Adoption status of artificial insemination in Indian dairy sector: application of multinomial logit model

ABSTRACT

India is blessed with the highest population of dairy animals, but has very poor productivity per se, which may be due to various factors such as lack of improved breeding and feeding services, access to markets, capital, inputs and technology. Although professionals have pointed out artificial insemination (AI) as an emerging technology of socio-economic importance, the ground realities or practices about AI is entirely different. With this theoretical background, an attempt was made to study the adoption status of AI and also to identify factors affecting adoption of AI in India by both qualitative and quantitative methods from 360 dairy farmers in four states of North India. The adoption status of AI pointed out that majority of the respondents had adopted AI since more than 9 years, which was mostly partial in nature. Multinomial logit model depicted the Chi-square value of 113.97, indicating that the model was highly significant (p < .001). Two variables (‘distance to veterinary institution’ and ‘education’) significantly influenced the probability of partial and full adoption of AI. In case of full adoption, in addition to the above variables, ‘size of livestock holding’ and ‘scientific orientation’ also emerged as significant variables. Since majority of the dairy farmers were in partial adoption category, scientists have to analyse the problems of dairy farmers and find suitable solutions for higher diffusion and adoption at field conditions by participatory technology generation and transfer approach in India.

KEYWORDS:

• Dairy innovation
• livestock technology
• multinomial logit model
• adoption

1. Introduction

Dairying is an effective tool for rural development, employment and sustained income and it acts as an insurance against several odds (Prasad 2011). Though, India is blessed with 190.09 million cattle and 108.7 million buffaloes (GOI 2012), the productivity per se is very poor. For instance, the average annual milk yield of Indian cattle is 1172 kg, which is only about 50% of the global average (FAOSTAT 2014), and much less than New Zealand (3343 kg), Australia (5600 kg), UK (7101 kg), US (9332 kg) and Israel (10214 kg). Chander et al. (2010) also pointed out that poor productivity as well as the quality of production and products remains a cause of concern in Indian livestock and dairy sector. However, the productivity in dairy sector is low, and smallholders are constrained by a lack of access to markets, capital, inputs, technology and services. Reduction of production risks faced by rural landless dairy farmers requires the availability of improved breeding services, targeted preventive animal health care, better feeding strategies and easy access to formal credit facilities (Torsten et al. 2003). Among all such bottlenecks, higher population of poor producing indigenous breeds and poor adoption of artificial insemination (AI) has to be addressed with greater emphasis. Among various dairy innovations in India, AI has been considered as an emerging dairy innovation of socio-economic importance in Indian dairy industry (Rathod & Chander 2014). Although professionals have pointed out AI as an emerging dairy innovation of importance, the ground realities or practices about AI is entirely different. With this theoretical background, an attempt was made to study the adoption status of AI and also to identify factors affecting adoption of AI in India. Further, the study also proposed certain policy implications for Indian dairy industry to improve the diffusion and adoption of AI.

2. Materials and methods

2.1. Study locale and sampling

Keeping in view the objectives of the study, the districts where the Veterinary Universities are situated were purposively selected for data collection. The existence of the veterinary universities and the concerned districts selected for study are presented in Table 1. Multistage random sampling and snow-ball method were followed to select 15 dairy farmers from each village making a final sample size of 360 farmers from 24 villages in four states. During the selection of respondents, care was taken to select the farmers who reared at least two dairy animals and practised dairy farming as major or subsidiary occupation.

Table 1. Locale of the study.

Districts under study (for dairy farmers)	States	Geographical location
Bareilly	Uttar Pradesh (UP)	28.36°N & 79.41°E
Udham Singh Nagar	Uttarakhand (UK)	28.98°N & 79.40°E
Karnal	Haryana	29.69°N & 76.98°E
Ludhiana	Punjab	30.91°N & 75.85°E

The primary data from the dairy farmers were collected either at their farm or home using pretested interview schedule by personal interview method during November 2013 to June 2014. Information through observation during interview and group discussion was also collected. Adoption of AI was categorized as ‘adoption’, ‘partial adoption’, ‘discontinuation’ and ‘non-adoption’, categories. Adoption status of AI by farmers was enquired to know the total number of years since when a particular practice was introduced in their farm. The responses were categorized into adoption since ‘last 0–3 years’, ‘last 3–6 years’, ‘last 6–9 years’ and ‘more than 9 years’. The data collected from sample respondents were coded, tabulated, analysed and presented in the form of tables. The statistical tools (namely frequency, percentage, mean, standard deviation and Chi-square test) were used for analysis of the data using SPSS version 20.0 package. The inferences were drawn in light of the results obtained, keeping in view the objectives laid in the study.

2.2. Application of multinomial logit model to identify the factors influencing degree of AI adoption

To identify the factors that influence the respondents’ degree of adoption of AI, a multinomial logit model (as used by Pundo & Fraser 2006) was fitted. The multinomial logit model not only focused on the most important decision (whether the farmer adopts AI or not), but also on the degree of adoption of AI. In the fitted model, the dependent variable assumed three discrete values – 0 (when the respondent did not adopt AI), 1 (when the respondent partially adopted AI) and 2 (when the respondent fully adopted AI). Given the alternatives before a respondent, the probability that an individual i choose alternative j, therefore, can be expressed by the following equation:(1)

where:

Pr[Yi = j] = Probability that an individual i belongs to either ‘No adoption’,‘Partial adoption’ and ‘Full adoption’ category.

j = 1, 2, 3

i = 1, 2, 3,  … , 360

Xi = Vector of the predictor variables and

βj = Vector of the estimated parameters

The multinomial logit model determines the effect of independent variable on the probability that a farmer will belong to one of the three categories (namely non-adopter, partial adopter and full adopter). This model was estimated by keeping the dependent variable 0 (i.e. non-adopter) as the reference category. The e^β was calculated, which gave the odds ratio (OR) associated with change in the independent variables. The odds mean the ratio of probability of happening of an event to probability of not happening of that event. The odds are expressed as single number to the ratio to 1. The odds of 2 associated with partial adoption, for example, means that the likelihood of partially adopting AI is twice that of not adopting. Zero-order correlation matrix was obtained to ensure that multi-collinearity did not pose any problem in estimating parameters of the mathematical model. The variables having higher multi-collinearity were dropped in the final model to improve the values of the variables. Table 2 depicts the variables used in the model with their expected signs.

Table 2. Variables used in multinomial logit model and their expected signs.

Variables	Definition and measurement	Expected sign
Age	Age of the respondent in years	±
Family size	Total number of members in a household	±
Land	Size of landholding of household in acres	±
Livestock	Number of livestock owned by the household in units	±
Distance	Average distance from veterinary institution or animal healthcare centre in kilometres	−
Information sources	Information-seeking behaviour of the respondent from various sources on three point continuum as ‘frequent’, ‘less frequent’ and ‘never’	±
Decision	Decision-making ability of the respondent related to dairy farming on three point continuum as ‘low’, ‘medium’ and ‘high’	±
Scientific	Scientific orientation of the respondent on three point continuum as ‘low’, ‘medium’ and ‘high’	±
Economic	Economic orientation of the respondent on three point continuum as ‘low’, ‘medium’ and ‘high’	±
Risk	Risk orientation of the respondent on three point continuum as ‘low’, ‘medium’ and ‘high’	±
educnew = 1	Illiteracy of the respondent	−
educnew = 2	Education up to 10 years of education	±
ocupnew = 1	Agriculture as major occupation and animal husbandry as subsidiary occupation of the household	±
ocupnew = 2	Animal husbandry as major occupation of the household	±
partinew = 1	No social participation of the respondents	−
partinew = 2	Respondent is the member of at least one social organization	±

3. Results and discussion

3.1. Adoption categories for AI

The adoption status of AI was studied to know whether the farmers had adopted AI or not, and the type of adoption followed. Table 3 indicates that among the pooled data, 62.5% of the respondents partially adopted AI, while 24.45% farmers fully adopted AI in their farm. It was considered partial adoption since in many instances, the farmers kept on switching from natural breeding to AI and vice-versa depending on the success or failure of AI. A highly significant difference (p < .001) among the respondents across the states with regard to adoption of AI was observed in the study area, which might be due to variation of socio-economic status, risk and economic orientation and livestock holding. Almost similar findings were also reported by Basunathe et al. (2010), Aulakh and Singh (2012) and Rathod et al. (2014) with regard to adoption of AI.

Table 3. Adoption categories of AI at field conditions (N = 360).

Categories	States				Pooled	χ^2
	UP	UK	Haryana	Punjab
Non-adoption	14 (15.56)	12 (13.34)	02 (2.22)	01 (1.11)	29 (8.05)	117.93**
Discontinuation	05 (5.55)	03 (3.33)	05 (5.56)	05 (5.56)	18 (5.0)
Partial adoption	68 (75.56)	68 (75.56)	61 (67.78)	28 (31.11)	225 (62.5)
Full adoption	3 (3.33)	7 (7.77)	22 (24.44)	56 (62.22)	88 (24.45)

Note: Figures in the parenthesis indicate percentage.

**Depict significance at 1% level of significance.

3.2. Adoption status of AI (Years since when AI was adopted)

Although the dairy farmers were categorized into full adoption, partial adoption, discontinuation and non-adoption with respect to adoption status of AI, an attempt was made to study the total number of years since when AI was adopted at the farmers’ field. The study also included the total number of years a farmer adopted AI even before discontinuing, if any. Table 4 indicates that among the pooled data, 47.50% of the respondents had adopted AI since more than 9 years, while 23.89% farmers adopted it since 6–9 years and 15.28% respondents were in the 3–6 year categories. Such distribution of AI adoption may be due to the variation in socio-economic status, information access and scientific orientation in the study area. This might also be the reason for highly significant difference (p < .001) among the respondents across the states with regard to adoption of AI in dairying. In a similar study, Rathod et al. (2014) also reported similar adoption status for AI in Karnataka state of India.

Table 4. Adoption status of AI (Since when AI was adopted) (N = 360).

Categories	States				Pooled	χ^2
	UP	UK	Haryana	Punjab
Non-adoption	14 (15.56)	12 (13.33)	02 (2.22)	01 (1.11)	29 (8.05)	124.9**
0–3 years	14 (15.56)	02 (2.22)	03 (3.33)	00 (0)	19 (5.28)
3–6 years	30 (33.33)	12 (13.33)	10 (11.11)	03 (3.33)	55 (15.28)
6–9 years	17 (18.89)	32 (35.56)	24 (26.67)	13 (14.45)	86 (23.89)
More than 9 years	15 (16.66)	32 (35.56)	51 (56.67)	73 (81.11)	171 (47.50)

Note: Figures in the parenthesis indicate percentage.

**Depict significance at 1% level of significance.

3.3. Reasons for adoption of AI

The reasons for adoption of AI in dairying as perceived by respondents has been enlisted.

• Timely AI is advantageous

• Good conception rate by AI

• Poor availability of good-quality bulls in the villages

• Human resource available for performing AI (AI Centre available)

• Nominal or low charges for AI

• Healthy calves born from AI

• Semen from males of high genetic merit can be used as per choice

• Enables breeding between animals of different geographic locations, or at different times

• Breeding can occur in the event of physical, physiological or behavioural abnormalities;

• AI can be used in conservation of required breeds or species

• Increased safety for animals and farmers since males can be large and aggressive

• Reduced transfer of venereal diseases

The adoption studies of Rezvanfar (2007) and Rathod et al. (2014) observed almost similar findings with regard to adoption of AI at field conditions.

3.4. Reasons for partial adoption/ discontinuation and non-adoption of AI

Following are the major reasons for partial adoption/discontinuation or non-adoption of AI as perceived by respondents in the study area.

• Estrous detection is very difficult leading to untimely AI

• Poor conception rate

• Costly/higher charges for AI

• Lack of skilled trained personnel

• More chance of getting male calf

• Unhealthy calves born from AI

• More chance of disease occurrence by AI

• Decline in fertility of dairy cattle associated with AI

• Poor infrastructure to store cooled and frozen semen for AI

• Distant AI centre from village

• More chance of reproductive disorders

• Focus on certain individuals may result in loss of genetic variation

• Poor post AI dairy animal management.

With regard to adoption of AI, Gandhi et al. (1998) reported that in spite of huge infrastructure and budgetary provisions over the years, a meagre (10%) bovine population in India was covered through AI. In another study, conducted in Bareilly district of Uttar Pradesh by Lal (2000), it was indicated that majority of the farmers perceived AI less profitable and less sustainable as compared to vaccination, deworming and fodder cultivation. In a similar context, The Asian Development Bank (ADB 1993) study on policies and strategies for livestock improvement in developing countries concluded that the primary policy failure was promotion of inappropriate technology. This was reflected in continuing problems experienced in livestock development programmes and projects.

3.5. Identifying factors influencing degree of AI adoption: application of multinomial logit model

Table 5 presents the results of the multinomial logit model fitted to identify the determinants of adoption of AI. The Chi-square value of 113.97 showed that likelihood ratio statistics are highly significant (p < .000), suggesting that the model has high explanatory value. Two variables, namely ‘distance to veterinary institution’ (p < .05) and ‘education’ (p < .05) significantly influenced the probability of partial and full adoption of AI. While the sign associated with the regression coefficient of the former variable was negative, that pertaining to the latter was positive. In case of full adoption, in addition to the above variables, ‘size of livestock holding’ (p < .05) and ‘scientific orientation’ (p < .05) emerged as significant variables and the signs of regression coefficients for both these variables were positive.

Table 5. Multinomial logit model for adoption of AI.

Variables	Partial adoption			Full adoption
	β (S.E)	Wald	Odds ratio	β (S.E)	Wald	Odds ratio
Intercept	−2.883 (4.683)	0.379		−25.538 (4.938)	26.747
Age	0.011 (0.013)	0.754	1.011	0.013 (0.015)	0.715	1.013
Land	−0.038 (0.051)	0.568	0.962	−0.053 (0.054)	0.955	0.948
Livestock	0.045 (0.069)	0.421	1.046	0.162** (0.077)	4.461	1.175
Distance	−0.082** (0.038)	4.586	0.921	−0.254** (0.064)	15.775	0.776
Information sources	0.16 (0.159)	1.024	1.174	0.25 (0.171)	2.128	1.284
Decision	0.005 (0.089)	0.003	1.005	−0.009 (0.104)	0.007	0.991
Scientific	−0.052 (0.16)	0.106	0.949	0.409** (0.194)	4.45	1.506
Economic	−0.139 (0.184)	0.565	0.871	−0.147 (0.199)	0.542	0.864
Risk	0.241 (0.26)	0.857	1.273	0.21 (0.281)	0.559	1.234
Family size	0.003 (0.051)	0.005	1.003	0.076 (0.059)	1.667	1.079
educnew = 1	0.219 (0.813)	0.073	1.245	0.73 (0.933)	0.612	2.075
educnew = 2	1.777** (0.787)	5.097	5.912	2.198*** (0.84)	6.845	9.005
ocupnew = 1	−0.489 (0.532)	0.844	0.614	−0.11 (0.637)	0.03	0.896
ocupnew = 2	−0.382 (1.221)	0.098	0.683	−0.224 (1.335)	0.028	0.799
partinew = 1	1.116 (1.488)	0.562	3.053	13.595 (0.521)	680.424	801,928.5
partinew = 2	1.448 (1.469)	0.973	4.256	14.407 (0.0)		1,806,582

Notes: Figures within parentheses indicate standard errors; Significant at ***1%, **5% and *10% levels of significance. Number of observations: 360; Wald chi²: 113.97; Prob < Chi²: 0.0000; Pseudo R²: Cox and Snell (0.311), Nagelkerke (0.36); Log pseudo likelihood: 579.40.

The negative influence of distance to veterinary institution implies that as the distance to veterinary institution increased, the farmer tends to move towards non-adoption from partial and full adoption. The OR suggests that with the increase in one unit of distance from veterinary institution, the likelihood of full adoption of AI decreased by 77.6% in the study area. Higher level of education was positively and significantly associated with both partial and full adoption, implying the importance of education in adoption of innovation. The finding is in line with the findings of Adeogun et al. (2008) and Odendo et al. (2009), who reported that education was positively and significantly related to adoption of hybrid heteroclarias and natural resources conservation technologies, respectively. The positive association of livestock holding with full adoption implies that adoption of this innovation increases with increase in scale of production. The odds of full adoption increased by 17.5% with one unit increase in livestock holding. With increase in scientific orientation, the likelihood of full adoption of AI increased by 50.6% in the study.

4. Conclusion and policy implications

The adoption status of AI pointed out that, majority of the respondents had adopted AI for more than 9 years, which was mostly partial in nature. Further, about 8% dairy farmers in the study area never followed AI in the study region. Multinomial logit model depicted the Chi-square value of 113.97, indicating that the model was highly significant (p < .000). Two variables (namely ‘distance to veterinary institution’ and ‘education’ significantly) influenced the probability of partial and full adoption of AI. In case of full adoption, in addition to the above variables, ‘size of livestock holding’ and ‘scientific orientation’ also emerged as significant variables.

The study suggests that researchers and extension experts need to make farmers more aware about the benefits of AI in the dairy sector. Since majority of the dairy farmers were in partial adoption category, a need-based long-run study under field conditions must be undertaken. Further, the scientists have to analyse the problems faced by farmers and find suitable solutions for higher diffusion and adoption of AI at field conditions. The study recommends for participatory technology generation and transfer approach with adequate representation of all categories of farmers, that is, small and marginal, medium and large so that their feedback can be taken into consideration. This can also emphasize on the generation and transfer of dairy innovations based on the socio-economic background of the dairy farmers.

Statement of ethical standards

The authors state that animals were not involved in the study. The study was based on the involvement of humans (dairy farmers) as respondents, and a prior consent has been taken for their inclusion in the study.

Acknowledgements

The authors offer their sincere thanks to Director, IVRI, Izatnagar for providing the necessary facilities in conducting this research work. The authors are also thankful to all the respondents for sharing their valuable views in the study.

Disclosure statement

No potential conflict of interest was reported by the authors.