A New Approach to Monitoring Farmer Prices: Method and an Application to Malawi

Abstract This paper proposes a new approach to monitoring farmer prices in low-income developing countries. This crowdsourcing approach involves broadcasting radio jingles inviting farmers to report the prices and locations at which they sold their crops to a toll-free call centre, with weekly prizes to incentivize reporting. An application to Malawi illustrates the feasibility of this approach in a setting where internet connectivity is limited but mobile phone coverage is reasonable. The majority of farmers reporting sold to assemblers or small traders and received substantially less than official minimum farm gate prices. Non-parametric analysis shows that farmer prices vary according to bargaining power and sales volume but not by distance to the point of sale. These findings may be explained by the fragmented and monopsonistic nature of food markets in Malawi, and farmers’ mode of transport to the point of sale.


Introduction
It is common for farmers, civil society organisations, government officials and journalists to accuse traders and parastatal marketing organisations of monopsonistic behaviour and 'price manipulation' in the post-harvest season (Baulch & Ochieng, 2020;Dillon & Dambro, 2017;Holtzman, 1989).Yet the evidence on which such accusations are made is largely anecdotal, as government efforts to collect and report farm gate prices are often patchy and inconsistent. 1 These weaknesses in the available data on farmers prices are widely documented within the extant development and agricultural economics literature (Aker, 2011;Zanello, Srinivasan, & Shankar, 2014) where there have been renewed efforts at exploring information and communications technology (ICT) as a means of reducing the informational asymmetries within the agricultural sector of low-income nations (van Campenhout, 2022).Such weaknesses in the currently available data result in a paucity of empirical studies on the factors influencing the prices that farmers are paid within low-income African nations.

The fundamentals of crowdsourcing
The term 'crowdsourcing' was first used in 2005 by Howe and Robinson, editors of Wired magazine, to describe how businesses were using the Internet to 'outsource work to the crowd'.Howe (2006) provided a definition for the term crowdsourcing in a follow-up article, 'The Rise of Crowdsourcing', in June 2006, which stated that crowdsourcing represents the act of a company or institution taking a function once performed by employees and outsourcing it to an undefined (and generally large) network of people in the form of an open call.
Although the earliest examples of crowdsourcing date back to ancient Greece and the Tang dynasty in China, the take-off of crowdsourcing was linked to the development of mobile phones and the internet in the late 20th and early 21st centuries (Brabham, 2013).As the wellknown examples of Pebble, Uber and Waze show, crowdsourced tasks are directly outsourced to individuals, who are not required to be employers or professionals in order to perform the outsourced work.In some, but not all cases, monetary rewards are offered to motivate individuals to supply the services or information that are outsourced but this is not always the case (Zeug et al., 2017).Prizes in competitions or the provision of airtime or text messages with reciprocal information are also frequently used as inducements in crowdsourcing exercises (Estelles-Arolas & Gonz alez-Ladr on- de-Guevara, 2012).
It is important to recognize that 'crowdsourcing' is a portmanteau term, which includes a variety of purpose and functions.It is possible to disentangle seventeen different types of crowdsourcing applications including crowd coding, crowd creating of content, crowd funding, crowd identification of pests and diseases (in agriculture), crowdsourcing of health care providers, crowd solving of problems, crowd shipping and even crowd voting (Brabham, 2013).For further information on the different activities comprising crowdsourcing, how it can be best implemented, plus its advantages and drawbacks see Brabham (2013) and Neto and Santos (2018).
In the current application to farmer prices in Malawi, paid price collectors with notebooks or tablets are replaced by farmers with mobile phones, who are motivated to supply information on the prices they are paid for their crops by the opportunity to win agricultural input vouchers in a weekly raffle.Internet connectivity in Malawi is currently too limited (and expensive) for internet based crowdsourcing methods to be feasible in rural areas of the country.The Economist Intelligence Unit's latest Inclusive Internet Index ranks Malawi as 114th out of 120 countries, with an affordability ranking of 116 and an accessibility ranking of 113 .(EIU, 2021).However, mobile phone ownership is common with 43.6 per cent of rural households and 82.4 per cent of urban households owning at least one mobile phone according to the 2019-2020 Integrated Household Survey (National Statistical Office, 2020).

Farmer prices in low-income African nations
In the immediate post-independence decades, many African countries pursued policies that undermined their rural economies by depressing prices for farmers in order to provide cheap food to politically more demanding urban consumers in addition to financial surpluses from exports of 'cash' crops (Bates, 2014;Caracciolo, Cembalo, Lombardi, & Thompson, 2013).This was often coupled with the maintenance of overvalued exchange rates, which further depressed the prices paid to farmers, and restrictive tariffs intended to promote agricultural processing and value-added (Kherallah, Delgado, Gabre-Madhin, Minot, & Johnson, 2002;Tsakok, 2019).While many African countries abandoned such policies in favour of freer agricultural markets in the 1980s and 1990s, elements of the old marketing arrangements such as parastatal marketing organisations still operate in many African countries to this day.Because of their high costs and inefficiencies, such parastatals often account for a significant share of government expenditures, particular when the government also try to regulate the prices paid to farmers (Kherallah et al., 2002;Timmer, 1986).
As noted above, it is common for traders and marketing parastatals in African countries to be accused of exploitative behaviour and price gouging, particular in the immediate post-harvest season when quick sales to meet farmers' immediate cash needs are the most common (Darko, Palacios-Lopez, Kilic, & Ricker-Gilbert, 2018).Most of the academic literature that has examined this issue quantitatively has concluded that traders operate in relatively competitive markets and that, after accounting for costs, their margins are relatively modest (inter alia Abbott (1990); Barrett (2008); Holtzman (1989); Timmer (1984Timmer ( , 1986))).However, there are also well-documented cases in which monopsonistic power has been shown to depress the prices farmers receive (Crow, 1989;Graubner, Salhofer, & Tribl, 2021;Harriss-White, 1996;Saitone, Sexton, & Sexton, 2008).
A common response by developing country government to such accusations, whether grounded or not, is to implement minimum crop or farm gate prices, often based loosely on the cost of production of the crop being sold.Often parastatal marketing boards are asked to support the prices that farmers receive for their food crops by buying at pre-determined 'above market' prices early on in the marketing season, only to sell at 'below market' prices to poor consumers later in in the agricultural seasons (Timmer, 1986).Of course, such buy high, sell low strategies are rather costly and parastatal agencies rarely have enough funding and storage capacity to buy enough to stabilise food prices effectively (Timmer, 1986).So, they resort to exhortation and enforcement to try to maintain MFGP.In the worst cases, mistimed procurement and sales by parastatals, along with trade policy interventions, may even destabilise prices for farmers and consumers (Kherallah et al., 2002;Timmer, 1986).

Data collection and empirical estimation
The following sections provide an explication of our crowdsourcing approach and the subsequent analysis of the collected dataset.Section 4.1.details the processes involved with the crowdsourcing data-collection highlighting our on the ground, collaborative work with IFPRI from initiating the pilot program to the running of the actual task.Our application of crowdsourcing is very much in line with the recent methodological literature, especially within the context of crop prices in Africa (Solano-Hermosilla et al., 2022).Section 4.2.presents our empirical analysis utilizing non-parametric bivariate conditional density estimations (CDEs) of our collected data on farmers' prices against several measures of monopsonistic power.

Crowdsourcing farmer prices in Malawi
In mid-2019, IFPRI and Farm Radio Trust (FRT) undertook a pilot study in southern Malawi to access the feasibility of collecting data on the sales prices farmers receive using crowdsourcing (Ochieng, 2019).Due to limited ownership of smart phones, poor telecommunications infrastructure and high internet costs, the crowdsourcing method selected involved a competition in which pigeon pea and chickpea farmers were invited to call or text a toll-free telephone number operated by FRT to report the prices and locations of their most recent crop sale. 2 To incentivise participation, the telephone numbers all farmers who called in were entered into a weekly raffle with a chance to win a voucher redeemable at any outlet of a major agro-input dealer.Between mid-August and the end of October 2019, 637 farmers from fifteen districts called the toll-free telephone line operated by FRT to report the prices they had received for their legumes.However, just eight farmers reported their sales of legumes via SMS.
Since the pilot study demonstrated the practicality of collecting data on the prices farmers received using telephone-based crowdsourcing methods, the study was upscaled to a nationwide exercise during the main marketing season of 2020.Between April 15 and July 31 of that year, the prices farmers received for maize, the main food staple, and soybean, a relatively new cash crop were crowdsourced by broadcasting jingles in local languages on three leading radio stations (the Malawi Broadcasting Corporation, the Voice of Livingstonia, and Zodiak Radio).
A new approach to monitoring farmer prices 877 The jingles invited farmers to report the prices and locations at which they had sold their maize and soybeans to a toll-free call centre operated by FRT. 3 The telephone numbers of farmers who called-in were entered into a weekly raffle with a chance to win one of three vouchers each worth Malawi Kwacha (MWK) 25,000 (approximately US$ 33), redeemable at any outlet of a major agro-input dealer.
In addition, to collecting data on the prices received, farmers were also asked about the volume of maize or soybeans they had sold, the location of sale and to whom they had sold (see online Supplementary materials for the text of the radio jingles, the call-centre checklist, and the questionnaire for the subsequent follow-up phone survey).In the four months of data collection between April and July 2020, a total of 1,048 maize and 1,265 soybean farmers called FRT to report the prices at which they had sold these crops.This timing happened to coincide with the first wave of the COVID-19 pandemic in Malawi, so the IFPRI-FRT crowdsourcing exercise was able to provide valuable information on the crop prices paid to farmers during a period when the official system for monitoring maize and other agricultural prices was operating erratically and under considerable strain.
A small number of duplicate calls (34 for maize, and 30 for soybeans) were then eliminated from the transactions level data. 4At the end of the marketing season, the farmers who had reported their sales prices were called back by the Farm Radio Trust call centre, and asked some additional questions about their age, education, farm and their crop marketing.The length of the follow-up calls was 10-12 min and 1,775 (77%) of the farmers who had reported sales to the call centre were re-contacted and interviewed.The total cost of the 2020 crowdsourcing exercise was approximately MWK 12.75 million (about US$ 17,250) of which 52 per cent was for the broadcasting of the radio jingles, 37 per cent for the operating costs of the call centre, and 11 per cent for vouchers.This equates to a unit cost of approximately MWK 5,550 (US$7.44)per call.We believe that costs could be reduced substantially by reducing the frequency of the radio jingles and by moving the call centre to another organisation.
We are mindful of the self-selecting nature of our crowdsourced data, which relies on farmers calling the FRT call centre to report their sales of maize and soybeans.We sought to minimize self-section through the weekly raffle for vouchers, which provided an economic incentive to encourage information disclosure.Since the analysis in Section 5 indicates that our sample is not representative of all maize and soybean sellers in Malawi, we also apply the non-parametric methods described in the next sub-section to analyze our crowdsourced data.
It should be noted that COVID-19 movement restrictions in Malawi were less severe than many other low-income countries.While schools were closed and social distancing measures were implemented for markets, public transportation plus social and religious gatherings, the outcoming Governments announcement of a 21-day lockdown in mid-April was blocked by the High Court and never enforced (Baulch, Botha, & Pauw, 2020).Moreover, due to favourable rains and no major cyclone events, a reasonably good maize crop of almost 3.8 million metric tons was harvested in early 2020.Abundant food supplies together with relaxed restrictions on agricultural markets kept daily retail maize prices during the post-harvest season more stable than in previous years (Figure 1) with the World Food Programme reporting similar trends for other food staples.Furthermore, while commercial flights were suspended from April 2000, land borders with neighbouring countries remained open until late December, and a rise in cross-border maize exports (primarily to Zambia and Tanzania) was recorded from June to October (FEWS NET, 2023).Nonetheless, it should be recognized that the four months during which the crowdsourced farmer prices analysed in this paper were ones of exceptional uncertainty in Malawi.This uncertainty is likely to have influenced both the timing and volume of maize and soybeans sales and hence the price expectations of farmers and traders.However, the effects of COVID-19 on farmer prices go beyond the focus of our study and present substantial opportunities for future research.
We provide a visualisation of our crowdsourcing data-collection process in Figure 2.  A new approach to monitoring farmer prices 879

Conditional density estimation (CDE) of farmer prices
Much of the empirical examination of farmer price behaviour assumes that prices converge to a single steady state (Baffes, Kshirsagar, & Mitchell, 2019;Lindenblatt & Feuerstein, 2015), however, the reality is more complicated given the potential impacts of monopsonistic buyer power and possibility of price manipulation within thinly-traded African crop markets (Dillon & Dambro, 2017;Kherallah et al., 2002).Indeed, the usual notion of steady-state food price convergence becomes more complex, factoring in both dynamic and spatial characteristics of said prices, and it is unlikely that we possess knowledge of the true distributional forms of farmers prices.As such, we elect to use a non-parametric distribution analysis, running kernel conditional density estimates (CDEs) on crowdsourced data set on farmers' crop sales enabling us to address the complexity of data without assuming any parametric form.Moreover, since the farmers who called in to report crop sales were self-selecting, they cannot be regarded as a random sample of farmers.In particular, farmers with mobile phones and radios are likely to be over-represented while farmers who did not sell any maize or soybeans are excluded. 5 As CDEs do not impose any assumptions on the underlying distribution of the data generating process, this allows us to capture the entire shape and dynamics of the price distribution including the presence of multi-modalities whilst also addressing the limitations of our crowdsourced data.We utilise the Hyndman, Bashtannyk, and Grunwald (1996) modified form of the original Rosenblatt (1956) estimator, which can be represented as: N is the number of farmers within our crowdsourced price dataset, and h y and h x are the optimal bandwidths for variables y and x, computed using the reference rules in Bashtannyk and Hyndman (2001) and Hyndman and Yao (2002) to ensure minimisation of the integrated asymptotic mean squared errors (IMSE).Y i and X i are the realisations of a random vector of variables y and x, and K denotes the Epanechnikov kernel density function (Hyndman et al., 1996).
For our examination of farmer prices in Malawi, we treat the prices of maize and soybean as our response variable (Y), whilst we have our a priori representations of monopsonistic power as the conditioning variable (X).We utilise three measures for monopsonistic power: bargaining power (Jaleta & Gardebroek, 2007), volume of sales, and time to market (Zant, 2018).Each of these conditioning variables are outlined below.
Following (Jaleta & Gardebroek, 2007) bargaining power (a) is calculated as, where P Ã is the final sales price, p I B and p I S represent the buyer's and seller's initial price offers, and a s [0, 1].A high a s , i.e. closer to 1, represents greater bargaining power for the seller (here the farmer), whilst a low a s , indicates more bargaining power for the buyer.Sales volume is measured by the number of 50-kg bags of maize or soybean sold in each transaction.Larger buyers, especially those with their own pick-up trucks or bigger vehicles, may be expected to purchase more bags of maize or soybeans in each transaction, which gives them greater leverage in price negotiations.Finally, time to market is expressed in terms of how long (in minutes) it takes the farmer to walk to the point of sale.As time to market increases, farmers will be more inclined to conclude sales at a given offer price from the buyer.
Additionally, to capture any temporal dynamics of prices, we subdivide our data into early and late season.April and May 2020 are described as the 'early season' and June and July 2020 as the 'late season'. 6We also engage in a spatial and temporal examination of our crowdsourced data through a within season, price-mapping exercise.Using a mapping dashboard developed by IFPRI, we generate color-coded price maps of central Malawi by month.

Findings
Examining our crowdsourced data, we discover that prices were reported from all districts in Malawi, except for Likoma island, with most callers coming from the main maize and soybean producing districts in the centre of the country (Figure 3).
The top panel of Table 1 illustrates that, in each transaction, maize farmers sold an average of 733 kg (approximately 14 bags) at MWK151/kg, while soybean farmers sold 518 kg (about 10 bags) at MWK232/kg.However, as shown by the medians, the sale volumes are influenced by the presence of a few large sales.Almost a fifth (19.7%) of farmers reporting to the call centre had sold two bags (100 kg) or less of maize or soybeans.Most farmers sold to assemblers, small traders and retailers (80% of maize and 90% of soybean sales).Just 2 per cent of maize farmers and 5 per cent of soybean sellers reported sales to larger traders or processors.About 18 per cent of maize farmers and 5 per cent of soybean farmers reported sales to the Agricultural Development and Marketing Corporation (ADMARC), Malawi's main agricultural marketing parastatal, which buys grain and other crops from farmers at fixed, pan-territorial prices that usually correspond to official MFGP.Less than 1 per cent of farmers reported sales to other value chain actors, such as retailers and consumers.Most sales took place at local markets (41% for maize and 51% for soybeans) or on farm (around 40% for both crops).
On average, farmers had been contacted by three potential buyers in the previous seven days and had about two previous transactions with the buyer they sold to.Around three-quarters of traders agreed with farmers on their assessment of crop quality. 7Some 74 per cent of maize farmers and 64 per cent of soybean farmers were aware of the MFGP, which were MWK200/ kg for maize and MWK300/kg for soybeans during the 2020 marketing season.However, when asked to state the MFGP, about 5 per cent of farmers mentioned incorrect prices.A new approach to monitoring farmer prices 881 Figure 4 illustrates the prices received by farmers who sold maize and soybeans during the main 2020 marketing season in panels (a) and (b) respectively.The blue line in each diagram shows the average price received by farmers (whether on their farm or in nearby assembly points and markets), while the red line shows the pan-territorial MFGP announced by the Ministry of Agriculture and Food Security in April each year.As can be seen, with the exception of a few days, the average daily prices received by farmers for both maize and soybeans remained substantially below their official MFGP throughout the 2020 marketing season.This is confirmed in Table 2 in which shows that average prices farmers received in 2020 were about 30 per cent below the MFGP for maize and 23 per cent below that for soybeans.The solid yellow and hollow black rings in panels (a) and (b), Figure 4 indicate the minimum and maximum price respectively, whilst the vertical grey lines indicate the range of prices paid to farmers on each day that more than one sale was reported.On most days, the minimum price paid to farmers was much further below the MFGP than the maximum price received by farmers was above it.
The average farmer received 75 per cent of the MFGP for maize and 77 per cent of the MFGP for soybeans (Table 2).Almost a quarter (24.5%) of maize farmers received a price equal to or greater than the MFGP, while a similar percentage of maize farmers received 60 per cent or less of the MFGP.For soybeans, just under a tenth of farmers (9.8%) received prices equal to or greater than the MFGP, while the bottom tenth of farmers received 67 per cent or less of the MFGP.
To assess the representativeness of our crowdsourced data, we compare the characteristics of the farmers who participated in the follow-up survey to the crowdsourcing competition to farmers with cell phones who sold maize and soybeans in the, nationally representative, fifth Integrated Household Survey (IHS5) conducted between April 2019 and March 2020.While 71 per cent of the 11,434 households enumerated in the IHS5 cultivated maize, only a fifth of them sold maize, with the remaining four-fifth of households retaining the crop for their own consumption.For soybeans, just over 8 per cent of household grew soybeans all of whom sold it, confirming soybeans role as a cash crop.Columns (a) and (b) of Table 3 compares the means for selected variables for maize and soybean sellers with mobile phones from the crowdsourced and IHS5 surveys, respectively.Column (c) shows the 95 per cent confidence interval of these means based on the IHS5.
Our data shows that farmers who participated in the crowdsourcing exercise were more likely to be male, younger and have some secondary school education than farmers who sold maize and soybeans in the IHS5.They also cultivate larger acreages of maize and soybeans and sold  A new approach to monitoring farmer prices 883 more than twice the amount of these crops during the main marketing season.Since the crowdsourcing survey means fall outside the 95 per cent confidence intervals from the IHS5 for all variables, we cannot conclude that the crowdsourced data is representative of farmers who sell maize and soybeans in Malawi.However, we do not believe that the prices reported in our crowdsourcing exercise differ from the prices paid to other types of farmers during the 2020 marketing season.Ideally, we would like to triangulate our crowdsourced prices by comparing them with those of the IHS5 and other data sources.Unfortunately, the timing of the IHS5 and the lack of availability of alternative price data do not allow for this.As indicated in the introduction, data on farmgate prices in Malawi is limited and sparse, especially for soybeans.

Maize price dynamics
We present the estimated CDEs for maize in Figure 5. Panels (a)-(c) show the price of maize conditioned by the alpha measure of bargaining power, panels (d)-(f) show the price of maize  Notes: this table presents a comparison between our crowdsourced data and the average results of IHS5.Gender is measured as 1 ¼ male, and 0 ¼ female; age is measured in years; level of education is measured as 0 ¼ no formal education, 1 ¼ some primary schooling, and 2 ¼ some secondary schooling; farm size given in acres; area of crop cultivated is measured in acres and is for the wet season harvest; quantity of maize sold is measured in kilograms.
conditioned against volume of sale, and panels (g)-(i) show the price of maize conditioned against the time to market.From panel (a) we observe that as bargaining power approaches zero, that is greater power for the farmer, the price of maize is higher.More specifically, we discern a cross-over above bargaining power of 0.5.We also see multiple statistical modes within the distribution of prices for maize, especially at lower values of alpha, but with a small number of sales occurring at or near MWK200/g, especially towards the end of the marketing season.We believe these multiple modes are related to known characteristics of maize markets in Malawi, especially during the early marketing season.Most maize and other annual food crops in Malawi are sun-dried with substantial price discounts for 'wet' maize that has not reached the required 12.5 moisture content. 8Early season trading tends to be dominated by assemblers and other small-scale traders, many of whom operate as commission agents for-or are directly financed-by larger trading companies, who do not start buying directly until later in the season.In addition, the parastatal National Food Marketing Agency (NFRA) usually does not have the funds to purchase maize in the early season while any grain that ADMARC purchases has to be financed by short-term, costly commercial borrowing. 9The bimodal distribution of the farmer prices, especially in the early season may be linked to these phenomena.
Examining our CDEs for price of maize conditioned against volume of sale (measured by the number of 50-kg bags sold), we see a more consistent pattern over time, with a primary peak around a sales price of MWK150/kg and a secondary ridge around MWK200/kg.There is also great dispersion of prices at the lower end of the CDEs, which accords with the greater variation in the moisture content for maize in the early season noted above.The secondary ridge centred around a price of MWK200/kg, which is the official MFGP, is more pronounced in the late season (panel f), which is as expected given the timing of ADMARC maize purchases.More than two-thirds of ADMARC maize purchases during the 2020 marketing season occurred in June and July.The secondary peak at sales volumes between 40 and 50 bags in the early season (panel e) is linked to a few purchases by larger traders and ADMARC.
Finally, panels (g) to (f) present our estimations of price against travel time to market (measured in minutes).A relatively stable relationship between price and time is observed with similar CDE in both the early-and late-season.This is confirmed by the lowess plots in Figure 6, from which it will be noticed that the prices paid are very similar for sales of maize (panel a) and soybeans (panel b) on the farm/roadside, at local markets or in the larger trading centres.The higher price paid for maize sales to ADMARC corresponds to the MFGP, although we know from Figure 5 (d) that these are only paid to a minority of maize sellers.
From a conventional point-space perspective, most agricultural economists would expect the prices to fall with distance to market as the buyers and sellers incorporate transportation and other transfer cost into their price (Tomek & Kaiser, 2014).In addition, within season prices should rise over time in line with the cost of storage (Timmer, Falcon, & Pearson, 1983;Tomek & Kaiser, 2014).However, our results suggest otherwise.This suggests that Malawian farmers-who typically travel to the point of sale by walking or by bicycle-absorb the cost of transportation when making crop sales.The distance that farmers travel to market is typically not far, as trading centres exist in all districts of Malawi and maize is actively traded in all of them.Furthermore, despite the maize export ban that has been in force, almost continuously since December 2011, Malawi's borders are long and porous.There is an active informal trade in maize along most of Malawi's western border with Zambia, along its northern land border with Tanzania, and also some informal trade with Mozambique in the south and east (Edelman & Baulch, 2016;Porteous, 2017).So, aside from the likely presence of monopsonistic power, the geographic dispersion and fragmentation of Malawian markets and the cross-border maize trade mitigates against conventional spatial and temporal price relationship.We further disaggregate our spatial and temporal argumentation in relation to geography by highlighting the price evolution of maize over the 2020 marketing season across Malawi in Figure 7, panels (a) and (b) represent the early season and panels (c) and (d) highlight late season price dynamics geographically at the traditional authority (TA) level.We observe from Figure 7, an increase not only in the number of TAs at which sales take place but also a progression from light to dark shading zones suggesting an increase in maize prices over time.These maps, which have been made available in interactive form to processors, traders and relevant government agencies, may help to inform their crop sourcing and procurement decisions in the future.

Soya price dynamics
The results of the CDEs for soya are presented in the Figure 8. Panels (a)-(c) show the price of soya conditioned by the alpha measure of bargaining power, panels (d)-(f) show the price of soya conditioned against volume of sales, and panels (g)-(i) is the price of soya conditioned against time to market.Examining price against bargaining power for soybeans, we see more transactions at lower prices as bargaining power moves towards buyers.This is the case for the all-sample estimation in panel (a), as well as the early-and late-season estimations in panels (b) and (c).However, like maize, the CDEs still exhibit the presence of multiple statistical modes, with no single steady state, point space relationship between prices and bargaining power for A new approach to monitoring farmer prices 887 soybeans.This may be linked to there being four large buyers and processors of soyabeans in Malawi, two of whom are located just north of Lilongwe, the capital, while the other two are located on the outskirts of Blantyre, Malawi's second city and commercial hub, which is around five hours drive to the south.In addition, there is known to be an active cross border trade in soybean with Zambia and Zimbabwe, where there is high demand both for processed soybean and its by-product soya cake (which is used in the manufacture of commercial animal feeds), although trade restrictions against soya exports are frequently muted and periodically implemented in Malawi (Edelman & Baulch, 2016).Unlike maize, the CDEs for soybean prices against the volume of sale in panels (d) and (g) exhibit a more unimodal distribution, albeit a relatively flat relationship.A priori, we would expect prices paid to be lower at higher volumes of trade, but the results suggest that prices remain relatively stable even when conditioned by volume of sale as measured by the number of 50-kg bags sold.Finally, we examine the CDEs for soybean price against time to market-panel (g)-and similar to maize, we observe steady prices which hardly vary with travel time to the point of sale.Early-and late-season estimations of price against time to market show similar behaviour as shown in panels (h) and (i).We offer a similar explanation for soybeans to that of maize, as it is possible for farmers-who mostly travel to the pointing of sale on bicycle or by foot -to absorb most of the transportation costs incurred in marketing their soybean.A telephone survey of 917 soybean farmers conducted by IFPRI in 2021 found that 52 per cent of farmers walked to village markets or trading centres to sell their soybeans, while another 32 per cent travelled to the point of sale by bicycle.The maps in Figure 9 show the evolution of soybean prices at the TA level between April and July 2020.They show considerable spatial variation in average prices with prices generally lower in TAs far from the major trading and processing centres.As time progresses, there were also less TAs shaded in red and more TAs shaded in green, indicating a tendency for prices to rise with time and the cost of carry-see panels (g) to (i) Figures 5 and 8 for maize and soybean respectively.Interactive maps similar to these have been made available to processors and large traders, and should prove useful in guiding their procurement decisions in the future. 10If larger buyers decide to source in lower price TAs, this should help to drive up the prices that farmers receive in these areas. 11In the future, it may prove beneficial to embed crowdsourced price maps in the National Agricultural Monitoring Information System or other web-based platforms (including Facebook). 12

Conclusion and future directions
This study has outlined a new and innovative approach for monitoring the prices that farmers in a low-income economy are paid for their crops.An application to Malawi is presented, which shows the feasibility of this telephone-based crowdsourcing method in a setting in which internet connectivity is limited but mobile phone coverage is reasonable.Our response rate in A new approach to monitoring farmer prices 889 Malawi was much higher than other internet-based crowdsourcing experiments to collect food prices in Kenya, Sierra Leone and Uganda using the Knoema platform (Donmez et al., 2017).As our competition-based method only requires making payments to a small proportion of those who report prices, the cost of crowdsourcing at MWK5,550 (US$7.44)per reporting farmer is also less than the official system involving price collectors, while the resulting data is more geographically complete.An additional unexpected advantage to the method, its compliance with the regulations for social distancing required during the COVID-19 pandemic, became apparent during the 2020 harvest/marketing season.
The results show that the vast majority of farmers in Malawi received substantially less than the MFGP announced by the Government.Only 24.8 per cent of the farmers who reported selling maize between April and July 2020 received at least the MFGP.The corresponding figure for soybeans is just 9.8 per cent.Furthermore, the average farmer received just 75 per cent of the MFGP for maize and 77 per cent of the MFGP for soybeans.Our non-parametric conditional estimates further disaggregate the prices paid to farmers using three indicators of monopsonistic power.Specifically, we discover a pattern of higher prices at higher levels of seller bargaining power and the presence of multiple equilibria.The conditional distributions of farmer prices against that of volume of sales for maize show the presence of twin peaks within the CDEs of price and volume of sales, with a primary peak at a price of MWK150/kg and a secondary ridge at the maize MFGP of MWK200/kg.The CDEs for soybean prices are more unimodal, especially for soya with a single peak at approximately MWK230/kg.We also observe that travel time to the point of sale exerts a negligible effect on the prices received by farmers, suggesting that farmers-who typically travel to the point of sale by bicycle or walking-absorb the cost of transportation when making crop sales.The highly fragmented and diverse geography of food markets in Malawi, which is crucially conditioned by Malawi's long and porous borders with neighbouring countries, must also be taken into account in understanding the spatial and temporal price patterns revealed.
Moving on to policy issues, the instinctive reaction of most farmers organisations, government officials and policy makers to our findings is that Malawi's MFGP should be enforced more strictly.While public policy concerns are certainly raised by localised monopsonistic buying power, we would argue that stricter enforcement of MFGP, often 'backfires' by creating opportunities for rent-seeking behaviour by the officials who are meant to be enforcing MFGP and other crop marketing regulations.Instead, we would argue that promoting competition is likely to do more to raise the prices farmers receive than imposing penalties on the minority of small traders who are observed not adhering to MFGP.Measures to promote competition might include improving roads and grain storage facilities, developing grading standards for the major crops, and encouraging grain traders to use hand-held moisture meters.Relaxing the cabotage rules that restrict haulage companies registered outside Malawi from competing with domestic firms, would also help to reduce road freight rates and marketing margins.Such reforms may well, however, run into institutional barriers as it seems likely that the sources of monopsonistic power in Malawian agricultural markets lie higher-up the marketing chain (Chirwa, 2010;Lall, Wang, & Munthali, 2009;Ncube, Roberts, & Vilakazi, 2016;Sitko & Jayne, 2014).
To conclude, a competition and phone-based crowdsourcing approach to collecting farmer prices has much to recommend itself in developing countries where internet connectivity is limited but mobile phone coverage is reasonable.Offering prizes to farmers who report transactions in a weekly competition can achieve wider geographical coverage and include more types of transactions than traditional trader surveys, as well as revealing features of agricultural markets (such as the non-binding nature of minimum farmgate prices) that are not apparent from conventional price monitoring exercises.However, these advantages come with both statistical and financial costs along with wider considerations for the sustainability of gamification of the data-collection process.The statistical cost is that since crowdsourced price reporting is self-selecting, one should

Figure 1 .
Figure 1.Daily percentage change in retail maize prices in Malawi from April to July: 2018-2020.Notes: this figure illustrates the daily percentage change in Malawian retail maize prices for the harvest period for 2018, 2019, and 2020.The figure is drawn from IFPRI Malawi retail price monitoring data.

Figure 2 .
Figure 2. Breakdown of our crowdsourcing data-collection process.Notes: this figure shows our crowdsourcing data collection framework.The framework was designed by IFPRI in collaboration with Farm Radio Trust for a pilot exercise undertaken in mid-2019, and then refined for the main 2020 harvest season.

Figure 3 .
Figure 3. Number of sales reported by district (April-July 2020).Notes: The figure indicates the number of transactions recorded within each district in Malawi over the period of data collection from April to July 2020.The bars in yellow and green indicate the total number of transactions reported to the Farm Radio Trust call centre for maize and soybean, respectively.

Figure 4 .
Figure 4. Daily maize and soybean prices per kilogram received by farmers, April-July 2020.Notes: this figure presents the spread of our crowdsourced prices over the 2020 marketing season.Panel (a) indicates the prices for maize, whilst Panel (b) highlights the prices for soybean.For both panels, the solid blue line represents the average daily price, the solid red line is the 2020 marketing season's set minimum farm gate price, and the solid grey circles and hollow black rings indicate minimum and maximum daily prices respectively.

Figure 5 .
Figure 5. Conditional density estimates for maize prices.Notes: this figure displays the bivariate conditional density estimates for maize using our crowdsourced data.In all panels, price is the response variable.Panels (a)-(c) estimate price of maize against a calculated alpha measure of bargaining power.Bargaining power is on a scale of 0 to 1 where values closer to zero indicate greater bargaining power for the seller and values closer to one indicating greater bargaining power for the buyer.Panels (d)-(f) estimate price of soybean against volume of sales.Volume of sales is measured as the number of 50-kg bags sold of maize.Panels (g)-(i) estimate price of maize against time to market, where time to market is the number of minutes taken to arrive at place of sale.Higher numbers indicate longer travel times.Panels (a), (d), and (g) are all-sample, which is data for the entire 2020 marketing season.Panels (b), (e), and (h) are early marketing season estimations.Panels (c), (f), and (i) are late marketing season estimations.We define early season as April-May 2020 and define late season as June-July 2020.

Figure 6 .
Figure 6.Lowess plots of price against time to market.Notes: Panel (a) and (b) represent the combined lowess plot for the price of maize, and soybean against time to point of sale, respectively.Note that time to market is measured in minutes.We disaggregate the point of sale for each crop into farm/roadside, market, trading centre, and the ADMARC.ADMARC is not present in the panel (b) as very little soybean is purchased by ADMARC.In both panels, a blue, long-dashed line represents sales at the farm/roadside, the solid yellow line indicates market sales, a solid red line is sales the trading centre, and the red, short-dashed line is sales at the ADMARC.Minimum farm gate prices are MWK200/kg for maize, and MWK300/kg for soya.

Figure 7 .
Figure 7. Geographical distribution of maize prices over the marketing season.Notes: this figure is a spatial representation of the changes in maize prices over the 2020 harvest season in Malawi.Panel (a)-(d) represent April 2020, May 2020, June 2020, and July 2020, respectively.Spatial mapping is for the traditional authorities in Malawi.Regions transition from light grey where prices at MWK60/kg to dark grey where prices are MWK200/kg

Figure 8 .
Figure 8. Conditional density estimates for soybean.Notes: this figure displays the bivariate conditional density estimates for soybean using our crowdsourced data.Panels (a)-(c) estimate price of soybean against a calculated alpha measure of bargaining power.Bargaining power is on a scale of 0 to 1 where values closer to zero indicate greater bargaining power for the seller and values closer to one indicating greater bargaining power for the buyer.Panels (d)-(f) estimate price of soybean against volume of sales.Volume of sale is measured as the number of 50-kg bags sold of soya.Panels (g)-(i) estimate price of soybean against time to market, where time to market is the number of minutes taken to arrive at place of sale.Higher numbers indicate longer travel times.Panels (a), (d), and (g) are all-sample, which is data for the entire 2020 marketing season.Panels (b), (e), and (h) are early marketing season estimations.Panels (c), (f), and (i) are late marketing season estimations.We define early marketing as April-May 2020 and late season as June-July 2020.

Figure 9 .
Figure 9. Geographical distribution of soybean prices over the marketing season.Notes: this figure is a spatial representation of the changes in soybean prices over the 2020 marketing season in Malawi.Panel (a)-(d) represent April 2020, May 2020, June 2020, and July 2020 respectively.Spatial mapping is done for the districts in Malawi.Regions transition from light grey where prices at MWK180/kg to dark grey where prices are MWK340/kg

Table 1 .
Descriptive statistics for transactions and farmersNotes: this table presents the descriptive statistics for the crowdsourced data over the period April-July 2022.Quantity sold measured in kilograms; final sales price is Malawi Kwacha per kilogram.

Table 2 .
Prices received by farmers versus minimum farm gate price this table breaks down the crowdsourced price relative to the minimum farm gate prices for the 2020 marketing season.Crowdsourced prices have been winsorised at the 99th percentile.Number of transactions remaining after outlier detection is shown in the final column.

Table 3 .
Characteristics of maize and soybean sellers: a comparison of crowdsourced and IHS5 data