© T. Brunelle and G.d.l.P. Bayiha, Hosted by EDP Sciences 2025

1 Introduction

Agriculture in Cameroon is characterized by a combination of food production systems for the national market and plantation agriculture for both the national market (oil palm) and the international market (cocoa, coffee, and rubber). It benefits from a large agro-ecological diversity, which provides an adequate environment for the development of a large number of plant species (Bayiha et al., 2019, 2020). However, Cameroonian agriculture is also affected by a large number of pests and diseases that can impede plant growth and reduce yields. In many cases, the use of synthetic chemical pesticides is the preferred method of control.

Since 1995, pesticide consumption has risen sharply, reaching 7300 tons of active substances in 2020 or nearly 1 kg/ha of cultivated area. This intensification of pesticide use has raised concerns about environmental impacts (particularly on soil fertility) and health. Although levels of pesticide use in Cameroon remain substantially below those observed in certain regions of intensive agriculture (Tang et al., 2021), there is a concern to prevent the country from following a path of intensive pesticide use that could lead to lock-in. Public policy instruments for pesticide reduction (standards, bans, taxes or subsidies) require in-depth knowledge of economic determinants to assess their effectiveness and financial impact on public finances and stakeholders (Baumol and Oates, 1988; Nielsen et al., 2023). This concerns in particular the elasticities (i.e. the variation in one quantity caused by the variation in another) of different economic components, such as the price of pesticides and the price of agricultural products.

There is a large literature on estimating the price elasticity of pesticides, which focuses almost exclusively on the United States or Europe. This literature has been summarized in the meta-analysis conducted by Böcker and Finger (2017). According to this study, pesticides appear to be price inelastic, but with highly heterogeneous values across different studies, ranging from complete inelasticity to relatively high elasticity. Using individual farms observed in the period 2006-2009 in the French Département de la Meuse, Bayramoglu and Chakir (2016) find a positive and significant effect of rapeseed price on the demand for pesticide inputs, but no significant relationship between pesticide use and its price. More generally, the literature emphasizes the importance of socio-economic factors as drivers of pesticide use compared to factors related to technological issues in the natural sciences (Hu, 2020). For instance, Rosenheim et al. (2020) show that crop values is the dominant driver explaining the variation in pesticide use across 93 crops grown in California.

To our knowledge, no study has examined the economic determinants of pesticide use in a tropical context. The reliability of data is an important challenge in estimating pesticide use elasticities in tropical countries. FAOSTAT data (https://www.fao.org/faostat/en/#data) is the main source of data on pesticide use and trade, but its reliability varies considerably across countries. In this article, we address this challenge by cross-checking FAOSTAT data with data from the recently published Global Pesticide Use and Trade Database (GloPUT) (https://osf.io/dyu38/?view_only=7e39ab440f104ed2b61591a086f89a0b), which improves estimates for 137 countries using bilateral paired mirror trade statistics and an index of reporter reliability (Shattuck et al., 2023). In addition to data quality concerns, the limited availability of data also poses a challenge to robust effect estimation. To address this issue, we use a specific methodology designed to mitigate small sample limitations. Another challenge is the inclusion of pesticide subsidies, which are not recorded in the databases. To address this issue, we complement our analysis with a survey of 30 farmers that provides a local estimate of the price elasticity of pesticides, that accounts for subsidies. Building on this approach, our paper aims to estimate the response (or elasticity) of pesticide use in the Cameroon context to a set of economic variables, including pesticide import prices, agricultural commodity prices, and labor availability. Our methodology is based on econometric estimation of national data from 1995 to 2018, which will be compared with elasticities calculated from survey data collected from a sample of 30 farmers in the coastal and western regions of Cameroon between 2021 and 2022.

2 Context of pesticides use in Cameroon

Pesticide reduction is an emerging issue in Cameroon, related primarily to concerns about soil quality rather than human health. This concern is particularly pressing in the northern regions, where severe soil degradation is attributed to excessive pesticide use. There is also a higher level of awareness in the cocoa sector, thanks to a number of international programs working in this field. On the other hand, there is less concern in the Adamao region, where soil fertility is being maintained. Overall, there is a "slight rise in awareness" of the issue, but this has not yet been translated into concrete action at national scale due to the absence of economically efficient technical alternatives.

According to FAOSTAT data (whose reliability is discussed in Sect. 3.2), pesticide consumption in Cameroon has increased sharply since the mid-1990s. Two-thirds of this increase is due to the use of fungicides, particularly in cocoa and coffee production.

The pesticides used in Cameroon are imported from partner countries, which have changed considerably over time (Fig. 1). In the 1990s, the main partners were European countries (especially France) and the United States. In 2018, most pesticide imports came from China (76% of total pesticide imports) and, to a lesser extent, India (9%). This dynamic is consistent with the pattern observed in international pesticide markets for several years (Wanner et al., 2020). Since the early 1990s, China has expanded its pesticide production capacity, becoming the world’s leading exporter by the mid-2000s. In the 2010s, China continued this strategy by buying up major agrochemical companies, such as Syngenta, which was acquired by the chemical conglomerate ChemChina in 2016. By 2020, China accounted for 34% of global pesticide exports, up from about 5% in the early 1990s (Shattuck et al., 2023).

A variety of public and private intervention instruments (programs, projects, interventions, or support) have been implemented in Cameroon to regulate the use of pesticides or to promote alternatives. Bayiha et al. (2025) distinguish three key periods:

• 1960–1990’s: policies are oriented primarily towards the use of synthetic chemical pesticides to improve the performance of agricultural systems through intensive use of pesticides;

• 2000–2018: A dynamic shift towards the emergence of regulatory instruments for pesticides, mainly driven by international donors, with an actual implementation (2008–2018) that remains poorly documented ;

• 2018–2025: Organic and ecological agriculture are institutionally recognized with the drafting of two key documents: (i) the prospective analysis note on organic farming; and (ii) the National Development Strategy 2020–2030. The latter document promotes two distinct trends in pesticide use: (a) encouraging the use of pesticides for yield performance; and (b) promoting the development of organic farming.

Currently, there are two main instruments designed to encourage the use of pesticides by influencing their price: a direct subsidy to farmers in the cocoa and coffee sectors of between 20% and 40% of the base price of pesticides, depending on the year. In 2023, the total amount of this subsidy was estimated at 6.3 billion CFA francs (around 10 million US dollars), corresponding to around 30% of the base price. The second type of subsidy is the VAT exemption (19.25%) on the purchase of pesticides by importing companies. These subsidies can be complemented by other regional interventions in specific product sectors. In addition, there are a variety of direct and indirect interventions on the price of pesticides (e.g. a pilot program to subsidise the purchase of inputs for cocoa in the Littoral region, launched in 2002), making it difficult to keep track of them all.

Fig. 1 Origins of pesticide imports into Cameroon in 1995 and 2018. Source: Shattuck et al. (2023) Origine des importations de pesticides au Cameroun en 1995 et 2018.

3 Data and methods

3.1 General approach

We study the effect of 3 variables on pesticide use: pesticide price, output price of agriculture and a proxy for labor cost. The coefficient associated to pesticide will yield the own price elasticity, which is expected to be negative, i.e., an increase in the price of pesticides is expected to reduce their use. An higher output price is expected to stimulate pesticide use, by making it more attractive for farmers to invest in protecting their crops against pests and diseases, as the income per unit produced is increased (Rosenheim et al., 2020). The impact of pesticide reduction on working time is still debated in the literature. Guo et al. (2022) find a positive relationship between agricultural labour wages and pesticide use, which indicates that pesticide use and working time can be considered substitutes, but Rödiger et al. (2024) conclude that reduced pesticide use in crop production does not necessarily lead to an increase in working time requirement. The inclusion of a proxy for labor costs in the econometric analysis is intended to assess this effect in the Cameroon context.

Given the complexity of the pesticide subsidy system in Cameroon (see Sect. 2), it is impossible to obtain comprehensive time series of subsidy amounts over a long period. To overcome this problem, our approach is based on a two-step estimation:

• First, we conducted an econometric estimation of elasticities based on pre-subsidy prices, which are the only data available to us (see Sect. 3.3);

• Second, we carried out an empirical estimation of local variation in pesticide price (after subsidies) and output price elasticities based on survey data from a sample of 30 farmers in 2 regions of Cameroon (West and Littoral) between 2021 and 2022 (see Sect. 3.2.2). This local empirical estimate is intended as a first approximation of the effect including subsidies, but is highly imperfect, in particular it can only be compared with the short-term econometric estimate and cannot be used to assess the accuracy of the long-term estimate.

3.2 Data

3.2.1 Data for econometric analysis

The FAOSTAT database provides data on pesticide use and imports, but its reliability varies greatly across countries (FAO, 2023). The GloPUT database was recently released to improve the reliability of pesticide use data using mirrored bilateral trade statistics and an index of reporter reliability (Shattuck et al., 2023). This database enables us to assess the quality of FAOSTAT data by comparing it with the GloPUT database. Although there may be significant differences for some countries (e.g. Ivory Coast), FAOSTAT and GloPUT data for Cameroon are fully consistent for total pesticide use, imports and exports in volume (net weight) and value between 1995 and 2018. The complete consistency between the two databases for Cameroon suggests that the FAOSTAT data for this country are reliable, supporting their use in our analysis. The good quality of Cameroon’s pesticide information system was also confirmed through personal communication by Cameroonian stakeholders competent on pesticide data produced in Cameroon. Changes in pesticide use between 1995 and 2018 and in the 3 pesticide categories (herbicides, insecticides, fungicides) are shown in Figure 2.

In addition to these data, we use commodity price data from the World Bank’s Pink Sheet database (World Bank, 2023b). All data used are annual time series covering the period 1995-2018, which is the period over which we were able to cross-check the FAOSTAT and GloPUT data.

Four categories of pesticides are considered: herbicides, insecticides, fungicides and total pesticides. All types of pesticides use are expressed in kilograms of active ingredient per hectare.

In the absence of publicly available pesticide price series, a standard method is to divide imports in value by imports in volume. The obtained price series, expressed in real terms using the World Bank deflator, shows that pesticide prices have fallen from $8.35/kg of formulated products (FP) in 1995 to $4.47/kg FP in 2018 (Fig. 3). In the FAOSTAT, only imports expressed in formulated products are available, which makes it impossible to distinguish price changes related to variations in product quality in terms of active ingredient concentration from changes in market conditions related, for example, to the level of competition between exporters. To address this issue, we compare the quantities of formulated products imported with the use of pesticides in active ingredients. Since Cameroon imports its pesticides, the two series should be parallel excluding stockpiling. We find that the ratio of use (in active ingredients) to imports (in formulated products) increases rapidly between 1995 and 2002 (from 0.01 to 0.15), and then becomes almost constant between 2002 and 2018 (∼0.36). This result does not indicate a downward variation in active ingredient concentrations, which leads us to consider other factors behind the price decline, including the dynamics of the global pesticide market. China’s growing importance on international pesticide markets, with its positioning on off-patent substances (Werner et al., 2022), seems a plausible explanation for the observed decline in pesticide import prices in Cameroon.

Another problem is the potential endogeneity of pesticide prices. Endogeneity refers to a situation in which an explanatory variable is correlated with the error term. It can arise as a result of different problems such as omitted variables or mutual influence between the response variable and the predictors (also referred to ʽsimulateneity bias’). The price of pesticides corresponds to the import price and could be influenced by negotiations with importers, which could have a varying effect depending on the demand for pesticides, and thus a reciprocal influence with pesticide use (simulateneity bias). However, further verification with experts from Croplife Cameroun — an international association of companies and professional organizations in the crop protection sector — confirmed that no such negotiations are taking place. Consequently, it is reasonable to treat pesticide prices as exogenous.

Due to the lack of data on pesticide use by crop types, this analysis focuses on the effect of aggregate crop prices on the different types of pesticides and on total pesticide use. For this reason, the price of agricultural products in Cameroon is estimated as the weighted sum of the world prices of the four main export commodities — Banana, Cocoa, Coffee, Cotton — by their share in the exports of these four products (see Fig. 4).

In theory, value added per worker in the agricultural sector is the most appropriate variable to capture labor costs. However, this measure is likely to suffer from significant endogeneity with respect to the dependent variable. To address this issue, the standard approach is to use an instrumental variable that is correlated with the potentially endogenous regressor but uncorrelated with the error term. In similar contexts, the number of job vacancies is often employed as an instrument. Unfortunately, this variable, along with other commonly used instruments, is not available in Cameroon’s national statistical databases.

As an alternative, we use the national unemployment rate across all sectors of the economy, sourced from World Bank (2023a). While this variable does not directly measure labor costs, it serves as a proxy for labor availability and can be interpreted as a shadow price of labor. Between 1995 and 2018, the unemployment rate in Cameroon fell from 8% to 3.6% of the population (Fig. 4). Under the assumption that the fall in the unemployment rate is not the result of an improvement in the quality of job matching, this decline can be interpreted as a reduction in the availability of labor, reflecting increased tightness in the labor market. As a result, the implicit or ʽshadow’ cost of labor is likely to have increased. Assuming that pesticides are substitutable for labor (see Sect. 3.1), such an increase in the implicit labor costs may have contributed to an increase in pesticide use. Importantly, national unemployment rate is expected to exhibit a lower degree of endogeneity than value added per agricultural worker, as it captures dynamics in the broader labor market rather than within agriculture alone. However, endogeneity cannot be completely ruled out, given that agriculture accounts for a large share of employment in Cameroon (43% in 2016 according to World Bank [2023a]).

Descriptive statistics of dependent and independent variables used in the econometric estimation are shown in Table 1, and sources and units are given in Table 2.

Fig. 2 Pesticide use in Cameroon in kilograms of active ingredients per hectare (kg AI/ha) Usage des pesticides au Cameroun en kilogrammes de matières actives par hectare (kg MA/ha).

Fig. 3 Pesticide price (real) in Cameroon Prix des pesticides (valeurs réelles) au Cameroun.

Fig. 4 Unemployment rate (left) and price of agricultural products (right) in Cameroon Taux de chômage (gauche) et prix des produits agricoles (droite) au Cameroun.

3.2.2 Survey data for local estimates

The surveys took place from May 23 to May 30, 2023, in the western region of Cameroon for rice, maize and tomato, and in the coastal region of Cameroon for cocoa (Tab. 3). The crops selected are among those prioritized by the Government of Cameroon. Given the time and cost constraints, a total of 30 farmers were selected in consultation with the managers of the Programme de consolidation et de pérennisation du conseil agropastoral − Amélioration de la compétitivité des exploitations familiales agricoles (PCP-ACEFA), a government agricultural extension organization in Cameroon. This organization assists hundreds of farmers throughout the country.

A two-part questionnaire was designed to gather both quantitative and qualitative information on plot characterization and farmers’ behavior towards changes in pesticide prices and market crops. The collected data pertained to the years 2021 and 2022 and included variables such as annual quantities of pesticides consumed, pesticide unit prices after subsidies, and selling prices of crops. Among the surveyed farmers, 67% are men and 33% are women, with more than half of the respondents aged over 50. Most of them farm small plots of between 0.5 and 1 hectare. When it comes to crop protection, the study reveals that the vast majority (90%) buy synthetic chemical pesticides from plant protection shops. As farm sizes are heterogeneous, data are rescaled to 1 hectare plot. For the purposes of the survey, all quantities were expressed in units to which farmers are accustomed (e.g., “bags”) and then converted to grams or liters. Given the data collected, it was only possible to calculate the own price elasticity of pesticide demand, nor was it possible to calculate elasticities by pesticide types. In comparison to the econometric estimation, the own price elasticity calculated here corresponds to short-term elasticities. The first step was to calculate the elasticity for each farmer, using equation (1). Subsequently, the mean and standard deviation were calculated for each crop types and for the entire sample of 30 farmers, with each farmer having the same weight, as data were collected on a standard area of 1 hectare.

$${}_{}ϵ=\frac{\left(PestUs{e}_{2022}-PestUs{e}_{2021}\right)/PestUs{e}_{2021}}{\left(PesticidePric{e}_{2022}-PesticidePric{e}_{2021}\right)/PesticidePric{e}_{2021}}$$(1)

An empirical elasticity provides a first approximation of the sensitivity of demand to price, but is subject to significant bias. First, it does not establish a causal relationship between price and demand and does not take into account other factors influencing demand that may bias the estimate. The price and quantity variations used to calculate elasticity are also specific to the time period and crop types, making extrapolation to other contexts risky. Finally, reporting errors and approximations in the quantities used can bias the estimate.

3.3 Econometric estimation

The estimation is carried out by regressing four different dependent variables − total pesticides, herbicides, insecticides and fungicides use per hectare in Cameroon − on a set of explanatory variables including the pesticide import price, the price of agricultural products and the unemployment rate (see Sect. 3.2).

3.3.1 Stationarity and cointegration

Stationarity means that the statistical properties of a time series i.e. mean, variance and covariance do not change over time. Ordinary Least Squares (OLS) model requires the series to be stationary to make estimations that are valid on the long-run. The stationarity properties of the series considered were tested with the Augmented Dickey-Fuller (ADF) test. All the time series expressed in log were found to follow a unit-root process, i.e., the mean and variance of the series increase over time and the series does not converge to a constant value, making it non-stationary.

When time series are non-stationary, cointegration is necessary to produce reliable longrun estimates. Cointegration implies that although variables may deviate from each other in the short term, there is a long-term relationship that relates them. We conduct the Engel Granger test to estimate the regression parameters of the long-run cointegration regression. This test determines whether there is a linear combination of these variables that is stationary. The OLS model was used to estimate the cointegration regression. The residuals were found to be stationary for all four regressions (p value <= 0.01 for the ADF test with one lag), thus confirming the cointegration relationship among the variables. The short-run relation was estimated with a Partial Adjustment Model (PAM) (Cuddington and Dagher, 2015). In this specification, the long-run level of pesticide use pu^* at each time t is:

$$p{u}_t^{*}={\beta }_0+{\beta }_{1}p{p}_t+{\beta }_{2}a{p}_t+{\beta }_{3}u{r}_t+{ϵ}_t$$(2)

With pp corresponding to the pesticide price, ap to the price of agricultural products, ur to unemployment rate and ϵ being the error term. The actual use pu_t adjust gradually towards the long-run value pu^* with speed of adjustment λ according to the following equation:

$$p{u}_t=p{u}_{t-1}+\lambda \left(p{u}_t^{*}-p{u}_{t-1}\right){}_{}$$(3)

λ reflect inertia, which may be related to habits of using a certain amount or type of product, as well as short-run disconnection between wholesale and retail pesticides prices. Substituting equation (2) in equation (3) yields:

$$p{u}_t=\lambda {\beta }_0+\lambda {\beta }_{1}p{p}_t+\lambda {\beta }_{2}a{p}_t+\lambda {\beta }_{3}u{r}_t+\left(1-\lambda \right)p{u}_{t-1}+{\nu }_t$$(4)

Following this specification, the short-term elasticities to the price of pesticide and agricultural products equals respectively λβ₁ and λβ₂ and the long-run elasticities equals β₁ and β₂. As 0 < λ < 1, the PAM specification implies that short-run elasticities are less than long-run ones by design.

3.3.2 Small sample size

Our sample consists of only 24 observations covering the period from 1995 to 2018. This sample size is insufficient to ensure the robustness of an ordinary least squares (OLS) regression. In particular, key OLS assumptions, such as the normality of residuals, are more difficult to assess with a small sample. Moreover, classic significance tests for the estimated coefficients (t test) become less reliable due to the limited sample size.

To address this problem, our strategy relies on two distinct approaches:

• We perform a block bootstrap for a more robust estimation of the standard errors and confidence intervals of the OLS regression. Block bootstrap is a statistical resampling method particularly suited to time series, to generate a distribution of the estimates. This method is recommended for small sample sizes because it provides a non-parametric way to estimate the variability and uncertainty of statistical estimates without relying on large-sample approximations or strong parametric assumptions. In our case, we use block sizes of 6 years with 256 iterations (maximum number of possible permutations) ;

• Additionally, we use a Bayesian estimation method, a statistical approach that combines prior knowledge with observed data to estimate the probability distribution of the parameters. This method is better equipped to deal with small sample sizes, as it incorporates additional information through the prior (McNeish, 2016).

3.3.3 Bayesian estimation

Bayesian estimation consists in generating an a posteriori distribution of the parameters to be estimated from an a priori distribution by following Bayes’ rule:

$$p\left(\theta \text{|}y\right)=\frac{p\left(y\text{|}\theta \right)}{p\left(y\right)}$$(5)

where θ is the coefficient to be estimated and y is the dependent variable. In the equation (5), p(θ) corresponds to the initial beliefs about the value of the parameter drawn from the a priori distribution, p(y|θ) to the likelihood, i.e. the function that measures how well a set of parameters explains the observed data, and p(θ|y) to the a posteriori value of the parameter. The a posteriori distribution is produced numerically using Markov Chain Monte Carlo (MCMC) methods with the stan_glm function from the "rstanarm" package in the R statistical software. The MCMC sampling is based on 2000 iterations.

As Bayesian estimation provides an a posteriori distribution of parameters rather than a point estimate, the results are presented as the mean of the a priori distribution including the confidence interval (ʽcredible interval’) corresponding to the 2.5% and 97.5% values of the a posteriori distribution.

In the case of a small sample, the choice of priors must be carefully considered in order to take full advantage of the properties of Bayesian estimation. In this regard, McNeish (2016) recommends consulting meta-analyses. Böcker and Finger (2017)’s meta-analysis finds a median estimate of the own-price elasticity of −0.18, which is close to the value obtained with our OLS regression (see Sect. 4). For this reason, we have chosen to use the OLS regression estimates as priors for the Bayesian estimation, because we believe that they may be more accurate in the Cameroon context.

The stan_glm function offers limited flexibility in the choice of the error distribution. Given our data, only a Gaussian distribution is available. As discussed earlier, this assumption is problematic due to our small sample size. To partially address this issue, we use priors following a Student’s t- distribution (with 1 degree of freedom) to make the estimation more robust to extreme values.

3.3.4 Heteroscedasticity and multicollinearity

The presence of heteroscedasticity results in biased estimates of standard error and in efficient estimates (i.e., their variance is not the lowest of all other unbiased estimators). This may affect statistical tests of significance and lead to biased inference. No significant heteroscedasticity is detected for the regression with herbicides and fungicides use as dependent variable (p-value = 0.131 and 0.090). However, the null hypothesis of homoscedasticity is rejected for total pesticides and insecticides (p-value = 0.035 and 0.014). To address this issue, we use a Bayesian estimation approach with priors drawn from a Student’s t-distribution. This approach allows for more robust estimates of the model parameters, as the Student distribution, with heavier tails, mitigates the influence of outliers and accounts for potential heteroscedasticity in the error terms.

Multicollinearity, which may affect the stability of individual predictor, is tested with the variance inflation factor (VIF). The VIF calculates how much the variance of a regression coefficient is inflated due to multicollinearity. Except for the short-run regression on herbicides, no critical level of multicollinearity was detected.

4 Results

4.1 Local estimates from survey data

Using the survey data, we estimate an average own price elasticity over the four crop types at −0.341, with a standard error (σ) of 0.649 (i.e., not significantly different from zero). However, we find notable difference among crops: tomato producers exhibit complete inelasticity to pesticides price (ϵ = 0), while rice producers display the highest price elasticity, with a mean estimate of −0.601 but with great variability among farmers (σ = 1.151). Maize and Cacao producers fall in an intermediate range, with own-price elasticities of −0.373 (σ = 0.515) and −0.345 (σ = 0.379) respectively.

These local empirical estimates, which account for pesticide prices inclusive of subsidies, suggests that in the short term, pesticides use is relative inelastic to their own price: a 1% increase in the price of pesticides results in only a 0.34% decrease in consumption. To explain this low sensitivity, farmers point out that they do not have alternatives that are as effective as chemical pesticides. For instance, when asked to prioritize among three land-clearing and maintenance technologies − herbicide, machete, and brushcutter − 74% indicated a preference for herbicides, even under rising price scenarios. This choice is primarily attributed to (i) the scarcity of paid labor for machete use, and (ii) the high cost of such labor when available.

4.2 Ordinary least squares estimation

The results of the short-run OLS estimation are shown in Table 4. The confidence intervals estimated using the block bootstrap method show that most of the estimates are not significantly different from 0. This results probably from the lack of statistical power due to the small sample size. Only the lagged variables for insecticide and fungicide use are statistically significant. Therefore, long-run elasticities are estimated only for these two dependent variables (Tab. 5).

We find a negative relationship between the use and price of pesticides, herbicides, insecticides and fungicides. Parameters values are greater than −1 both in the short- and long-run, indicating a weak elasticity of pesticide use to its own price. By construction, as a direct consequence of the partial adjustment model used to estimate the short-run relationship (see Sect. 3.3.1), long-run elasticities are necessarily larger than short-run ones.

4.3 Bayesian estimation

The results of the Bayesian estimation are presented in Table 6. The Monte Carlo Standard Error (MCSE) is below 0.01 for all estimated parameters across the four tested relationships, and the R^̂ statistic is systematically close to 1, indicating good convergence. The posterior mean estimates are overall close to the OLS estimates.

The price of pesticides is significant at 5% for all categories, except herbicides. Since lagged variables are statistically significant, long-term elasticities can be consistently estimated (see Tab. 7). In the long term, herbicides are the most price elastic pesticide category and insecticides the least. All values remain above −1 confirming that there is little elasticity between pesticides use and their own price in Cameroon.

The price of agricultural product is not significant for all pesticide categories, suggesting that pesticides use is inelastic to the output price. This result is consistent with interviews conducted with farmers as part of the field survey, in which they stated that they did not modify their pesticide consumption in accordance with anticipated prices of agricultural products.

The unemployment rate is significant at 5% or 10% for all pesticides categories except herbicides. The value of the parameter is close to or even below −1, indicating some elasticity of pesticide use to labor availability. The non-significance of the parameter for herbicides prevents us from validating the assumption that herbicides are used as a labor substitute to a greater extent than insecticides and fungicides, as they are cheaper and more readily available than labor for hand weeding (Gianessi and Reigner, 2007; Köhler, 1979; Sharma et al., 2018).

4.4 Robustness checks

The estimation was conducted from 1995 to 2018, which is the period over which we were able to cross-check the FAOSTAT and GloPUT data. Nevertheless, 1995 represents a price peak compared with 1996 and even more 1994. The dynamics of pesticide prices between the two periods are markedly disparate: a decline of −2.7% per year between 1995 and 2018, in contrast to a decline of −1.4% per year between 1994 and 2018. To test the robustness of our results, we re-estimate the short-term Bayesian model over the period 1994–2018 instead of 1995–2018. The results of the robustness test, as depicted in Figure 5, show slightly higher significance levels for the own-price elasticity in the estimation over 1994–2018, particularly for herbicides, where the price of pesticides becomes significant. Regarding the price of agricultural products, the results are broadly similar with both time period. Note that the parameter becomes significant in the estimation over 1994–2018 for insecticides and herbicides.

Finally, we assess the influence of a time trend on the results by estimating a model that includes a trend at three levels, corresponding to the periods 1995–2001, 2002–2010, and 2011–2018. These three periods reflect distinct phases observable in pesticide use trends (Fig. 2), with two growth phases separated by a period of relative stability between 2002 and 2010. The results are presented in Table 8. In this specification, the price of pesticides is no longer significant for insecticides, but remain significant for total pesticides and fungicides. While the lagged variables remain highly significant, the unemployment rate is no longer significant for any pesticide category.

Fig. 5 Robustness test: value and 95% confidence interval of the elasticities of the 4 types of pesticide demand (total pesticides, herbicides, insecticides, and fungicides) estimated on 1994-2018 vs. 1995-2018 data. Own price elasticity (left) and elasticity to the price of agricultural products (right) Test de robustesse : valeur et intervalle de confiance à 95 % des élasticités des 4 types de demande de pesticides (pesticides totaux, herbicides, insecticides et fongicides) estimées sur les données 1994-2018 vs. 1995-2018. Elasticité par rapport à son propre prix (gauche) et par rapport au prix des produits agricoles (droite).

5 Discussion and conclusion

Estimating the economic determinants of pesticide use in Cameroon is challenging due to data limitations, including reliability issues, insufficient observations with relatively short time series, and incomplete information, especially regarding subsidy amounts. To address these obstacles, we employed several strategies: (i) comparing available pesticide use and trade data with a recent database using a data reliability assessment method; (ii) combining econometric methods with a field survey to better account for the potential effects of subsidies; and (iii) using a bootstrap and Bayesian estimation approach to deal with issues related to small sample size. While our results may not provide a definitive answer, we hope they can offer a valuable foundation for informing pesticide use regulation in Cameroon, especially in the context of a rapid increase in pesticide consumption and its associated environmental and health risks.

The results of our econometric estimation and the survey of farmers in Cameroon show that pesticide use is relatively inelastic to its own price, with elasticity values greater than −1. The elasticity estimated locally from the farmer survey, which accounts for prices inclusive of subsidies, is slightly higher than the econometric short-term estimate (–0.341 vs. −0.194). However, this result should be interpreted with caution, as the variance of the local estimate is substantial, and the estimation methods differ. Our econometric estimate, based on gross prices (before subsidies), may underestimate farmers’ true price sensitivity if their decisions are driven by net prices (after subsidies). If subsidies remain broadly proportional to price, the estimated elasticity remains valid. We find no significant response of pesticide use to agricultural product prices, but pesticide use appears to be more sensitive to a proxy for labor costs, although this result is not robust when a time trend is included. The use of prices relative to formulated products rather than active ingredients may also introduce inaccuracies, even if the data do not suggest a reduction in the concentration of active ingredient per unit of formulated product. Overall, the significance levels of our estimates are relatively low due to lack of statistical power, warranting caution in interpreting the results and drawing conclusions.

Böcker and Finger (2017)’s meta-analysis provides values for own-price elasticities from which we can compare our results. It is important to note that Böcker and Finger’s meta-analysis only includes values for the United States and Europe. The agricultural contexts are therefore not directly comparable with that of Cameroon, which is characterised by a tropical climate, agricultural production with an important share of perennial crops (cocoa, bananas, coffee) and a lower level of mechanisation. For the short-term, they give a median estimate of the own-price elasticity of −0.18, with an inter-quartile range from −0.10 to −0.30.

In comparison, our results (ϵPesticides = –0.194, ϵHerbicides = –0.176 not significant, ϵInsecticides = –0.136 and ϵ_Fungicides= –0.215) are relatively close to this median estimate. For the long term, Böcker and Finger (2017) find a median own-price elasticity of −0.39 with an inter-quartile range between −0.10 and −0.68. Our results for all pesticides categories (ϵ_Pesticides= –0.351, ϵHerbicides = –0.393, ϵInsecticides = –0.198, ϵFungicides = –0.346) are also in good accordance with this meta-analysis. On the other hand, contrary to Böcker and Finger, our analysis does not confirm that herbicides are more price elastic than other categories of pesticides.

From the perspective of public policy, our findings indicate that implementing a tax on pesticide prices could potentially influence the reduction of pesticide use, although the impact would likely be limited. To achieve a notable reduction, the tax would have to be set at a relatively high level. This conclusion aligns with Böcker and Finger (2016), which suggest that, in the context of European agriculture, pesticide taxes are only effective if set at sufficiently high levels. However, the effectiveness of the tax remains controversial, as Skevas et al. (2012), in the context of Dutch agriculture, show that even high tax levels result in only a modest reduction in pesticide use and lead to environmental spillovers. Finger et al. (2017) emphasize the varying price sensitivity across agricultural systems and propose a differentiated tax scheme, so that more risky pesticides are taxed at higher rates. Our results do not allow us to shed light on this debate because we do not have data according to pesticide risk level and crop type. This is an important limitation of our study.

More fundamentally, the key recommendation of this study is to enhance the reliability of Cameroon’s information system on pesticide use, particularly to better document subsidies granted to farmers. Since 2019, the government (Ministry of Agriculture and Rural Development) has implemented a tool called "Manuel de procédures de subvention des intrants et équipements agricoles productifs au Cameroun". This manual provides comprehensive guidelines on financing mechanisms for pesticides. However, its dissemination among farmers remains limited and it is not completed by a statistical monitoring system to track the amount of pesticide subsidies granted. Our analysis indicates that the quality of data on pesticide use and imports is of a high standard. It is therefore crucial to maintain this quality to enable more robust estimates in the future.

Our preliminary results were presented at a workshop held in Yaoundé, bringing together participants from a variety of backgrounds (national and international public institutions, academics, professional organizations, civil society and business). During this event, the question of pesticide quality has been the subject of much debate. It has been reported that some Cameroonian farmers need twice as much pesticides for the same result. The comparison of import data expressed in formulated products and use expressed in active ingredients, as conducted in this study, did not reveal any notable decline in product concentration. A decline in the quality of pesticides, expressed in terms of the concentration of active ingredients, is therefore unlikely to be the cause of the decline in import prices highlighted in this study. On the other hand, China, which now accounts for three-quarters of Cameroon’s pesticide imports, has developed significant production capacity for off-patent pesticides, such as glyphosate (Werner et al., 2022). The price decline could therefore be explained by an increasing share of imports of older generation products, which are associated with greater resistance and lower efficacy.

The insignificant or positive response of pesticide use to the prices of agricultural products, measured in both econometric methods, is difficult to explain from an economic point of view. Considering an aggregate index of crop prices without distinguishing pesticide use by crop type introduces a degree of imprecision that may explain the observed result. This approach does not capture the potential shifts in production choices between different crops. Also, our survey among farmers revealed that, with the exception of cocoa, they are poorly informed about market prices and therefore do not incorporate them into their decision-making regarding pesticide use.

The increase in rainfall observed in Cameroon since the early 1990s could also have an influence on pesticide use, by promoting the growth of fungal diseases and diluting the concentration of pesticides in plants (Frederic and Mesmin, 2017). Adding biophysical variables to the econometric estimation model to capture this effect could thus be a possible model improvement. More fundamentally, econometric measures of elasticities have the disadvantage, when it comes to assessing the impact of long-term policies, of being based on a constant production technology (Carpentier and Weaver, 1997; Lansink and Carpentier, 2001). The low elasticity of pesticides mainly reflects the absence of economically viable alternatives within a given technical framework, and its magnitude could vary significantly within another technical framework, for example one that includes a higher proportion of agroecological practices.

Acknowledgments

The research leading to these results received funding from Agropolis Fondation and Fondation FARM under the Pesticide Reduction for Tropical Agricultures project (PReTAg). Authors acknowledge fundings from the French Agence Nationale de la Recherche within the CLAND project (ANR-16-CONV-0003).

Authors wish to thank Franck Nadaud, Jean Joel Ambagna, Armel Awah, Lydie Bamou, Frederic Ngwack, Clémence Ngah Abanda, Cyrille Dominick Bitting and Nathalie Jas for their helpful suggestions to improve this paper. We are grateful to the three anonymous reviewers for their constructive comments and suggestions, which helped improving the quality of this paper.

Data availability statement

Code and data used in this paper are available on the following GitHub repository: https://github.com/thierryb2020/Pretag-Pesticide-Assessment.

References

All Tables

All Figures

	Fig. 1 Origins of pesticide imports into Cameroon in 1995 and 2018. Source: Shattuck et al. (2023) Origine des importations de pesticides au Cameroun en 1995 et 2018.
In the text

	Fig. 2 Pesticide use in Cameroon in kilograms of active ingredients per hectare (kg AI/ha) Usage des pesticides au Cameroun en kilogrammes de matières actives par hectare (kg MA/ha).
In the text

	Fig. 3 Pesticide price (real) in Cameroon Prix des pesticides (valeurs réelles) au Cameroun.
In the text

	Fig. 4 Unemployment rate (left) and price of agricultural products (right) in Cameroon Taux de chômage (gauche) et prix des produits agricoles (droite) au Cameroun.
In the text

Fig. 5 Robustness test: value and 95% confidence interval of the elasticities of the 4 types of pesticide demand (total pesticides, herbicides, insecticides, and fungicides) estimated on 1994-2018 vs. 1995-2018 data. Own price elasticity (left) and elasticity to the price of agricultural products (right) Test de robustesse : valeur et intervalle de confiance à 95 % des élasticités des 4 types de demande de pesticides (pesticides totaux, herbicides, insecticides et fongicides) estimées sur les données 1994-2018 vs. 1995-2018. Elasticité par rapport à son propre prix (gauche) et par rapport au prix des produits agricoles (droite).

In the text