Analysing the farm level economic impact of GM corn in the Philippines

Abstract

This paper analyses the farm economic viability of genetically modified (GM) corn in the Philippines. Data was collected from 114 farmers in Isabela province including non-GM, Bacillus thuringiensis (Bt), herbicide tolerant (HT) and BtHT corn farmers. Results of univariate analysis showed that non-GM corn was not statistically different from Bt, BtHT and HT corn in terms of production output, net income, production-cost ratio and return on investment. Multivariate econometric analysis for the agronomic input variables showed a higher return on investment for Bt corn as the only significant difference between seed types. Next, pest occurrence and severity variables were included in the regression to address endogeneity. The Blinder-Oaxaca decomposition method was used to further investigate differences between growers of BtHT corn and non-GM corn into an endowment and a coefficient effect. The decomposition analysis showed that BtHT corn has a negative impact on return on investment as revealed by the negative signs of the overall mean gap and the characteristics and coefficient components. In contrast, the overall mean gap for net income indicated that adopting BtHT corn could potentially increase non-GM growers’ income mainly from better control of corn borer pest even though mean levels of borer occurrence are lower for non-GM growers.

Keywords:

• corn
• genetically modified crops
• farm economics
• smallholder farms
• decomposition analysis
• Philippines

1 Introduction

The adoption of genetically modified (GM) corn, cotton and soybean improves yields and reduces pesticides usage (e.g., [1], [2], [3], [4], [5] and [6]). Recent meta-analyses by Finger et al. [7] and by Areal et al. [8] of GM corn, cotton and soybeans provides evidence that these crops lead, on average, to a higher economic performance than conventional crops. Other studies have confirmed these higher averages for specific countries and specific crops. For developing countries the 2007 farm income gains from Bacillus thuringiensis (Bt) and herbicide tolerant (HT) corn were estimated at $302 and $41 million, respectively [9].

Contrariwise, the results reported in studies such as those listed above remain a matter of great controversy as noted as early as 2000 by Zadoks and Waibel [12]. The performance of GM crops is variable, socio-economically differentiated, and contingent on a range of agronomic and institutional factors [11], [14], [15], [16]. There is particularly a need for further evidence on the benefits for resource-poor farmers. The most obvious pecuniary benefit for resource-poor farmers is increase in yield [7], [11] and [13]. But results from field trials or partial analysis of gross returns and/or cost for pest control do not necessarily imply that farmer's incomes and welfare are improved.

In this paper we focus on the Philippines, the first and so far only country in Asia to have approved the commercial cultivation of GM corn. After Bt corn was first commercialized in the Philippines in 2003, there was a dramatic increase in its adoption. By 2010, GM corn was grown on over a quarter million hectares by 270,000 Filipino farmers [19]. Since GM corn seeds cost are higher than that of the available commercial iso-hybrid corn in the market, high income and large-scale farmers were the first adopters of this technology. More recently small-scale and poor farmers have also adopted the technology.

In an earlier study we investigated Filipino small-scale farmers’ reasons for adoption and their experiences growing GM corn [21]. Results from this survey work showed that these farmers were motivated to switch to GM corn for economic reasons (perceived yield increase, better insect control, reduced costs of inputs) and also out of curiosity. The experience of using GM corn however led to statistically significant changes in the respondents’ opinion. Sixty-eight percent of the Bt corn respondents did not agree that their economic status had improved after they had started using the technology or that Bt corn is worth investing in. A significant number of Bt respondents now perceived negative effects of Bt corn on farmers’ economic status. Of the BtHT adopters 21% and 29% said they had changed their standpoint to disagreement in regard to the statements that BtHt corn is worth investing in and could improve the lives of farmers, respectively.

These findings motivated us to a conduct a comparative assessment specifically for small-scale farmers in the Philippines. The changes in farmers’ opinion after using GM corn justify a more in-depth economic assessment of the GM technology under the conditions and skill level of these farmers. In addition, previous studies on the economic impact of GM corn in the Philippines focused only on Bt corn [16], [17] and [44]. BtHT and HT corn are now also widely grown in the Philippines. Hence in our paper we compare conventional, Bt, HT and BtHT corn.

We focus at the farm level and at the variability across farms/farmers. We take explicitly into account that genetically modified (GM) corn seed is substantially higher in price and hard to afford by a resource poor farmer [22]. This price can be up to 84% higher than for non GM-corn depending on the type and number of transgenic traits included in the seed. Thus, to deal with the farm economic issues we seek to answer the research question: Is GM corn more economically viable and worth the investment than non-GM corn at the farm level? We investigate farm level differences by corn variety in expenditure for agricultural inputs (labour, seed, and fertiliser costs), gross and net return, production-cost ratio and return on investments. Following previous studies on profitability [17] and adoption [23] and [24] of GM corn we use econometric analysis to evaluate if and how agronomic variables (i.e. labour costs, agricultural inputs, corn types and farm area) affect production cost, total return, net income, production-cost ratio and return on investment.

This paper further contributes to the discussion by employing the Blinder-Oaxaca decomposition method [25] [26] to decompose the observed differences in economic performance between GM adopters and non-adopters into two components, namely a characteristics effect and a coefficients effect. This decomposition technique is widely used in economic applications to study mean outcome differences between groups [27], [28] and [29]. The counterfactual exercise answers the question, what would happen to the GM adopters if their distribution of characteristics was as for the non-GM adopters but if they maintained the returns to their characteristics? A comparison of the counterfactual and estimated performance distribution for the GM group and the non-GM group yields the part of the performance difference that is attributable to differences in covariates (farm and farmer endowments). The remainder of performance difference is then attributable to differences in returns to covariates. To the best of our knowledge, no other study has employed this decomposition technique to investigate the GM-economic impact nexus.

2 Material and methods

2.1 Area description: GM Corn and the family farm in the Philippines

The Philippines has a total of 9.6 million hectares (32%) agricultural land area of which 51% and 44% are arable and permanent croplands, respectively [18]. There are ∼1.8 million corn farmers in the country and 60% of these cultivate yellow corn. Mostly, these farmers are categorized as small, semi-subsistence farmers with a farm area of less than 4 hectares [30]. All corn in the country is grown on rain fed non-irrigated land. The cornfields of these small farmers are mostly situated in marginal places. In contrast, most of the large-scale plantations of yellow corn are found in well-situated lowland or upland areas.

Small-scale farmers plant one corn variety, sometimes intercropped with tobacco, fruits (pineapple) and vegetables. Post-harvest activities include de-husking, shelling and grain drying which is done manually by both family and hired labour. Harvested corn is sun-dried immediately after harvest [30]. The small-scale farmers are dependent on trader-financiers for full-season input financing because they lack the necessary capital. Farmers repay their loans with a certain interest (∼7-15%) either in cash or in corn product upon harvest. The trader-financier decides on the terms of condition of the payback agreement. For large-scale farmers that have large cornfields (cornfield size of more than 3 hectares) hired labour and mechanized farming are common practices.

Among the sixteen regions in the Philippines, the Cagayan Valley region ranks first in terms of corn production. Isabela province in the Cagayan Valley region was chosen as the case study area for the farm level economic assessment. In Isabela province average yield of Bt corn per ha was reported to exceed yield of conventional corn by up to 33% in the 2003-2005 seasons. In 2008-2009, Bt and BtHT corn yields surpassed conventional corn by 4-5% and by 13-22%, respectively [20]. Recall that the 2003/2004 crop year is the first year that Bt corn became available to Philippine farmers. Hence, data from 2003-2005 crop years gives information about the “initial” impacts of Bt corn [16]. The surplus yield of BtHT corn compared to Bt corn for 2008-2009 was due to its combined traits which resolved problems due to both Asian Corn borer and weed pest severity.

In Isabela province, farm demonstrations showcased the advantages of using GM corn including both its pecuniary and non-pecuniary benefits. One of the non-pecuniary benefits of GM corn, especially of BtHT (insecticide plus herbicide tolerant) corn is that less labour inputs are required for weed management. With proper spraying of herbicide, the weed problem can be reduced or totally controlled. However many poor farmers may still resort to manual weeding in BtHT corn employing the labour force of the (extended) family on a cooperative basis as they cannot afford to buy herbicides [21].

2.2 Survey

The survey was conducted from October to December 2010 to obtain data for the wet growing season. In order to select our respondents within the group of general farmers who were best able to give us the first-hand information we needed, we applied a purposive sampling technique. Purposive sampling was accomplished of 114 corn farmers in the province of which 42, 8, 44 and 20 were non-GM, Bt, BtHT and HT corn adopters, respectively. Ninety-percent of the respondents were classified as small-scale farmers with farm sizes of not more than 3 ha. Only 10% of the respondents were large-scale farmers with farm sizes of 4 to 8 ha. Almost all farmers (98%) in the sample practiced mono-cropping and 25% was female. A self-structured questionnaire was used during the face-to-face interview of the respondents who were from 10 municipalities and 33 villages of the province.

The respondent did not differ significantly in age, household size, residency in their respective municipality, and level of education (Table 1). All farmers encountered weed problem but their level of concern varies. Further analyses revealed large differences between non-GM and Bt farmers in encountering weed and Asian corn borer (ACB) problems and likewise for the level of concern about this problem in their fields (Table 1). All non-GM correspondents confirmed to have encountered the ACB problem in the wet growing season of 2010 while only part of the Bt farmers had observed ACB problems in their fields.

Table 1 Respondents’ information and farming background. Similar superscript letters represent no differences between corn varieties at P < 0.05 after post-hoc analysis using Bonferroni test.

Respondents’ Information	non-GM (n = 42)			Bt (n = 8)			BtHT (n = 44)			HT (n = 20)			F-value	Sign
	Mean	±	sd	Mean	±	sd	Mean	±	sd	Mean	±	sd
Age	49.29	±	8.97	43.38	±	8.90	46.84	±	12.49	49.50	±	12.34	0.939	0.424^ns
Household size	6.43	±	3.12	5.00	±	0.54	5.70	±	2.46	6.30	±	2.64	0.980	0.405^ns
Years residing in the area	41.95	±	16.88	34.63	±	18.10	41.00	±	15.89	43.35	±	16.91	0.560	0.643^ns
Highest educational attainment^1	3.93	±	1.96	4.75	±	1.91	4.00	±	1.95	4.70	±	2.20	1.040	0.378^ns
Cornfield information					±			±			±
Area of farm devoted to the new variety^2	1.33	±	1.36	1.56	±	0.82	2.06	±	1.43	1.85	±	1.44	2.104	0.104^ns
Weed problems^3	0.71	±	0.46	0.75	±	0.46	0.59	±	0.50	0.60	±	0.50	0.657	0.580 ^ns
Concerns on weeds pest incidence^4	2.61^a	±	0.67	3.50^b	±	0.93	2.91^ab	±	0.94	2.95^ab	±	0.97	2.794	0.044*
Asian corn borer (ACB) problem^3	1.00^a	±	0.00	0.71^b	±	0.49	0.82^ab	±	0.39	0.89^ab	±	0.32	3.431	0.020*
Concerns on ACB incidence^4	3.12	±	0.83	2.75	±	1.39	3.07	±	0.77	2.95	±	0.83	0.517	0.671 ^ns
Severity of ACB damage^5	3.72	±	1.02	3.13	±	1.55	3.42	±	1.03	3.40	±	1.23	0.968	0.411 ^ns

¹Scale: 1-No schooling; 2- Elementary level; 3- Elementary graduate; 4- High School Level; 5-High School graduate;

²Scale: in hectare

³Scale: Yes-1, No-0

⁴Scale: 5- Highly concern; 4- moderately concern; 3- concern; 2-unconcern; 1- Highly unconcern

⁵Scale: 5- Highly severe; 4- severe; 3- moderately severe; 2- negligible; 1- Highly negligible

Input cost covered in the survey included the payment for the seeds, fertilisers, pesticides and other expenses entailed from land preparation to post harvest (Table 2). The labour cost encompasses the labour service fee for man-machine day, man-animal day and man day entailed during land preparation and cultivation practices (ploughing, harrowing, furrowing, off-barring and hilling-up), chemical application (fertiliser application and spraying of insecticide and herbicide) and pre- and post-harvesting practices (seed planting, harvesting, threshing, hauling and drying) for the 2010 wet growing season. The service fees for man day include both paid labour (hired labour) and non-paid labour (labour by family members). The corresponding wage per farming practice (e.g. harvesting, spraying) employed was calculated by multiplying the number of labourers to the existing standard service fee given per labourer per day (e.g., harvesting cost = 10 persons [paid and unpaid labourers] x $4.65 per man day). Related research of the sample [21] found that reducing the labour requirement was not a major adoption argument for the growers of BtHT and HT corn. This highlights, as mentioned above, that poor farmers may still resort to manual weeding of BtHT and HT corn because they cannot afford the requested herbicide.

Table 2 Cost of production (labour, agricultural inputs, other expenses. Values are in US$ per ha at 1US$:42.50 Philippine pesos). Similar superscript letters represent no differences between corn varieties at P < 0.05 after post-hoc analysis using Bonferroni test. P values: *** = p < 0.001, ** = p < 0.01, * = <0.05, (*) = <0.10).

Production cost	non-GM (n = 42)			Bt (n = 8)			BtHT (n = 44)			HT (n = 20)			F-value	p-value
	Mean		sd	Mean		sd	Mean		sd	Mean		sd
Total Production Cost	504.70^b	±	138.68	669.92^a	±	70.67	633.39^a	±	110.94	603.12^a	±	115.95	9.982	0.000***
1.Total labour cost	208.60	±	52.12	231.90	±	61.90	228.36	±	47.58	220.17	±	39.61	1.336	0.267 ^ns
Ploughing	42.13	±	20.91	38.09	±	13.69	36.47	±	16.82	37.41	±	14.44	0.776	0.510^ns
Harrowing	8.64	±	7.44	7.53	±	11.77	9.85	±	10.58	10.91	±	11.96	0.362	0.780^ns
Furrowing	1.34	±	6.84	2.94	±	8.32	1.30	±	3.48	1.59	±	4.04	0.220	0.882^ns
Second harrowing	1.01	±	6.54	2.94	±	8.32	1.12	±	3.64	0.00			0.664	0.576^ns
Basal application	4.91	±	4.98	9.03	±	4.93	5.50	±	4.44	7.34	±	6.52	2.165	0.096(*)
Side dress	5.90	±	4.96	6.62	±	3.44	7.85	±	5.29	6.55	±	6.01	1.025	0.385^ns
Planting	25.04	±	10.37	24.82	±	7.41	29.39	±	10.06	28.05	±	15.14	1.264	0.291^ns
Off-baring	8.10	±	7.97	8.59	±	9.55	5.88	±	8.46	3.79	±	7.84	1.496	0.220^ns
Hilling-up	3.03	±	5.89	4.12	±	8.03	2.98	±	6.71	1.76	±	4.51	0.336	0.799^ns
Total spraying	8.49	±	9.18	8.65	±	7.48	8.94	±	7.72	5.89	±	4.31	0.736	0.533^ns
Insecticide Spraying	2.87	±	3.97	2.35	±	4.53	2.86	±	4.93	1.06	±	2.70	0.990	0.400^ns
Herbicide spraying	5.17	±	6.85	6.29	±	6.00	5.60	±	5.76	4.84	±	4.70	0.150	0.929^ns
Fungicide Spraying	0.45	±	1.82	0.00			0.48	±	2.25	0.00			0.478	0.698^ns
Harvesting	30.84	±	8.67	37.06	±	14.38	33.88	±	6.98	36.29	±	10.60	2.344	0.077(*)
Thresher	35.30	±	13.65	39.58	±	16.52	39.84	±	13.96	42.49	±	16.16	1.338	0.266^ns
Hauling	14.23	±	9.94	18.85	±	7.67	19.46	±	12.86	17.11	±	7.76	1.770	0.157^ns
Drying	18.84	±	10.36	20.32	±	10.41	22.79	±	10.57	19.48	±	8.92	1.177	0.322^ns
2.Total input cost	262.84 ^b	±	106.92	393.68 ^a	±	79.84	369.10 ^a	±	88.46	354.51 ^a	±	88.08	11.071	0.000***
Seed costs	90.01^b	±	32.46	152.35^a	±	33.55	163.71^a	±	32.11	168.00^a	±	26.14	49.120	0.000***
Total Fertiliser	165.44	±	88.71	238.38	±	79.93	198.73	±	81.08	181.86	±	72.57	2.309	0.080(*)
Organic	126.86	±	12.85	157.12	±	29.79	148.13	±	12.70	156.06	±	18.39	0.821	0.485^ns
Inorganic	36.59	±	85.09	81.26	±	68.39	50.60	±	77.41	34.89	±	79.07	0.886	0.451^ns
Pesticides	7.39	±	8.57	2.94	±	8.32	6.67	±	11.73	4.65	±	8.71	0.678	0.567^ns
3.Others expenses	33.26	±	51.79	44.35	±	26.27	35.93	±	27.10	28.45	±	26.58	0.380	0.768 ^ns

Prior to employing statistical analyses of the data, the total cost of production, gross income or production output, net income, production-cost ratio and return on investment were computed in US$ per hectare (Table 3). Production output is the total yield in kg of the 2010 wet season multiplied by the prevailing prize of corn grain per kilogram. The PO values shown in Table 3 reflect farmers’ gross revenue i.e. total amount of yield in kilogram per hectare multiplied by the prevailing prize of corn per kilo in their respective municipality which ranged from 0.25 to 0.28 US$ per kilo. Net income (NI) was calculated by subtracting total cost of production (TCP) from production output (PO). The production-cost ratio (M) was computed as the quotient of the production output and the total cost of production per hectare (M = PO/TCP). Finally, the return on investment (RoI) was calculated as the difference between net income and the product of interest rate (IR) paid on loans and the total cost of production (TCP) i.e. RoI = NI-(IR x TCP). The IR varies per farmer depending on their current financial capability and agreement with the financier/middleman and it ranges from 7-15% on loans to pay for seasonal inputs. There is no uniformity in interest rates across Isabela province or across seed type. Farmers differ in IR due to variations in their economic status. For farmers who finance their own variable inputs, zero (0) IR was used in the computation. The high average RoI for Bt corn growers is because of lower interest payments (IR x TPC) as Bt farmers on average are more economically capable and are borrowing less money.

Table 3 Production output, net income, production-cost ratio and return on investment between corn types categories using univariate analysis. (Values are in US$ per ha at 1US$:42.50 Philippine pesos).

	non-GM (n = 42)			Bt (n = 8)			BtHT (n = 44)			HT (n = 20)			F-value	p-value
	Mean	±	sd	Mean	±	sd	Mean	±	sd	Mean	±	sd
Production Output (PO)	1,103.98	±	539.36	1,071.84	±	455.98	1,299.17	±	372.12	1,272.12	±	442.09	1.671	0.177^ns
Net Income (NI)	612.28	±	489.98	436.12	±	456.77	687.54	±	345.17	684.21	±	410.63	0.940	0.424^ns
Production-cost ratio (M)	2.28	±	1.01	1.698	±	0.70	2.158	±	0.59	2.208	±	0.71	1.231	0.302^ns
Return on Investment (RoI)	503.23	±	341.66	885.64	±	676.05	618.39	±	417.93	572.16	±	424.53	2.046	0.112^ns

2.3 Univariate and multivariate analysis

A univariate analysis was first employed to evaluate differences on the respondents’ information, farming background and production cost and to deal with the single response variables (i.e. corn types, agronomic inputs). A Holm-Bonferroni post hoc test [31] was used to assess significant differences of the responses between GM and non-GM adopters.

While the means for production cost, total return, net income, production-cost ratio and return on investment provide the realistic farm economic result of the corn types under farm conditions, a comparison of these means by seed type would be misleading. A correct comparison needs to account for the fact that it is not just the corn type that differs but at the same time many other agronomic inputs and farm characteristics as well. This confounds the impact of seed type on the economic results.

Multivariate analysis was used to assess how production output (PO), net income (NI), cost-production ratio (M) and return on investment (RoI) by seed type are directly or indirectly affected by other agronomic input variables. For comparison of the individual response variable between corn types the following conventional production function specification was estimated:(1) $y_i=\alpha +{\beta }_n{X}_{ni}+{\epsilon }_i$(1) where y_i denotes the response variables (i.e., the natural logs of PO, NI, M and RoI) in US$ ha⁻¹ of farm i; α is the intercept and x _ni is a vector of the natural logs of the explanatory variables 1…, n of farm i, including labour cost in US$ ha⁻¹, agricultural input cost (fertiliser, seeds or pesticides) in US$ ha⁻¹, area planted, corn type, and ɛ _i is the error term with the usual classical properties. The estimated model was formulated following the Cobb-Douglas production function approach of Yorobe and Quicoy [17] which is linear in the natural logs of the variables.

Starting from the full model for each the response variable, step-wise regression analyses were performed using the backward selection procedure. The full models included 16 types of labour cost, three types of input cost, and the area planted (see Table 2). Starting from the full model, each explanatory variable in this model was removed and an F-test performed comparing the full model with the model without the specific regressor (i.e. the reduced model). The regressor with the largest p-value (or smallest F value) was then removed and the result declared the current model. The process was then repeated. This iterative procedure led to our preferred estimated models. These models were selected on the basis of parsimoniousness in the number of predicators and a better fit as measured by the R² statistic. We present only the results from the final models. All econometric analyses were performed using R stat. version 2.12.2 [32].

2.4 Blinder-Oaxaca (BO) decomposition between GM and non-GM corn

The agronomic production function in eqn (1) above covers only part of the heterogeneity among the farmers that is expected to affect their input and output decisions. To proxy farmers’ individual production environment a common approach is to include addition variables in the production function. Particularly important in this context is that GM seed and pesticides are applied in response to pest problems. This can give rise to endogeneity of pesticide use decisions and seed type selection and thus inconsistent parameter estimates. Following Mutuc et al. [16] we included a pest occurrence and a severity variable in the production function to eliminate this potential bias.

Next, to the extended equations the Blinder-Oaxaca decomposition technique was applied to further investigate the mean differences in the response variables between GM and non-GM corn farmers. We assumed that GM corn has advantages compared to non-GM corn in terms of the responses because farmers will not shift to GM corn otherwise. Thus, we expect that the non-GM corn growers have a lower mean for the response variable compared to GM corn growers.

For the decomposition, the extended equation is estimated separately for two groups of farmers (by seed type):(2) ${\overline{y}}_{GM}={\overline{x}}_{GM}{\hat{\beta }}_{GM}+{\hat{\alpha }}_{GM}\text{for}{n}_{1}obs$(2) (3) ${\overline{y}}_{nonGM}={\overline{x}}_{nonGM}{\hat{\beta }}_{nonGM}+{\hat{\alpha }}_{nonGM}\text{for}{n}_{2}obs$(3)

Recall that residuals sum to zero in eqs (2) and (3). Next, the mean gap in performance between the two groups, ${\overline{y}}_{GM}-{\overline{y}}_{nonGM}$, is split into two parts:where ${\overline{x}}_{GM}$ and ${\overline{x}}_{nonGM}$ refer to the means of the explanatory variables, and α and β are the intercept and the coefficient estimates on the explanatory variables for the two samples, respectively. Equation 4 follows the proposed decomposition formulation of Neumark [33]. Subtracting and adding ${\overline{x}}_{nonGM}{\hat{\beta }}_{GM}$ to the right hand side of eqn. (4) and rearrangement gives the decomposition in the characteristics and coefficients effects. An alternative and equally valid formulation in eqn (4) multiplies differences in mean observables characteristics by difference in non-GM coefficient estimates and multiplies differences in coefficient estimates by GM mean observable characteristics.

In eqn. (4), the first term of the right-hand side is the part of the performance differential ‘explained’ by group differences in the predictors, i.e. the part of the gap attributed to differences in observed individual characteristics. The second term is attributable to differences in returns to co-variates; this is the unexplained “coefficient” part. It is important to recognize that this second term includes also all potential effects of differences in unobserved variables. In our case, it is the part of the gap that is due to different returns to the field characteristics and input levels. This second part answers the question if the growers of non-GM corn were to switch to GM corn overnight but nothing else observable changed (i.e. the field/farmers’ characteristics remained the same) would this lead to better results? A further detailed decomposition examines the percentage contribution of each individual explanatory variable to the total raw differential between the two samples to assess the comparative impact.

A decomposition of the mean gap as discussed above is only useful if the two compared equations are significantly different. Thus, first a Chow test for the difference between eqns. (2) and (3) is required; the null hypothesis is that the parameters of the two equations are equal, meaning that all the independent variables have uniform effects for both subgroups. The formula of the Chow test is:

(5) $F=\frac{\raisebox{1ex}{$\left(RS{S}_{Pooled}-\sum RS{S}_j\right)$}\!\left/ \!\raisebox{-1ex}{$k+1$}\right.}{\raisebox{1ex}{$\sum RS{S}_j$}\!\left/ \!\raisebox{-1ex}{$n{}_1+n_2-2k-2$}\right.}$(5) where RSS _pooled is the residual sum of squares (RSS) in the pooled regression, ΣRSS _j is the sum of the RSS from the two subgroup regressions, k is the number of predictor variables in the model and n ₁ and n ₂ are the number of observations in the subgroups [34]. The Chow test statistic follows an F-distribution with k + 1 and n ₁ + n ₂-2k-2 degrees of freedom.

3 Results and Discussion

3.1 GM vs. non-GM corn: Production Cost

The total cost of production (TPC) was obtained by summing up the variable cost for one hectare of corn production by category of corn farmer. Table 2 shows significantly lower mean total cost of production for non-GM corn. Agricultural input cost between GM corn types, i.e. Bacillus thuringiensis (Bt) vs. herbicide tolerant (HT) vs. BtHT, did not differ but all these GM corn types differed from non-GM corn. This corresponds to the large difference in seed cost between GM and non-GM corn. Seeds costs of all the GM corn types were more than 60% higher than non-GM corn. This study shows that cost of seeds per hectare was far higher for GM corn than for the leading conventional corn hybrids available on the market. This is also one of the main factors influencing the high level of total production cost for GM corn (Table 2). Aside from seed cost, there was a (less) significant difference among corn types in the cost of fertiliser with Bt corn having the highest fertiliser cost followed by BtHT, HT and finally non-GM corn (Table 2). The higher fertiliser cost for hybrid GM corn is consistent with the increased fertiliser input requirement to maximize the purported yield benefits when using any hybrid corn seed. Specifically, Bt corn requires a higher input of nitrogen-containing fertiliser especially at the early growth stage to enhance its δ-endotoxin production and the corn plant's ability to resist insect predation [45].

Total labour cost per hectare of production showed no difference between corn types. Reduction of pesticides usage is one of the benefits associated with using GM corn [3], [10], [16], [35], [36], and [37]. Our study showed however that the difference in pesticide cost across corn types was statistically insignificant. Our result confirms the findings of Afidchao et al. [21] that BtHT and HT farmers perceived no reduction in pesticides usage and exposure. This is in contrast to the US and Europe where GM corn reduced pesticide usage by as much as 14% [38] and where savings of $25-$75 per acre have been reported for insecticide use on Bt corn [37]. This reduction in pesticide usage was not observed in Isabela province. The explanation is that due to Bt farmers’ fear and anticipation of yield loss by pests other than Asian corn borer they might still spray insecticides even with Bt seed [21].

3.2 GM vs. non-GM: Production and Income

In terms of yield, our result for conventional and Bt corn was similar to the yield reported for 2004-2005 and 2007-2008 in the Philippine provinces of General Santos City and Isabela, respectively where conventional corn was statistically higher than GM corn [20].

Of the sample farmers, BtHT and HT corn growers out yielded non-GM corn growers by 8% and 7% but non-GM corn out yielded Bt corn by 1%. However, there was no statistical significant difference in production output between GM and non-GM corn. The computed averages showed that BtHT corn growers had the highest net income followed in descending order by HT, non-GM and Bt corn hybrids (Table 3). Average net income for BtHT and HT was higher than for non-GM corn by 7% and 5%, respectively and the average net income of non-GM corn growers was 5% higher than that of Bt growers. However, none of these differences were statistically significant. The lowest production cost ratio was observed for Bt corn; yet, this did not differ statistically from other GM corn types and was found to be not significantly different from non-GM corn. Finally, the return on investment (RoI) for Bt, BtHT and HT corn was 28%, 10% and 6% higher than for non-GM corn, respectively (Table 3). Yet, these differences were not statistically significant.

The results in Table 3 show that GM corn has no absolute straightforward overall advantages compared with non-GM corn. GM corn may produce higher yields [12], [39], [40] and [37] but additional points should be taken into account when assessing economic returns. As stated by Dilehay et al. [39] and Stanger & Lauer [40], Bt corn has higher grain moisture, lower test weight and higher harvest & seeds cost; these aspects counterweigh increased yield and might result in less or no benefits when using GM corn. Ma and Subedi [41] show that on the same maturity, non-Bt corn accumulates more nitrogen and leads to highest grain yield. According to Wolf and Albisser-Vögeli [42], low to moderate infestation of corn borer provides no advantage in using Bt corn and conventional maize hybrids are superior when appropriately grade-selected.

Past studies by Yorobe and Quicoy [17] and Gonzales et al. [20] stated that the Filipino farmers that adopted GM corn found it profitable, i.e. that farmers with high risks of Asian corn borer (ACB) damage have adopted GM corn by now. In the present study we find that with moderately severe ACB infestation as observed by the respondents (see Table 1); GM corn did not manifest a strong advantage in terms of profit.

3.3 Multivariate analysis

We applied production function analysis, using eqn. (1), to production output, net income, production-cost ratio and return on investment. Before the analysis, we first evaluated the residual plots (residual vs. fitted, normal Q-Q, scale-location and residual vs. leverage) for its normal distribution. Data that were non-normally distributed were ln (x + 1) transformed. Results presented are for the final models of the stepwise regression analyses.

In explaining the variation in production output, costs of threshing, harvesting and ploughing were found to have the largest effect. Among input cost, differences in seed cost seems to be important as expected from the summary statistics in Table 2. The R² value was estimated 0.53 for the final model (Table 4 column A). Table 4 (column B) shows the multi-agronomic variables explaining variation in net income with area planted and fungicide spraying being most important (R² = 0.39 for the final model).

Table 4 Estimates of agronomic variables identified to affect PO, NI, M and RoI ha⁻¹ employing stepwise regression analyses. All data was natural log (ln) transformed. P values: *** = p < 0.001, ** = p < 0.01, * = <0.05, (*) = <0.10); L_i= man labour cost ha⁻¹; I_i= agricultural input cost ha⁻¹; se = standard error.

	(A) Production Output (PO) [r^2 = 0.526]	(B) Net Income (NI) [r^2 = 0.391]	(C) Production-Cost Ratio (M) [r^2 = 0.386]	(D) Return on Investment (RoI) [r^2 = 0.053]
	Estimate ± se	Estimate ± se	Estimate ± se	Estimate ± se
Intercept
Corn types (Ci): Non-GM corn	3.698^** ± 0.743	-1181.230 ± 297.630	1.029 ± 0.262	503.190^*** ± 63.990
Contrast with intercept
-Bt corn	-0.011 ± 0.144	-193.200 ± 133.710	-0.085 ± 0.086	382.560^* ± 159.970
-BtHT corn	0.036 ± 0.103	-15.450 ± 77.290	-0.018 ± 0.050	115.220 ± 89.460
-HT corn	-0.002 ± 0.121	-34.550 ± 96.260	-0.037 ± 0.061	69.010 ± 112.660
Covariates
Ploughing cost (L1i)	-0.002^** ± 0.050
Furrowing cost (L2i)	-0.056^* ± 0.047	-92.860^* ± 41.630	-0.047^* ± 0.026
Second harrowing cost (L3i)	-0.124^* ± 0.058
Insecticide spraying (L4i)	0.081^* ± 0.031
Harvesting cost (L5i)	0.112^** ± 0.118
Thresher cost (L6i)	0.575^*** ± 0.081	125.050^(*) ± 77.030	0.286^*** ± 0.052
Side dress cost (L7i)		-66.590^(*) ± 31.130	-0.048^* ± 0.020
Fungicide spraying (L8i)		504.290^*** ± 82.010
Seed cost (I1i)	0.143* ± 0.125
Fertiliser cost (I2i)			-0.178^*** ± 0.041
Area planted (Ari)	0.077** ± 0.024	83.940** ± 25.000	0.045** ± 0.016

The relationship between production output (PO) and net income (NI) was strong and positive as shown in Table 5. Note however that the results in Table 4 reveal that the observed variance in production output and net income is attributed to distinct input variables. For net income fungicide spraying is the most important input variable, followed by the area planted. For production output this is threshing cost, also following by area planted.

Table 5 Correlation values (upper triangular part of the table) and p-values (lower triangular part of the table) between corn agronomic variables ha⁻¹. (PO= production output/yield; NI = Net income; M= Cost-production ratio; RoI= Return on investment; TPC= Total Production Cost). P values: *** = p < 0.001, ** = p < 0.01, * = <0.05, (*) = <0.10), ns = not significant.

Variables	Pearson's correlation coefficient
		PO	NI	M	RoI	TPC	Labour cost	Fertiliser cost	Seed cost	Pesticide cost	Area planted	ACB	Weeds
P- values	PO		0.898	0.724	0.096	0.425	0.344	0.237	0.273	0.006	0.300	-0.025	-0.031
	NI	< 2.2e-16^***		0.863	0.075	0.160	0.209	0.014	0.138	-0.017	0.287	-0.090	-0.016
	M	< 2.2e-16^***	< 2.2e-16^***		0.104	-0.014	0.001	-0.116	-0.007	-0.077	0.171	-0.067	-0.060
	RoI	0.309^ns	0.428^ns	0.271^ns		0.079	0.009	0.025	0.157	-0.064	0.000	-0.022	-0.109
	TPC	2.48e-06^***	0.090^(*)	0.035^*	0.401^ns		0.579	0.795	0.506	0.272	0.222	0.129	0.074
	Labour cost	0.000^***	0.026^*	0.994 ^ns	0.926^ns	1.5e-11^***		0.188	0.073	0.073	0.061	0.026	0.052
	Fertiliser cost	0.011^*	0.884^ns	0.002^**	0.794^ns	<2.2e-16^***	0.045^*		0.270	0.076	0.346	0.111	0.084
	Seed cost	0.003^**	0.144^ns	0.939^ns	0.095^(*)	9.6e-09^***	0.441^ns	0.004^**		-0.089	0.310	0.224	0.003
	Pesticide cost	0.952^ns	0.857^ns	0.414^ns	0.497^ns	0.324^ns	0.437^ns	0.423^ns	0.086^(*)		-0.089	-0.091	0.150
	Area planted	0.001^**	0.002^**	0.069^(*)	0.997^ns	0.018^*	0.516 ^ns	0.065^(*)	0.001^**	0.349^ns		0.083	0.209
	ACB	0.788^ns	0.341^ns	0.476^ns	0.816^ns	0.172^ns	0.784^ns	0.240^ns	0.016^*	0.337^ns	0.378^ns		0.069
	Weeds	0.741	0.868	0.527	0.248	0.431	0.586	0.374	0.973	0.110	0.025	0.468

The variables area planted, fertiliser and labour cost for threshing explained most of the variation in the production-costs ratio among the corn growers in the sample (Table 4 column C). Fertiliser cost, which constitutes around 33 to 44% of total cost of production depending on corn types, showed a significant negative correlation with the production-costs ratio. This means that an increase in fertiliser inputs does not warrant higher economic results. Note also the correlation of -0.116 with p-value 0.002 for the production-costs ratio (M) and fertiliser cost in Table 5.

Finally the analysis showed that growing Bt corn had a significant positive impact on the return on investment while none of the other agronomic variables did have a significant effect (R²: 0.05). Note that the overall effect of corn type was not significant (Table 3).

Overall the econometric analysis revealed that among the tested agronomic variables, area planted is the variable with an encompassing positive influence on production output, net income and production cost ration (Table 4).

3.4 Blinder-Oaxaca decomposition

Next the regression equations as above were extended with a pest occurrence and a severity variable to eliminate potential endogeneity bias. The equations for two response variables, i.e. return on investment (RoI) and net income (NI) for BtHT and non-GM corn were selected for the Blinder-Oaxaca decomposition analysis on the basis of the results obtained after subjecting the extended regression models for all the response variables to a Chow test as shown in Table 6.

Table 6 Chow test outcome for production output, net income, return on investment and cost-production ratio. P values: ** = p < 0.01, * = <0.05, ns = not significant.

Response variable	df numerator	df denominator	RSS BtHT	RSS non-GM	ΣRSS	F-value	p-values
Production output	8	70	3.252	9.120	13.210	0.59260	0.78077^ns
Net Income	8	70	21.380	198.940	298.978	3.12391	0.00447**
Return on Investment	8	70	177.493	106.396	358.910	2.31229	0.02913*
Cost-Production Ration	8	70	50.719	119.100	168.703	0.05750	0.99989^ns

RSS = residual sum of squares in the pooled regression; ΣRSS= sum of the RSS

For the decomposition, the RoI and NI equations are estimated separately as discussed above. The regression results for return on investment (Table 7, column 4 and 5) show that among the assessed variables, corn borer occurrence and costs of labour, seeds and pesticides manifested significant negative effects on GM corn. It is interesting to note that together with farm size and fertiliser, corn borer severity showed positive effects on GM corn's RoI. For non-GM corn, the costs of seeds and pesticide have significant positive effects. All other variables including corn borer occurrence and severity show significant negative effects on non-GM corn's RoI.

Table 7 The Blinder-Oaxaca decomposition of the return on investment (ROI) of GM and non-GM corn types. P values: ** = p < 0.01, * = <0.05, (*) = <0.10), ns = not significant.

Explanatory Variables	Characteristics		Coefficients		ROI GM evaluated at means	ROI non-GM evaluated at means	Mean Gap	Characteristics Effects (share in %)	Coefficients effects (share in %)
	${\overline{X}}_{GM}$	${\overline{X}}_{nonGM}$	${\hat{\beta }}_{GM}$	${\hat{\beta }}_{nonGM}$	${\hat{\beta }}_{GM}{\overline{X}}_{GM}$	${\hat{\beta }}_{nonGM}{\overline{X}}_{nonGM}$		$\left({\overline{X}}_{GM}-{\overline{X}}_{nonGM}\right){\hat{\beta }}_{GM}$	${\overline{X}}_{nonGM}\left({\hat{\beta }}_{GM}-{\hat{\beta }}_{nonGM}\right)$
Intercept	1.000	1.000	11.413	6.806	11.413	6.806	4.607		4.607 (-171%)
Farm size	2.182	1.500	0.042^ns	-0.341^ns	0.091	-0.512	0.602	0.028 (-4%)	0.574 (-21%)
Labour cost	228.341	208.571	-0.007^ns	-0.002^ns	-1.598	-0.417	-1.181	-0.138 (20%)	-1.043 (39%)
Seed cost	163.818	90.024	-0.014^ns	0.033**	-2.293	2.971	-5.264	-1.033 (147%)	-4.231 (157%)
Fertiliser cost	198.750	165.405	0.004^ns	-0.007^(*)	0.795	-1.158	1.953	0.133 (-19%)	1.819 (-68%)
Pesticides	6.727	7.429	-0.287^ns	0.318^ns	-1.931	2.362	-4.293	0.201 (-29%)	-4.495 (167%)
Corn borer severity	3.659	3.439	0.127^ns	-0.730*	0.465	-2.510	2.975	0.028 (-4%)	2.947 (-109%)
Corn borer occurrence	1.091	1.119	-2.689^(*)	-0.123 ^ns	-2.934	-0.138	-2.796	0.075 (-11%)	-2.871 (107%)
Sums					4.007	7.405	-3.397	-0.705 (100%)	-2.692 (100%)

The estimated models were then used to split the observed gap between corn types in two portions (Table 7, last three columns). The sum in the bottom row of Table 7 shows that of the overall raw gap of -3.397 for RoI only 21% (-0.705) can be explained by differences in characteristics of the two samples. The remaining 79% (-2.692) can be attributed to the coefficient or unexplained effect. Notice that the gap is negative and thus the switch to GM corn would mean a drop in RoI for the farmers on average. The last two columns of Table 7 present the contribution of each explanatory variable to the explained and the unexplained component, respectively. In terms of the explained part, most important contributions to explaining the negative gap come from the seed cost (147%) followed by some distance by labour costs (20%). Notice that all the other characteristics reduce the gap (negative percentages).

For net income (NI) the regression results in Table 8 show that among the assessed variables, seed cost, fertiliser cost and corn borer severity carry negative signs. These are variables which manifest negative effects on NI. Farm size, labour cost and pesticide cost have positive signs hence, exhibit significant positive effects on NI for both corn types. Further analysis shows that the two main parts of the mean gap (1.144) have opposite signs; we find a small negative characteristics effect (-23%) and a large positive coefficient effects (123%). In particular, among the explanatory variables of the negative characteristic components, seed cost has the largest percentage (112%) followed by fertiliser cost (61%). Except for farm size and labour cost, the remaining characteristics contribute to increasing the negative gap (positive percentages). Contrary to RoI, the overall gap indicates that adopting GM corn could potentially increase the growers’ income. The results in the last two columns of Table 8 show that the mean income advantage from switching to BtHT corn is mainly due to better control of corn borer pest.

Table 8 The Blinder-Oaxaca decomposition of the Net Income (NI) of GM BtHT and non-GM corn types. P values: (*) = <0.10), ns = not significant.

Explanatory Variables	Characteristics		Coefficients		NI GM evaluated at means	NI non-GM evaluated at means	Mean Gap	Characteristics Effects (share in %)	Coefficients effects (share in %)
	${\overline{X}}_{GM}$	${\overline{X}}_{nonGM}$	${\hat{\beta }}_{GM}$	${\hat{\beta }}_{nonGM}$	${\hat{\beta }}_{GM}{\overline{X}}_{GM}$	${\hat{\beta }}_{nonGM}{\overline{X}}_{nonGM}$		$\left({\overline{X}}_{GM}-{\overline{X}}_{nonGM}\right){\hat{\beta }}_{GM}$	${\overline{X}}_{nonGM}\left({\hat{\beta }}_{GM}-{\hat{\beta }}_{nonGM}\right)$
Intercept	1.000	1.000	4.522	5.321	4.522	5.321	-0.799		-0.799 (-57%)
Farm size	2.182	1.500	0.113^ns	0.526^ns	0.247	0.788	-0.542	0.077 (-29%)	-0.619 (-44%)
Labour cost	228.341	208.571	0.010^(*)	0.012^ns	2.215	2.461	-0.246	0.192 (-73%)	-0.438 (-31%)
Seed cost	163.818	90.024	-0.004^ns	-0.001^ns	-0.655	-0.108	-0.547	-0.295 (112%)	-0.252 (-18%)
Fertiliser cost	198.750	165.405	-0.005^ns	-0.003^ns	-0.954	-0.568	-0.386	-0.160 (61%)	-0.226 (-16%)
Pesticides	6.727	7.429	0.074^ns	0.013^ns	0.495	0.093	0.402	-0.052 (20%)	0.454 (32%)
Corn borer severity	3.659	3.439	-0.014^ns	-0.182^ns	-0.053	-0.627	0.574	-0.003 (1%)	0.577 (41%)
Corn borer occurrence	1.091	1.119	0.762^ns	-1.660^ns	0.831	-1.858	2.689	-0.021 (8%)	2.710 (193%)
Sums					6.648	5.504	1.144	-0.263 (100%)	1.407 (100%)

The decomposition technique serves to distinguish an observable characteristics effects and an unexplained coefficient effect. The coefficient component can have a different sign from the characteristics component and this can give insightful information in particular. If both components have the same sign, differences in return on investment (RoI) or net income (NI) are as expected. The last two columns of Table 7 show that for RoI the sums of the two components have identical signs (negative). However for individual variables differences in signs do occur. For both pesticide costs and for corn borer occurrence there is a negative impact on the RoI which is unexpected given the lower average for these variables for the non-GM sample. Finally, the intercept is responsible for most of the coefficients effects indicating the contribution of unobservable characteristics (such as physio-chemical characteristics of cornfields) to the difference in RoI.

In contrast, in case of NI, the sums of the two components shown in the last two columns of Table 8 do not have identical signs. The characteristics effect making up a small portion (23%) of the gap bears a negative sign. This indicates a negative effect on NI from the differences in BtHT and non-GM farmers’ observable characteristics which is mainly attributed to seed costs and costs of fertiliser inputs. However this is counteracted by the coefficients or unexplained component which carries a positive sign and is mainly due to pesticide input, corn borer severity and occurrences. In general, this shows that BtHT has disadvantages for NI based on observable characteristics, yet could provide economic advantage overall due to better pest control even for cornfields less heavily infested with corn borer pest and also due to savings on pesticide costs.

Finally, the research results from the decomposition analysis and the positive and negative overall mean gaps for income (NI) and return on investment (RoI), respectively, conveys an important overall message. These results suggest that when farmers are more economically capable and have the ability to finance their own farm expenses, they are more likely to exploit the opportunities of GM corn to a larger extent. Farmers who cannot afford to finance their field expenses (seed and pesticides) are obliged to borrow money from financiers with high interest rates. These farmers are more likely to be unable to grow the GM corn in the intended way. This is because they are more financially constraint and likely to face shortfalls in input supplies [16]. Related research of the sample [21] found indeed that reducing the labour requirement was not a major adoption argument for the growers of BtHT and HT corn. This highlights, as mentioned above, that poor farmers may still resort to manual weeding of BtHT.

4 Conclusion

This study focused on the economic impact of genetically modified (GM) corn as compared to non-GM corn with especial consideration for small-scale farmers in Isabela province, Philippines. We used uni-variate, multivariate, and econometric decomposition analysis to determine which corn type is worth investing in by resource-poor farmers considering the production output, net income, return on investment, cost of production and significant agronomic variables.

The vast increment and wide-scale cultivation of GM corn in the Philippines is attributed to its purported economic benefits such as increased yield and higher profits for farmers. Yet, after comprehensively considering the economic results in the context of actual field experiences, our study shows a less optimistic picture. The current socio-economic and agronomic conditions in Isabela province negatively affect the purported economic advantages of GM corn as achievable under ideal conditions. Important reasons are, first, the financial constraint of high seed costs in combination with the expensive credit system which means farmer may pay interest of 7 to 15% to finance their inputs. Secondly, there is the issue of technical inefficiency exemplified by the continuation of weeding practices and pesticide use by some farmers despite the use of the GM crop. Poor farmers may still resort to manual weeding in BtHT corn as they cannot afford to buy herbicides.

Under the prevailing socio-economic conditions, this study found no significant difference in net income between corn varieties. This shows that the economic advantage of GM corn tends to be relative, at least in the present context of resource-poor small scale farmers in the Philippines. Bt corn was shown to be effective in controlling the Asian Corn borer problem in Isabela province [43]. This finding in combination with the results from our decomposition analysis suggest that small-scale and poor farmers whose fields are experiencing heavy corn borer infestations but who can hardly afford to buy Bt corn seeds would best benefit from this technology through seed subsidies by the Philippine government via its Department of Agriculture.

Acknowledgement

We gratefully acknowledge the constructive comments by the editor and an anonymous reviewer. We would like to extend our appreciation and gratitude to the farmers for their cooperation and unselfishly sharing their first-hand information with us. In addition, we thank to the following people: RG Salazar, RG Mabutol Jr., AM Vanyvan, G. Persoon, J van der Ploeg, P.M. Afidchao Jr., RR Quilang, MD Masipiqueňa, T Minter, MTR Aggabao, M. van Weerd, A. Domingo Sr., EF Macaballug, RC Ramirez, WC Medrano, AHG Aggabao, CB Mabutol, RM del Rosario, R Aquino, EM Ariola, DM Mabutol, KC Villegas and WB Saliling. This study was supported by the Louwes scholarship program of Leiden University in the Netherlands.