But the problem is, the government channel has a ceiling value for the amount purchase from a farmer as 2500 kgs. Private channel does not have such a ceiling value. Which means, even if a farmer prefer to sell more, via government channel, he or she can't due to the existing legislation.
in such a scenario, can I still use logit or probit to analyze farmers' marketing channel choice? or is there any other suitable method? Thank you.
Thank you very much for your answers.
As some collogue asks, I would like to explain it further. The sample size was 350 farmers of which some have sold their entire quantity through government channel while some have sold their entire quantity through private channel and the rest have used both channels to sell their quantity.
A) For a farmer who has less than 2500kg marketable surplus, can choose either government channel, private channel or both in selling their produce.
B) However, for a farmer who has a marketable surplus over 2500kg, selecting private channel as one of the marketing channels is not a choice but a must as government put a celling value of 2500kgs for the amount purchased. In other words, they have selected the government channel voluntarily but pushed to select private channel to sell the rest.
C) On the other hand, there were around 50% farmers who did not use the government channel at all due certain reasons (inefficiencies ect) and chose only private channel, although they are perfectly eligible to sell 2500kgs via government channel.
In such a situation, can I still use logit or probit to forecast farmers channel choice?
I feel that, decision of selecting channels are not mutually exclusive in the case B. Could you please comment?
Dear colleague
I have difficulty imagining it because I do not know exactly the application in the country. However, the logit or probit model is suitable for a binary comparison. I think it is necessary to include an independent variable of 2500 kg limit to include the condition in the model. So buying rice can be a private and dependent variable in the state sector. But the independent variable should also be constrained by the sale of rice. So the model gets closer to reality. In such a study, it can be expected that this variable has a high coefficient. This is not a problem. The important thing is the solution you will give to this fiction.
Achievements
First of all, how does your data looks like? do all the farmers sell either via government or private channel? Or sometimes they use both channels?
Dear Champika,
If I understood you properly You have farmers with two marketing outlets: government and private ones. If the distribution of your dependent variable is purely on these two outcomes you can use logit or probit and it does not have serious estimation problem. However, you have to be able to control the factors affecting the choice of these market outlets (prices of the two buyers, location, product quality requirement, and other factors). If the outcome variable consists of the third alternative as mentioned by Peter, then you have to use mlogit/mprobit.
I disagree with previous suggestion. Even if there are only two possible outcomes, it still makes sense to model this as a two stage problem:
1. Some variables determine size of your output.
2. If your output is above 2500, then you have to choose private channel. If your output is below 2500, then you can choose between them.
One possibility would be to use logit/probit only for those observations where the output is below 2500. However, this would still not be optimal, because some farmer, who could hava capacity for example for production of 2550 might decide to cut production to 2500 in order to be able to use governmental channel.
I am not sure if any simple standard econometric technique would be sufficient for proper modelling of your problem. I think You might need to resort to strucutral modelling.
Dear Jamamini Champika,
This is my candid opinion on your question.
From the question and the explanation you have provided in points A, B and C, the dependent variable cannot be binary and therefore the binary logit/probit may not be the appropriate technique to use.
If I get you right in the explanation, you are saying that a farmer can either choose government channel, private channel or even both. You mentioned again that some farmers did not choose government at all but rather chose private channel. In this case you may have to consider multinomial logit/probit. This is because, a farmer has more than one choice at any point in time regardless of the ceiling value (2500kg) you are talking about.
You can only use the binary if a farmer has only one choice (may be only government or private) and for that matter the farmer either chooses it or otherwise; then you will also have binary response as well: Yes (for may be choosing government channel) and No (for not choosing government channel). But where the choices are many as you have in your case, binary logit or probit may not help.
Again, if you think the choice of these channels are interrelated, then you may also consider multivariate logit/probit where these choices can be estimated simultaneously. On the other hand, if you think they are not interrelated then you may consider the multinomial logit/probit. If you think both (multinomial and multivariate) are likely to be the case, you may estimate both multinomial and multivariate logit/probit regressions and compare the results if you wish.
You may read the following two papers for more information on the usage of the multinomial and the multivariate:
1. Akpandjar, M. G., Quartey, P. & Abor, J. (2013). Demand for Financial Services by Households in Ghana. International Journal of Social Economics, 40(5), pp. 439-457. doi:10.1108/03068291311315322.
2. Bendig, M., Giesbert, L. & Steiner, S. (2009). Savings, Credit and Insurance: Household demand for Financial Services in Rural Ghana. Working paper, No. 94/2009, German Institute of Global and Area Studies, 2009.
Thank you.
I tend to agree with Samuel. it is question of overlap of farmers across two channels dictated by state law. So, you must think about it before going for binomial model. Or you treat same farmer selling in two channels as two separate farmers for knowing their choice.
U can go for multinomial logit/probit by coding different channels. if u want to run binary probit u can take up each channel separately viz. government channel if yes-1, other wise zero and so on
I agree with the first part of the response by Feroze. The best thing is to use mlogit or mprobit since the choice is multiple rather than taking them separately. Besides I think the data has been collected already and some farmers have even indicated they use both government and private channels. In this case it cannot be coded yes and no for a particular channel used by a farmer.
Thank you
Dear Champika
From your detailed query, it is clear your dependent variable is rather a multiple (3) representing marketing through the government channel, private channel or both. In this case, you have to model this using a mutinomial logit or probit model.
To control for the government restriction of 2500kg, you can include it as a covariate in your model. This will then explain how this legislation affects the marketing choice which ofcourse it does.
Lastly, should there be some correlation between the dependent variables, a multivariate logit or probit does the trick. As a matter of fact, you can estimate both the multinomial model and the multivariate model and compare both.
Dear Champika
I found this paper (Agricultural Technology Adoption under Climate Change in the Sahel: Micro-evidence from Niger by Solomon Asfawa, Federica Di Battistab, and Leslie Lippera [see the attached]) on the choice of farming technology by farmers and I see it to be very useful for your case (i.e. if only you agree with me that your dependent variable is not binary following the further explanation you provided in A, B and C and are also related). In this paper, the author examined the determinants the choice of farmers regarding farming technologies. The author further added that the farmers adopt or choose mix of technologies to deal with a multitude production constraints than to adopt a single practice/technology. The choices are crop residues, organic fertilizer and modern inputs. The author found these choices to be related and so used the multivariate probit model for analysis. You can see the section 4 of the paper for the modelling and Table 5 of the paper for a guide on the results and interpretation.