline and µ i is a zero - mean household - specific error 1 2 provide estimates of the differences in Y i of the video - only and th e 16 multichannel group relative to the control group , respectively . 1 and 2 are statistically different to assess any imbalances between the multichannel and video - only groups. I test whether there are any differences between the video - only , multichannel and control groups along several dimensions , including gender of the household head, the number of people in the household, a livestock index (number of different livestock multiplied by their market values), main income source s (agriculture or non - agriculture), total land size, household distance to the nearest market selling seed, main maize grower ' s education level , and whether the household was located in a FIPS villages or in a non - FIPS villages . Sample means of the se var iable for the video - only , the multichannel , and the control group s as well as estimates for Equation (1) are presented in Table 2. The sample is relatively well - balanced in terms of gender of the household head, characteristics of the main maize manage r within a household (age, gender, and education), land size , distance to seed markets and trading center s , number of members within a household, wealth (livestock index) and mobile phone ownership. Three dimensions of the baseline comparisons are statisti cally significant at the 10% level or lower : (a) households in the multichannel group have a higher livestock index than the video - only group; (b) the number of household heads who have no formal education is significantly higher in the video - only group th an it is in the multichannel group though in total, farmers with no education constitute only 6.67% of the total sample ; and (c) the proportion of households in the control group who reside in FIPS villages is significantly smaller than it is for both the video - only group and the multichannel group . This is because two non - FIPS villages which were suppos ed to be assigned to treatment groups were accidently assigned to the control group during the field work implementation (see Table 1) . However, t hese diffe rences are along the lines of what would be expected: out of 36 pairwise comparisons (i.e., three group comparisons along 12 variables), 11% are 17 significant with a p - value of 10% or less. Overall, it does not seem that there are systematic differences bet ween the three groups. 3.5. Empirical Estimation S trateg y maize and about the management practices. We measured knowledge by giving respondents an exam that tested wheth er they retained various details of DT maize and the complementary practices. This knowledge test has two parts: in the first part, I asked respondents to recall the names of the DT varieties they learned from the video . I n the video, seven DT maize variet ies that are suitable to the local conditions were mentioned . Therefore, the maximum knowledge score is seven points if we assume the farmers did not receive new information about name s of varieties from sources other than the video and audio messages. Fa rmers earned one point every time they g a ve correct full name of a DT variety and half of a point for each DT maize name they could partially recall. Two questions were featured in the v ideo. For each correct answer, , they earned one point . Other outcome variables are grow DT maize , which is measured both as a dichotomous variable indicating whether they plan to test DT maize or not and a continuous variable captu ring the proportion of maize land on which they plan to grow DT maize . We asked farmers to report the total size of cultivated and fallow land their households used to cultivate maize . We then asked what maize varieties they planned to grow in the 2016 pri mary planting season , followed by size of the plot s on which they intended to grow each variety. We added up the total size of plots on which they planned to grow DT maize , if any, in the primary season. If farmers plan ned to allocate any of their land to grow DT maize, we . we . 18 the land allocation measure , we divided the quantity of land that farmers planned to allocate to plant DT maize by their total farm size. Throughout the a nalysis, I use the following definitions for treatment variables: Video takes a value of 1 if a household is in a video - only treatment. It takes a value of 0 otherwise. Multichannel takes a value of 1 if a household is in a multichannel treatment. It tak es a value of 0 otherwise. The remaining households (i.e. , those with Video=0 and Multichannel =0) are in control villages. I calculate the impact of the two treatments by comparing househol endline knowledge about DT maize and management practice, an d their uptake of DT . Namely, I estimate the following regression: i = 0 + 1 Video i + 2 Multichannel i + Z i + i (2) w here i are outcomes including (1) level of knowledge about DT seed (2) knowledge about the complementary p ractices retained by household i , ( 3 ) farmers willingness to plant DT maize , and (4) proportion of land farmers were willing to allocate to DT varieties. Further, we hypothesize that farmers in the multichannel group earned more knowledge and were more wi lling to grow DT maize than farmers in video - only group. Z i is a vector of household - level controls, including gender of the household head, household size , a livestock index (number of different livestock multiplied by their market values) , the main incom e sources (agriculture or non - distance to the nearest market selling seed, the main maize grower ' s education level, and whether the households were located in a FIPS village. In this specification, the reference g roup is the control group . The identification of treatment effects relies on the assumption that the error term i is uncorrelated with the variables of interest , Video i and Multichannel i . In our setting, this is plausible 19 because the treatment was randomly assigned . Equation 2 is based on the original treatment assignment ( Video i and Multichannel i ) regardless of whether households actually watched the video or received audio messages and provides an Intention - to - Treat (ITT) estimate of the intervention. Standard errors are clustered at the village level. Due to the small number of clusters in the study, we follo w Cameron et al. (2008) and compute the standard error s using wild cluster bootstrap procedures. Around half of the househo lds in the video - only group attended the video screening s . In the multichannel group , one third of sampled households watched the v ideo and received the audio messages. There are several reasons that can explain the relatively large number of non - compliers. The project only had resources to screen the video once in each village. Therefore, households did not have the opportunity to wa tch the video if they missed the scheduled screening in their village. In the multichannel group, we only sent the phone messages to those who attended the video , mainly because of technical iss ues, such as phones being turned off or having no reception . 14 None of these reasons seem to be there would be any systematic differences between complie rs and non - compliers. We compare the baseline characteristics of the compliers (i.e., those who attended the screening in the video treatment , and those who attended the screening and received the messages in the multichannel group) and non - compliers using Equation (1 ) . There are significant difference s between the compliers and non - compliers in two of these characteristics: (a) compliers are younger than the non - compliers , which suggests that younger farmers were more interested in attending the trainings than older farmers , and (b) the proportion of compliers are higher in the non - FIPS villages than in the 14 20 FIPS villages . W e speculate that farmers in the non - FIPS villages appreciated the trainings more than the farmers in the FIPS villages because they rare ly receive them . However, these differences are along the lines of what would be expected: out of 24 pairwise comparisons (i.e., two group comparisons along 12 variables), 8% are significant with a p - value of 10% or less. Overall, it does not seem that th ere are systematic differences between the compliers and non - compliers (see Table 3 ) . As I do not find major differences between compliers and non - compliers, w e can estimate Local Average Treatment Effects (LATE) using the treatment assignment s as inst ruments for compliance. In particular, I estimate the following system of equations: (3) (3a) (3b) where Watch i and ReceiveWatch i are indicator variables for whether the households are compliers in the video - only group and in the multichannel group, respectively, and is a vector of outcome variables , including knowledge about DT seed, knowledge about the practices and the uptake of DT , and and are the LATE esti mates on those who are compliers in the two assigned treatment groups . We designed the interventions to assess whether communication strategies providing information knowledge and uptake of DT maize . However, we are also interested in investigating if the treatment effects are relatively homogeneous between households who live in the FIPS villages and those who do not. In other words, we intend to test whether the treatments can be a stand - alone approach that 21 increase s farmers knowledge and willingness to grow the DT maize. Farmers in FIPS and non - FIPS villages can be different in diverse ways . For example, FIPS, through their village - based advisors ( VBAs ), provides trainings an d demonstration s , distributes free sample seed , and sells inputs. FIPS trains their VBAs on new maize varieties and complementary practices. Some VBAs manage a small , ( community meetings ) to distrib ute sample seed, including DT maize seed . Moreover, these advisors work as small agro - dealers who supply inputs, such as DT seed , Farmers in the non - FIPS villages do not receive these services. Ther efore, in the FIPS villages, the interventions are not the only information sources relating to DT maize. I test for the presence of FIPS effects within the treatment groups by comparing the outcome variables of households in FIPS villages with those of ho useholds in the non - FIPS villages. We expected that households in FIPS villages would receive services from FIPS in addition to the interventions in the treatment group s . Therefore, farmers in FIPS villages could gain more knowledge about DT maize and be m ore willing to grow it . I estimate the following variation of Equation (4) to calculate the ITT effect on households who resided in FIPS village versus those who did not : Y i j = 1 Video i j + 2 Multichanne l i j 3 Video i j W j + 4 Multichann el i j W j + 5 W j + Z i j + µ i j (4) where W j is NGO status (i.e., live in an FIPS village and lived in a non - FIPS village). 22 CHAPTER 4 . RESULTS In this section , we present the impact of t he about DT maize and the management practices, on the probability of purchasing improved maize seed, and on farmers' willingness to plant the varieties in the next primary growing season. We also present results o f heterogeneous treatment effects between farmers who are with the FIPS services and those who are not. 4.1 . The Effect of Treatments on Knowledge about DT Maize and Management P ractices about D T seed and the management practices are reported in Table 4 . We used two outcome variables to measure knowledge : (1) a continuous variable measuring with a range from 0 to 6 in increments of 0.5 , and (2) a continu the management practices with range from 0 to 11 in increments of 1 . Estimates of the i mpact s of DT seed are in c olumn s 1 - 4 , on their knowledge about proper ways to apply fertilizer are in columns 5 - 8, and on their knowledge of the recommended amount of fertilizer are in columns 9 - 12. Knowledge measurement of these two practices are included because the y were featured in the video and reinforced in the audio message s. Therefore, practices. Columns (1), (2), (5), (6), (9) and (10) in Table 4 show ITT effects of treatments. Columns (3), (4), (7), (8) , (11), (12) show the LATE est imation of treatment effects , which I will discuss below. As shown in column (1), on average, farmers in the multichannel group earned knowledge score of DT maize that was 0.72 points , or 34.5% higher than those in the control group . These 23 differences ar e statistically significant. Farmers in the video - only group earned nearly the same knowledge score as the control group , and the difference is not statistically significant. . A lthough the percentage differences of knowledge about DT maize between treatmen t groups is relatively high, the actual difference is relatively small. This is because farmers in the control group, which we used as the reference to compare with farmers in the treatment group s, retained a substantial ly low er level of knowledge about DT maize. For example, only one third of farmers could recall one or more varieties of DT maize . This fact should be consider ed when interpreting the differences of knowledge about DT maize. Since some farms in the treatment groups did not comply with the i ntervention, we used the random assign ment of the treatment s as instrumental variable s for compliance . The compliance rate in the video treatment is 47.8% , and the compliance rate in the multichannel group is 37%. The results in Table 4 , column (3) suggest that there were sizeable impacts of the multichannel treatment on level of knowledge on DT seed and the complementary practices if the farmers received the treatment. Column (3) shows that farmers who received both the video and four audio messag es earned knowledge scores on DT that were 94.5 % higher than farmers in the control group and 86.7 % higher than farmers in the video - only group . Both difference s are statistically significant. However, t he difference in knowledge on the two management pr actices 15 between the multichannel group and the control group is not statistically significant, regardless of the estimation strategies . Column (5) in Table 4 shows the ITT estimation on the probability of giving correct answers to questions about the meth ods of fertilizer application among those in the multichannel group. We find an ITT effect of 13 percentage points in the multichannel group , and an ITT effect 15 ge score on the proper ways of applying the fertilizer and the recommended amount of fertilizer are binary variables and take a value of 1 if household i gives correct answers to the knowledge test questions and 0 otherwise. I used a linear probability mod el to estimate the effects on these variables. However, the coefficients estimated by the linear probability model are quantitatively similar to those from the Probit model. . 24 of 6 percentage points in the video - only group. Column (9) in the same table shows that the ITT estimation of the impact of the two treatments on household i about the recommended quantity of fertilizer is close to zero. These results indicate that the knowledge about DT maiz e seed compared with the control. Although the video - only approach increase d in the score between farmers in the treatment group and those in the control group is not statistical ly significant. R eminders about the managem ent about the two fertilizer application practices as much as the y did knowledge about DT maize . 4.2 . The Effect of Treatment on Willingness to G row DT maize Far DT maize is modeled as a function of the treatment conditions socio - economic characteristics , which include the same variables as in the regressions of knowledge scores and a new variable that uptake of DT maize in the 2016 first planting season . Due to the project schedule, the endline was c ollected in September 2016. Therefore, I ss to grow DT maize in the next primary season: (1) a dichotomous variable measuring whether farmers plan ned to grow the varieties in the next primary season or not , and (2) a continuous variable ranging from 0 to 1 measuring the proportion of land farmers were willing to allocate to grow DT maize. Regression results from analysis using the dichotomous variable are presented in Table 5 , while results from using the continuous outcome variable are similar and are thus excluded. 25 Column s 1 and 2 in Table 5 s how that the multichannel treatment ha d a positive and significant impact on intention to plant DT maize in the next primary planting season. Farmers in the multichannel group were more likely to plant DT maize than the farmers in the control and in the video - only groups . Th e ITT estimation shows that farmers in the multichannel treatment were 14 percentage points more willing to test DT maize than were those in the control group . This difference is statistically significant at 10% level . The video - only treatment ha d a positive but . It increase d the willingness to plant DT maize by 3 percent age points . The LATE estimation reported by column 3 and 4 in Table 5 suggest s that the multichanne l treatment increase d the willingness o f uptake by 38 percentage points, and the video - only treatment increase d it by 7 percentage points. This finding shows that the multichannel treatment increase s knowledge about DT seed and that it helps to reinforce knowledge about complementary practices and increase s their willingness to try DT seed , relative to farmers in the control group and in the video - only group. e and effectively bring their attention to the new varieties of maize seed. These reminders appear to be positively associated 4.3. D ifferential Impacts of the Intervention s The treatment effects were relatively heterogeneous between households who live in the FIPS villages and those who do not. Table 6 shows th e ITT effects of both treatment s and potential adoption of DT seed . My results show that the impa ct of the intervention was mostly driven by households living in FIPS villages. Households who live in FIPS villages and were assigned to one of the two treatments gained higher knowledge scores on DT maize than did 26 households who received the treatments b ut live in non - FIPS villages. H ouseholds who were assigned to the multichannel treatment group and live in FIPS villages retained a 47.3 % higher knowledge score on DT seed than households in the control group without the services. This difference is statistically significant at 1% level . Households living in FIPS villages that were in the video treatment retained 27.3 % higher knowledge score s on DT than did households in the control group . This difference is not statistically significant. Households in both the video - only group and the multichannel group ha d a higher willingness to plant DT maize in the next primary season if they li ve in FIPS villages than their counterparts who received the same treatments and live in villages with no service from FIPS . Relative to the control group, the multichannel treatment increase d the probability of uptake by 20 percentage points when a household live d in a FIPS - supported village , and t he video - only treatment increase d the willingness to plant DT by 13 percent age points if a household reside d in a FIPS - supported village. Both impacts are statistically significant at 5% level . FIPS distributes free sample DT maize seed to farmers , and VBAs sell seed to communities and provide trainings and demonstration s about D T maize and modern management practices, so households in FIPS villages have better access to DT seed and related information than do households in non - FIPS villages. We speculate these makes the uptake more likely. To conclude, w e found that farmers in t he multichannel group retained more knowledge about DT maize and the management practices, and they were more willing to grow DT maize in the next primary season than the farmers in the video - only group and the control group. Moreover, FIPS , the local NGO , increases availability of seed and provides services that advise farmers about management practices, and these follow - up activities enhance the multichannel treatment effect s. 27 CHAPTER 5. DISCUSSION AND CONCLUSION I examined the effect of participato ry video s and audio message reminders in enhancing about DT maize and comple me ntary practices and their willingness to grow DT maize. This study finds that timely audio messages reinforced the video training content and reminded farmers about the management practices based on the maize growing stages. The multichannel treatment that included both video and audio reminders wa s more effective than the video - only treatment in communicating relatively complex agricultural technologies with mu ltiple steps. The video content provide d local context of the technical messages that enable f armers to understand the benefits of the new varieti es and the management practices. The demonstration in the video also provide d visual guidelines to farmers abo ut steps to plant and manage DT maize. Moreover, the short audio messages that were sent out prior to key decision - making points in the growing season reminded farmers about the management practices at proper times . For example, two audio messages reminded farmers of the need to engage in fertilizer micro - dosing more than a month after the video screenings, a topic the farmers learned about and viewed during the video screening. Hence, one explanation for the heightened effectiveness of the multichannel tre atment is that the audio messages reminded farmers of the video content at appropriate times over a relatively long growing season. Further, although we did not include a group that received audio reminders only, we speculate that the impact of a reminders - only approach would be less than the synergy created by the multichannel strategy tested in this study . If the above interpretation is correct, it has a number of implications for the external validity of these findings. The findings provide insights for the design of ICT strategies aimed at communicating knowledge to farmers about new agricultural technologies and practices . These insights include selection and the arrangement of ICTs and other agricultural extension approach es 28 within a strategy to creat e a synergy. Moreover, how best to design these strategies is based on Farmers commonly need to understand multiple steps and various components in order to adopt new agricultural technologies . Therefore, they need consistent information and coaching to learn and to reinforce the new knowledge. For example, Digital Green, an Indian NGO discussed in the literature review, has developed the participatory video training approach, which has been adopted by vari ous organizations in India to provide information to farmers. The success of this approach depends on a substantial number of skilled frontline extension workers who consistently work with communities to help them learn from the training videos and resolve issues that impede adoption. However, in the majority of Sub Saharan African countries (including our study area), training videos are normally only shown to a community once , due to the lack of video playing devices and proficient personnel to handle the se devices and to moderate screening sessions. Further, d ue to a deficiency of skilled extension workers and other training resources in the majority of Sub Saharan African count r ies, e xtension workers are not able to provide frequent follow - up to enhance farmers' knowledge about a technology after the video screening. The multichannel strategy we tested in this study create s a synergy between various ICT extension approaches to provide farmers contextualized information using the video approach . Without en gaging with extension workers, the audio messages remind farmers about the content of the video before key decision point s in the maize growing season. Therefore, the audio reminders in this study enhanced the video approach based on the local context to c ompensate for the deficiencies of knowledge reinforcement after the video screening. The i mportance of the reminders ha s become more widely recognized by scholars and practitioners because of its impact on the reinforcement of knowledge and capacity for nudging 29 behavior in areas such as agricultural production. In our study, we found that the reminders and increase d nings . This finding corroborates the conclusion by Larochell e et al. (2016) in Northern Ecuador that reminders could reduce inattention, especially when farmers need to make complex decisions. Th e result of th is study suggests that tailored reminders sent out before decision - making points specific behavior are effective, a suggestion similarly made by Cole & Fernando (2012). Although this study area has more than a 90% mobile phone subscription rate, we experienced several challenges in delivering the audio mobile phone reminders to farmers. We delivered the audio messages by making automated phone calls with an audio recordings in the local language. This is different from studies (e.g. , Larochelle et al. 2016 ) that sending SMS to fa rmers because previous research among rural farmers in Kenya had found low mobile phone literacy and limited ability to read SMS among this population (Wyche & Steinfield 2016). Despite our use of audio recordings, however, we still encountered difficulti es in getting messages delivered. The main challenges we faced were network failures and phone numbers that were not functioning. Additionally, in the focus group discussions, some farmers told us that they did not pick up the calls because they were busy and did not expect to receive the messages. We suggest that in any future attempts to use this approach, hints should be sent to farmers prior to sending out the messages to , where farmers can trigger a system to send the messages based on their demand. Lastly, although we found some farmers were able to recall the content in the audio messages, it was impossible for them to revisit these aud io messages when they wished. Therefore, it would be useful to provide a means of allowing farmers to re - listen to the messages or to provide information in a reviewable form such as in a handout containing the information in the audio messages. 30 A numbe r of other l imitations should be kept in mind when interpreting our results and actual uptake of DT maize and the management practices in their primary gro wing season. Future studies should plan a longer - term assessment of this multichannel approach to understand its impact treatment which only sends out audio r eminders without videos. Future studies can incorporate this uptake of agricultural innovation in this multichannel approach. More studies should focus on inv estigating what types of reminders are more likely to work and when to send them to farmers, and how to design different multichannel approach es that can match the characteristics of various agricultural innovations, as well as capacity. 31 APPENDICES 32 APPENDIX 1. IMPROVED MAIZE SEED AND MANAGEMENT PRACTICES How many Drought Tolerant varieties are you aware of? Ni mithemba yiana ya mbeu wisi ila itonya ku miisya sua? _____varieties (if 1. Please name all of the Drought Tolerant varieties that you know ( Kwa ndaia ndavya mtihemba yonthe ya mbeu ila wisi itonya kumiisya sua ___. ( Enumerator : Write down all the names the farmers can remember, remember to includ e all numbers and letters) IF A15=1 OR A22=1 to ask you how much you know about ways of growing maize showed in the video and the audio messages IF A15=2 AND A22=2 [E numerator: this is the instruction for the farmers who did not watch video] experters from NGOs. You can also just tell us good ways you know about growing maize 2. How many seeds to plant per planting hole? Ni mbeke syiana sya kuvanda kwa kila yiima? ________seed(s) 3. What is the distance between each hole within a row? Utaaniu wa yiima na yingi mustalini wiana ata? ______ (feet) 4. What is the distance be tween each row? Utaaniu wa mustalli na ungi wiana ata?______ (unit_____ feet) 5. How many times fertilizer is applied during a growing season? Ni mavinda meana vatalisa wikiawa ivindani yila liu ukwiana? ______times (if farmers think no need to apply, write 6. What is the method to apply fertilizer during planting ? IS IT READ OPTION Mwikiile wa atalisa ni wiva? [enumerator: read options] a. mix the fertiliser and the seed together ( ikia vatalisa mbekeni) (yes/ no) b. Separate the seeds from the fer tliser when applying fertilizer ( ikia vatalisa utee wa mbeke) (yes/ no) 7. Do you need to cover the fertilizer you applied with soil? Niwailite kuvwika vatalisa ula weekia na muthanga? a. yes ( yii) b. no ( Aiee) 8. How much fertilizer is applied? Ni vatal isa wiana ata wikiawa? (visual aid4) a. Less than half a bottle top b. half a bottle top c. One a bottle top d. Two bottle tops e. Two or more bottle top 33 9. When is the time to do the first fertilizer application? Ni ivinda yiv a yailite kwikia vatalisa yambee? a. during planting ( yila unuuvanda) b. when maize is germinated ( yila mbemba yamea) c. 1 - 2 weeks after maize is germinated ( katikati wa kyumwa 1 - 2 kana itna wa mbemba kumea) d. 3 - 4 weeks after maize is germinated( katikati wa kyumwa kya 3 - 4 itina wa mbemba kumea) e. more than a month after maize is germinated 10. When is the time to do the second fertilizer application? Ni indii yila vailite kwikia vatalisa yakeli? a. during planting( yila unuuvanda b. when maize is germinated( yila mbemba yamea c. 1 - 2 weeks after maize is germinated ( katikati wa kyumwa kya katatu na kya kana itina wa mbemba kumea) d. 3 - 4 weeks after maize is germinated (katikati wa kyumwa kya thanthatu na muonza itina wa mbemba kume a) e. more than a month after maize is germinated 11. How many times crops are weeded? Liu uimiawa keana? _______times 34 APPENDIX 2. FIGURES AND TABLES Figure 1 . Timeline of the Study 35 Figure 2 . The Study A rea 36 Table 1 : Experimental Design Treatments Number of villages (original assignment) Number of villages (final assignment) Number of households at the baseline Number of household in the post test FIPS Video only 6 6 104 97 Multichannel 6 6 102 99 Control 6 6 100 100 Non - FIPS Video only 3 2 68 64 Multichannel 3 2 67 63 Control 3 5 169 159 37 Table 2 : Household Characteristics in Baseline by Treatment Groups Overall Mean Co ntrol Video Multichann el Mean Differences (C) (V) (M) (C) - (V) (C) - (M) (V) - (M) Age of household head 49.9 48.464 51.017 50.686 - 2.554 - 2.223 0.331 (15.97) (0.91) (1.31) (1.29) (1.55) (1.54) (1.83) Number of household members ages 15 - 65 3.11 3.117 3.1 74 3.036 - 0.058 0.081 0.139 (1.7) (0.10) (0.13) (0.13) (0.17) (0.16) (0.19) Number of household members 5.481 5.595 5.401 5.379 0.194 0.216 0.022 (2.64) (0.16) (0.21) (0.20) (0.26) (0.25) (0.29) Household's distance to nearest market selling seed (km ?) 6.791 9.02 8.1 7.56 0.92 1.46 0.54 (8.00) (0.67) (0.77) (0.62) (1.04) (0.98) (0.99) Household's distance to nearest trading center (km?) 6.783 6.34 6.827 7.486 - 0.488 - 1.146 - 0.658 (8.05) (0.43) (0.58) (0.75) (0.71) (0.80) (0.94) Landholdings (acr es) 3.27 3.769 3.619 4.195 0.151 - 0.425 - 0.576 (2.67) (0.27) (0.30) (0.45) (0.41) (0.49) (0.54) Livestock index 177.78 176.394 157.483 198.852 18.912 - 22.458 - 41.37 * (170.04) (10.02) (12.54) (13.77) (16.09) (16.73) (18.62) Female headed household (1= yes) 0.236 0.212 0.297 0.201 - 0.085 0.01 0.095 - (0.03) (0.04) (0.03) (0.04) (0.04) (0.05) Own mobile phone (1=yes) 0.929 0.964 0.936 0.964 0.027 - 0.001 - 0.028 - (0.01) (0.02) (0.01) (0.02) (0.02) (0.02) Household head with no formal education (1=yes ) 0.067 0.055 0.116 0.036 - 0.062 ** 0.019 0.081 ** - (0.01) (0.03) (0.01) (0.03) (0.02) (0.03) Household head with secondary education or higher (1=yes) 0.020 0.168 0.174 0.136 - 0.007 0.032 0.038 - (0.02) (0.03) (0.03) (0.04) (0.04) (0.04) Household in FIPS village (1=yes) 0.506 0.386 0.64 0.515 - 0.254 *** - 0.129 *** 0.125 - (0.03) (0.04) (0.04) (0.05) (0.05) (0.05) N 615 274 172 169 446 443 341 Note 1. For the first three columns, the means and standard deviations of each variable in the control, vid eo - only and multichannel groups are reported. In the last three columns, the differences were calculated using the following regression: Y i 1 Video i 2 Multichannle i + µ i . . Standard errors clustered at the village level (right?) and in parentheses below. Significance levels of the differences between the two treatment groups and control group denoted by *** 99%, ** 95%, * 90%. 38 Tabl e 3 : Household Characteristics in Baseline by Compliance with Treatments Non - Complier Complier Mean Difference Age of household head 51.737 49.041 2.696 * (1.31) (0.75) (1.42) Number of household members ages 15 - 65 3.07 3.139 - 0.069 (0.13) (0.08) (0.15) Number of household members 5.301 5.59 - 0.289 (0.20) (0.13) (0.24) Household's distance to nearest market selling seed (km?) 7.27 6.589 0.682 (0.66) (0.37) (0.71) Household's distance to trading center (km?) 7.269 6.57 0.699 (0.66) (0.38) (0.71) Landholdings (acres) 4.189 3.706 0.483 (0.43) (0.21) (0.43) Livestock index 181.43 176.056 5.374 (12.56) (8.54) (15.13) Female headed household (1=yes) 0.247 0.23 0.017 (0.03) (0.02) (0.04) Own mobile phone (1=yes) 0 .941 0.965 - 0.024 (0.02) (0.01) (0.02) Household head with no formal education (1=yes) 0.081 0.058 0.022 (0.02) (0.01) (0.02) Household with secondary education or higher (1=yes) 0.129 0.175 - 0.046 (0.03) (0.02) (0.03) Household in FIPS village (1 =yes) 0.602 0.466 0.136 * (0.04) (0.03) (0.04) N 186 395 581 Note: The second column reports the means and standard deviations of each variable among farmers who did not received the treatments in the treatment villages and those who received the treatm ent in the control villages (Non - Compliers). The third column reports who received the treatments in the treatment villages and who did not receive the treatment in the control (compliers), . In the last column, the differences were calculated using the fo llowing regression: Y i 1 Complier i + µ i . Standard errors clustered at the village level and in parentheses below. Significance levels of the differences between the two treatment groups and control group denoted by *** 99%, ** 95%, * 90%. 39 Table 4 : Knowledge Scores Regression by Treatment Groups 1 Knowledge on DT 4 Knowledge on fertilizer application 4 Knowledge on fertilizer quantity 4 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Video only (1) 0.02 0.02 0.06 0.0 5 0.004 0.01 (0.09) (0.09) (0.13) (0.18) (0.003) (0.02) Multichannel (2) 0.19 0.14 0.13 0.12 0.02 0.02 (0.10) * (0.09) (0.17) (0.28) (0.04) (0.07) Video Complier 0.05 0.02 0.12 0.11 0.01 0.03 (0.21) (0.16) (0.12) (0. 11) (0.08) (0.07) Multichannel Complier 0.52 0.40 0.36 0.32 0.06 0.06 (0.25) ** (0.24) * (0.25) (0.25) (0.16) (0.15) Additional baseline controls 2 N Y N Y N Y N Y N Y N Y Observations 582 582 582 582 582 582 582 582 582 582 582 582 Cont rol Mean 0.55 0.55 0.55 0.55 0.35 0.35 0.35 0.35 0.22 0.22 0.22 0.22 Share of compliers in the video - only group 3 0.48 0.48 0.48 0.48 0.48 0.48 (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) Share of compliers in the multichannel group 3 0.37 0.37 0.37 0.37 0.37 0.37 (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) P - values for Video only = Multichannel 0.17 0.28 0.13 0.14 0.30 0.39 0.36 0.39 0.93 0.91 0.73 0.82 Note 1. Columns 1, 2, 5, 6, 9 & 10 report the ITT estimate of the intervent ions. Columns 3, 4, 7, 8, 11 & 12 report the Average Treatment Effect or Local Average Treatment Effect (LATE) estimates using the treatment assignment as an instrument for the compliers in each treatment conditi on. 2. Additional baseline controls in col umns 2, 4, 6, 8, 10 & 12 include gender of household head, number of people in the household, livestock index, main income sources (agriculture or non - agriculture), total land size, household distance to the nearest market selling seed main maize, grower's education level, and whether the household was located in a FIPS village 3. Indicator variable for whether the households in the two treatment conditions participated in the interventions. For the mult ichannel condition, households both watched the video and received the phone messages. The results are the first stage of the Instrumental Variable Regression (following equation 3) 4. The range of the knowledge on DT is 0 - 6 in increments of 0.5. The knowledge measurement of two practices is binomial and tak e the value of 1 if household i gives correct answers to the knowledge test questions and 0 otherwise. 5. The standard errors of the ITT coefficients are estimated by using the Wild Cluster Bootstrap (Cameron et al., 2008). Standard errors clustered at the village level and in parentheses below. Significance levels of the differences between the two treatment conditions and contr ol conditions denoted by *** 99%, ** 95%, * 90%. 40 Table 5 M aize in the Next Season by Treatment Groups 3 (1) 2 (2) (3) 2 (4) Video (1) 0.03 - 0.01 (0.06) 4 (0.04) Multichannel (2) 0.14 0.10 (0.08) ** (0.05) * Video Complier 0.07 - 0.02 (0.14) (0.09) Multichannel Complier 0.38 0.28 (0.18) ** (0.13) ** Sha re of compliers in the video group 5 0.48 0.48 (0.02) (0.02) Share of compliers in the multichannel group 5 0.37 0.37 (0.04) (0.04) Additional baseline controls 1 N Y N Y Observations 582 582 582 582 Control Mean 0.37 0.37 0.37 0.37 P - val ues for Video only = Multichannel 0.140 0.061 0.107 0.017 Note 1. Additional baseline controls in columns 2 & 4 include gender of household head, number of people in the household, livestock index, main income sources (agriculture or non - agriculture), tot al land size, household distance to 2. Columns 1 & 2 report the ITT estimate of the interventions. Columns 3 & 4 report the Average Treatment Effect or Local Average Treatment Effects (LATE) estimates using the treatment assignment as an instrument for the compliers in each treatment condition. 3 - linear models such as random effects Probits yield similar coefficient magnitudes (not reported). 4 . The standard errors of the ITT coefficients are estimated by using the Wild Clus ter Bootstrap (Cameron et al., 2008). 5 . Indicator variable for whether households in the two treatment conditions participated in the interventions. For the multichannel condition, households both watched the video and received the phone messages. The res ults are the first stage of the Instrumental Variable Regression (following equation 3) Standard errors clustered at the village level and in parentheses below. Significance levels of the differences between the two treatment conditions and control cond itions denoted by *** 99%, ** 95%, * 90%. 41 Table 6 : Differential Impacts of the Treatment on K nowledge about DT maize by NGO Status 1 Knowledge about DT 2, 3 Willingness to try 3 (1) 4 (2) (3) (4) (5) 4 (6) 1 ) - 0.1 7 - 0.13 - 0.12 - 0.11 (0.22) (0.19) (0.05) * (0.06) * 2 ) 0.09 0.06 0.05 0.02 (0.17) (0.12) (0.06) (0.05) 3 ) 0.32 0.27 0.24 0.23 (0.23) (0.20) (0.08) *** (0.09) ** 4 ) 0 .16 0.16 0.15 0.17 (0.24) (0.21) (0.11) (0.11) * 5 ) - 0.30 - 0.24 (0.20) (0.19) 6 ) 0.34 0.23 (0.69) (0.55) Video Complier x FIPS ( 7 ) 0.65 0.56 (0.33) * (0.29) * Multichannel Comp 8 ) 0.25 0.30 (0.73) (0.60) Combination of coefficients 1 3 0.15 0.16 0.13 0.12 (0.12) (0.10) (0.06) ** (0.07) * 2 4 0.26 0.25 0.20 0.19 (0.08) *** (0.90) *** (0.09) ** (0.09) ** 5 7 5 0.35 0.32 (0.24) (0.23) 6 8 5 0.59 0.52 (0.22) *** (0.22) *** Additional baseline controls N Y N Y N Y Observations 582 582 582 582 582 582 Control Mean 0.55 0.55 0.55 0.55 0.37 0.37 Note 1. Additional baseline controls in columns 2, 4 & 6 includ e gender of household head, number of people in household, livestock index, main income sources (agriculture or non - agriculture), total land size, household distance to the nearest market selling seed and main maize grower's education level. 2. The range of the knowledge about DT maize is 0 - 6 in increments of 0.5. 3 . The coefficients are estimated in the regression Y i = 1 Video i + 2 Multichannle i 3 Video i W i + 4 Multichannle i W i Z i + µ i where W i is a binary variable that indicates whether household is in a FIPS or non - FIPS village. We use the treatment assignments and their interaction with W i as instruments. 4 . Columns 1 - 4 present the impact of the treatments on knowledge about DT: 1 & 2 use the ITT estimation and 3 & 4 use the LATE estimation. Columns 5 & 6 present the impact of the treatments on willingness to grow DT. 5. Indicator variable for households wh o received the interventions. For the multichannel condition, households both watched the video and received the phone messages. The results are the first stage of the Instrumental Variable Regression (following equation 3). Standard errors clustered at t he village level and in parentheses below. Significance levels of the differences between the two treatment conditions and control conditions denoted by *** 99%, ** 95%, * 90%. 42 REFERENCES 43 REFERENCES Abate, T., Fisher, M., Abdoulaye, T., Kassie, G. T., Lu nduka, R., Marenya, P., & Asnake, W. (2017). Characteristics of maize cultivars in Africa: How modern are they and how many do smallholder farmers grow? Agriculture & Food Security , 6 (1), 30. http://d oi.org/10.1186/s40066 - 017 - 0108 - 6 Barber, R.G. Moore, T.R. Thomas, D. B. (1979). The erodibility of two soils from Kenya. Journal of Soil Science , 30 , 579 591. Bernard, T., & Taffesse, A. S. (2014). Aspirations: An approach to measurement with validation using ethiopian data. Journal of African Economies , 23 (0), 189 224. doi:10.1093/jae/ejt030 Bernard, T., Dercon, S., Orkin, K., & Seyoum Taffesse, A. (2015). Will Video Kill the Radio Star? Assessing the Potential of Targeted Exposure to Role Models throu gh Video: Table 1: The World Bank Economic Review , 29 , S226 S237. http://doi.org/10.1093/wber/lhv014 Bery, R. (2003). Participatory video that empowers. In S. White (Ed.), Participatory Video . Ithaca, New York: Cornell University.Byerlee, D., & Heisey, P. W. (1996). Past and potential impacts of maize research in sub - Saharan Africa: a critical assessment. Food Policy , 21 (3), 255 277. doi:10.1016/0306 - 9192(95)00076 - 3 Cai, T., Chiwasa, H., Steinfield, C. & Wyche, S. (2015). Participatory videos for smallholder farmer training in Malawi: an analysis of knowledge gain and uptake. In Seventh International Conference on Information and Communication Technologies and Development . Singapore. Casaburi, L., Kremer, M., Mullainathan, S., & Ramrattan, R. (2014). Harnessing ICT to Increase Agricultural Production: Evidence From Kenya . Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008). Bootstrap - Based Improvements for Inference with Clustered Errors. The Review o f Economics and Statistics , 90 (3), 414 - 427. Cole, S. A., & Fernando, A. N. (2012). The Value of Advice: Evidence from Mobile Phone - Based Agricultural Extension. SSRN Electronic Journal . http://doi.org/10. 2139/ssrn.2179008 Diagne, A., & Demont, M. (2007). Taking a new look at empirical models of adoption: average treatment effect estimation of adoption rates and their determinants. The Africa Rice Center (AfricaRice). Agricultural Economics, 37, 201 210. doi:10.1111/j.1574 - 0862.2007.00266.x Duflo, E., Kremer, M., & Robinson, J.(2008). How high are rates of return to fertilizer? Evidence from field experiments in Kenya. American Economic Review , 98 , 482 488. doi:10.1257/aer.98.2.482 Fafchamps, M., & Minte n, B. (2012). Impact of SMS - based agricultural information on Indian farmers. The World Bank Economic Review , 26 (3), 383 414. 44 Fisher, M., Abate, T., Lunduka, R. W., Asnake, W., Alemayehu, Y., & Madulu, R. B. (2015). Drought tolerant maize for farmer adapt ation to drought in sub - Saharan Africa: Determinants of adoption in eastern and southern Africa. Climatic Change , 133 , 283 299. doi:10.1007/s10584 - 015 - 1459 - 2 modern maize in Southern Malawi. Experimental Agriculture , 50 , 533 548. doi:10.1017/S0014479714000027 Foster, A. D., & Rosenzweig, M. R. (2010). Microeconomics of Technology Adoption. Annual Review of Economics , 2 , 395 424. doi:10.1146/annurev.economics.1 02308.124433 Gandhi, R., Veeraraghavan, R., Toyama, K., & Ramprasad, V. (2009). Digital Green : Participatory Video and Mediated Instruction for Agricultural. In Information Technologies and International Development (Vol. 5, pp. 1 15). Government of Ken ya. (2013). Makueni County . Data Sheets Kenya . Hassan, R. M. (1998). Maize Technology Development and Transfer: A GIS Application for Research Planning in Kenya . Hendrix, C. S., & Glaser, S. M. (2007). Trends and triggers: Climate, climate change and civ il conflict in Sub - Saharan Africa. Political Geography , 26 (6), 695 715. http://doi.org/10.1016/j.polgeo.2007.06.006 Kabunga, N. S., Dubois, T., & Qaim, M. (2012). Heterogeneous information exposu re and technology adoption: The case of tissue culture bananas in Kenya. Agricultural Economics , 43 , 473 486. doi:10.1111/j.1574 - 0862.2012.00597.x Kassie, M., Jaleta, M., & Mattei, A. (2014). Evaluating the impact of improved maize varieties on food secur ity in rural Tanzania: Evidence from a continuous treatment approach. Food Security , 6 , 217 230. doi:10.1007/s12571 - 014 - 0332 - x Khonje, M., Manda, J., Alene, A. D., & Kassie, M. (2015). Analysis of adoption and impacts of improved maize varieties in Easter n Zambia. World Development , 66 , 695 706. doi:10.1016/j.worlddev.2014.09.008 La Rovere, R., Abdoulaye, T., Kostandini, G., Guo, Z., Wilfred, M., MacRobert, J., & Dixon, J. (2014). Economic, production, and poverty impacts of investing in maize tolerant to drought in Africa: An ex - ante assessment. The Journal of Developing Areas , 48 (1), 199 225. Langyintuo, A. S., Mwangi, W., Diallo, A. O., MacRobert, J., Dixon, J., & Bänziger, M. (2010). Challenges of the maize seed industry in eastern and southern Africa : A compelling case for private - public intervention to promote growth. Food Policy , 35 (4), 323 331. doi:10.1016/j.foodpol.2010.01.005 45 Larochelle, C., Alwang, J., & Travis, E. (2016). Did you really get the message ? Using text reminders to stimulate adopt ion of agricultural technologies . Lovo, S. (2013). Tenure insecurity and investment in soil conservation. Evidence from Malawi. Clinate Change and Environment , 1 25. Medhi, I.; Toyama, K.; "Full - context videos for first - time, non - literate PC users," Infor mation and Communication Technologies and Development, 2007. ICTD 2007. International Conference on , vol., no., pp.1 - 9, 15 - 16 Dec. 2007 doi: 10.1109/ICTD.2007.4937400 URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4937400&isnumber=4937385 Muhammad, L., Mwabu, D., Mulwa, R., Mwangi, W., Langyintuo, A., & Rovere, R. La. (2010). Characterization of Maize Producing Households in Machakos and Makueni Districts in Kenya . Muyanga, M., & Jayne, T. S. (2008). Private agricultural extension system in Kenya: Practice and policy lessons. The Journal of Agricultural Education and Extension , 14 (May), 111 124. doi:1 0.1080/13892240802019063 Nakasone, E., & Torero, M. (2014). Soap Operas for Female Micro Entrepreneur Training (pp. 1 49). the arid and semi - arid lands of lower eastern Kenya. Agriculture Food Security , 41 , 1 13. Ongachi, W., Onwonga, R., Nyanganga, H., & Okry, F. (2017). Comparative analysis of video mediated learning and farmer field school approach on adoption of striga control technologies in Western Ke nya. International Journal of Agricultural Extension , 5 (November 2013), 1 10. Setimela, P. S., Magorokosho, C., Lunduka, R., Gasura, E., Makumbi, D., Tarekegne, A., Cairns, J. E., Ndhlela, T., Erenstein, O., & Mwangi, W. (2017). On - farm yield gains with s tress - tolerant maize in eastern and Southern Africa. Agronomy Journal , 109 (2), 406 417. http://doi.org/10.2134/agronj2015.0540 Shiferaw, B., Prasanna, B. M., Hellin, J., & Bannziger, M. (2011). Crops that feed the world 6. Past successes and future challenges to the role played by maize in global food security. Food Security , 3 , 307 327. doi:10.1007/s12571 - 011 - 0140 - 5 Smale, M, Byerlee, D. and Jayne, T. (2011). Maize Revolutions in Sub - Saharan Africa. The World Bank: Washington DC. doi:10.1007/978 - 94 - 007 - 5760 - 8_8 Suri, T. (2011). Selection and Comparative Advantage in Technology Adoption. Econometrica , 79 (1), 159 209. doi:10.3982/ECTA7749 eptions of and adaptations to climate change in the Nigerian savanna. Regional Environmental Change , 13 , 375 388. doi:10.1007/s10113 - 012 - 0351 - 0 Van Mele, P. (2006). Zooming - in zooming - out : a novel method to scale up local innovations and sustainable tech nologies. International Journal of Agricultural Sustainability , 4 (2), 131 142.