in

A perspective of scale differences for studying the green total factor productivity of Chinese laying hens

Minimum distance to weak efficient frontier

Briec and Charnes et al. first proposed the Minimum distance to weak efficient frontier (MinDW) model39,40, which can be expressed as (m + n) linear programming ((m) is the number of input indicators and (n) is the number of output indicators), assuming that the input variable is (x) and the output variable is (y). The specific formula is shown in Eq. (1):

$$ begin{aligned} & max beta_{z} ,z = 1,2, ldots ,m + n & s.t.left{ begin{gathered} sumnolimits_{j = 1}^{q} {alpha_{j} x_{rj} + beta_{z} e_{r} le x_{rk} ,r = 1,2, ldots ,m} hfill sumnolimits_{j = 1}^{q} {alpha_{j} x_{ij} + beta_{z} e_{i} ge y_{ik} ,i = 1,2, ldots ,n} hfill alpha_{j} ge 0 hfill end{gathered} right. end{aligned} $$

(1)

(e_{r}) and (e_{i}) are constants. In the programming formula, only one (e) is equal to 1, and the others are 0, that is shown in Eq. (2):

$$ begin{aligned} & e_{r} = 1;{text{ if}}; , r = z; , e_{r} = 0 , ;{text{if}}; , r ne z & e_{i} = 1 , ;{text{if}}; , i = z – m; , e_{r} = 0 , ;{text{if}}; , i ne z – m end{aligned} $$

(2)

The efficiency value of model is expressed as Eq. (3):

$$ theta_{z}^{*} = frac{{1 – frac{1}{m}sumnolimits_{r = 1}^{m} {frac{{beta_{z}^{*} e_{r} }}{{x_{rk} }}} }}{{1 + frac{1}{n}sumnolimits_{i = 1}^{n} {frac{{beta_{z}^{*} e_{i} }}{{y_{ik} }}} }} $$

(3)

The efficiency value of MinDW model is expressed as (theta_{max }^{*} = max (theta_{z}^{*} ,z = 1,2, cdots ,m + n)), and the maximum efficiency value corresponds to the minimum (beta^{*}), that is the nearest distance to the frontier.

This paper uses the MinDW model with negative output to conduct empirical analysis. The method can be expressed as (m + n + d) linear programming ((m) is the number of inputs, (n) is the number of desirable output, (d) is the number of unexpected output), assuming that the input variable is (x), the desirable output variable is (y), and the undesirable output variable is (f). The specific formula is shown in Eq. (4):

$$ begin{aligned} & max beta_{z} ,z = 1,2, ldots ,m + n + d & s.t.left{ begin{gathered} sumnolimits_{j = 1}^{q} {alpha_{j} x_{rj} + beta_{z} e_{r} le x_{rk} ,r = 1,2, ldots ,m} hfill sumnolimits_{j = 1}^{q} {alpha_{j} x_{ij} – beta_{z} e_{i} ge y_{ik} ,i = 1,2, ldots ,n} hfill sumnolimits_{j = 1}^{q} {alpha_{j} x_{lj} + beta_{z} e_{l} le f_{lk} ,l = 1,2, ldots ,d} hfill alpha_{j} ge 0 hfill end{gathered} right. end{aligned} $$

(4)

(e_{r}), (e_{i}) and (e_{l}) are constants. In the programming formula, only one (e) is equal to 1, and the others are 0, that is shown in Eq. (5):

$$ begin{aligned} & e_{r} = 1;{text{ if}}; , r = z; , e_{r} = 0 , ;{text{if}}; , r ne z & e_{i} = 1 , ;{text{if }};i = z – m; , e_{r} = 0 , ;{text{if}}; , i ne z – m & e_{l} = 1 , ;{text{if}}; , l = z – m – n; , e_{l} = 0 , ;{text{if}}; , l ne z – m – n end{aligned} $$

(5)

The efficiency value of model is expressed as Eq. (6):

$$ theta_{z}^{*} = frac{{1 – frac{1}{m}sumnolimits_{r = 1}^{m} {frac{{beta_{z}^{*} e_{r} }}{{x_{rk} }}} }}{{1 + frac{1}{n + d}left( {sumnolimits_{i = 1}^{n} {frac{{beta_{z}^{*} e_{i} }}{{y_{ik} }}} + sumnolimits_{l = 1}^{d} {frac{{beta_{z}^{*} e_{l} }}{{f_{lk} }}} } right)}} $$

(6)

The efficiency value of MinDW model is expressed as (theta_{max }^{*} = max (theta_{z}^{*} ,z = 1,2, cdots ,m + n + d)), and the maximum efficiency value corresponds to the minimum (beta^{*}), which means the nearest distance to the frontier.

The efficiency value of MinDW model will not be less than the efficiency value of directional distance function model with any direction vector or other distance types (such as radial model and SBM model). In other words, the efficiency value of MinDW model is the largest. Combined with the above process, we can define the common boundary ((beta^{meta*})) and the model is as Eq. (7):

$$ begin{aligned} & beta^{meta*} = max frac{{1 – frac{1}{m}sumnolimits_{r = 1}^{m} {frac{{beta_{z} e_{r} }}{{x_{rk} }}} }}{{1 + frac{1}{n + d}left( {sumnolimits_{i = 1}^{n} {frac{{beta_{z} e_{i} }}{{y_{ik} }}} + sumnolimits_{l = 1}^{d} {frac{{beta_{z} e_{l} }}{{f_{lk} }}} } right)}} & s.t.left{ begin{gathered} sumnolimits_{j = 1}^{{q_{m} }} {alpha_{j} x_{rj} + beta_{z} e_{r} le x_{rk} ,r = 1,2, cdots ,m} hfill sumnolimits_{j = 1}^{{q_{m} }} {alpha_{j} x_{ij} – beta_{z} e_{i} ge y_{ik} ,i = 1,2, cdots ,n} hfill sumnolimits_{j = 1}^{{q_{m} }} {alpha_{j} x_{lj} + beta_{z} e_{l} le f_{lk} ,l = 1,2, cdots ,d} hfill alpha_{j} ge 0 hfill end{gathered} right. end{aligned} $$

(7)

Similarly, the efficiency value of DMU relative to the scale frontier ((beta^{scale*})) can be obtained by the Eq. (8):

$$ begin{aligned} & beta^{scale*} = max frac{{1 – frac{1}{m}sumnolimits_{r = 1}^{m} {frac{{beta_{z} e_{r} }}{{x_{rk} }}} }}{{1 + frac{1}{n + d}left( {sumnolimits_{i = 1}^{n} {frac{{beta_{z} e_{i} }}{{y_{ik} }}} + sumnolimits_{l = 1}^{d} {frac{{beta_{z} e_{l} }}{{f_{lk} }}} } right)}} & s.t.left{ begin{gathered} sumnolimits_{j = 1}^{{q_{s} }} {alpha_{j} x_{rj} + beta_{z} e_{r} le x_{rk} ,r = 1,2, ldots ,m} hfill sumnolimits_{j = 1}^{{q_{s} }} {alpha_{j} x_{ij} – beta_{z} e_{i} ge y_{ik} ,i = 1,2, ldots ,n} hfill sumnolimits_{j = 1}^{{q_{s} }} {alpha_{j} x_{lj} + beta_{z} e_{l} le f_{lk} ,l = 1,2, ldots ,d} hfill alpha_{j} ge 0 hfill end{gathered} right. end{aligned} $$

(8)

Finally, in the common frontier model, the technology gap ratio (TGR) is equal to the ratio of the efficiency value of the common frontier to the scale frontier41. The formula is as Eq. (9):

$$ TGR^{MinDW} = frac{{beta^{meta*} }}{{beta^{scale*} }} $$

(9)

(beta^{meta*}) and (beta^{scale*}) represent the optimal solution of formula (7) and formula (8), respectively. Obviously, (0 le TGR le 1). TGR is used to measure the distance between the optimal production technology and the potential optimal technology of a group, and identify whether there are any differences in LHG under different groups. The closer the TGR is to 1, the closer the technology level is to the optimal potential technology level. Conversely, it shows the larger gap between the technology level and the potential optimal technology level.

Metafrontier-Malmquist–Luenberger index

Malmquist productivity index is widely used in the study of dynamic efficiency change trend, and has good adaptability to multiple input–output data and panel data analysis. The actual production process often contains unexpected output. After Chung et al. proposed Malmquist–Luenberger (ML) index, any Malmquist index with undesired output can be called ML index42. Oh constructed the Global-Malmquist–Luenberger index43. All the evaluated DMUs are included in the global reference set, which avoids the phenomenon of infeasible solution in VRS. The global reference set constructed in this paper is as Eqs. (10)–(11):

$$ Q^{G} left( x right) = Q^{1} left( {x^{1} } right) cup Q^{2} left( {x^{2} } right) cup cdots cup Q^{T} left( {x^{T} } right) $$

(10)

$$ Q^{t} left( {x^{t} } right) = left{ {left( {y^{t} ,f^{t} } right)left| {x^{t} ;can;produce} right.;left( {y^{t} ,f^{t} } right)} right} $$

(11)

This paper takes MML index as the LHG.

$$ begin{aligned} MML_{t – 1}^{t} & = sqrt {frac{{1 – D_{t – 1} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{t – 1} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} times frac{{1 – D_{t} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{t} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}}} & = sqrt {frac{{1 – D_{t – 1} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}}{{1 – D_{t} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} times frac{{1 – D_{t – 1} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{t} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}} & ;;;;; times frac{{1 – D_{t} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{t – 1} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} end{aligned} $$

(12)

Next, it further decompose the MML index into efficiency change (EC) and technology change (TC). The specific formula is shown in Eqs. (13)–(14):

$$ TC_{t – 1}^{t} = sqrt {frac{{1 – D_{t – 1} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}}{{1 – D_{t} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} times frac{{1 – D_{t – 1} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{t} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}} $$

(13)

$$ EC_{t – 1}^{t} = frac{{1 – D_{t} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{t – 1} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} $$

(14)

where (left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} } right)) and (left( {x^{t} ,y^{t} ,f^{t} } right)) represent the input, expected output and unexpected output of t-1 and t, respectively. (TC_{t – 1}^{t}) is the devotion to LHG raise of DMU’s technical progress from (t – 1) to (t). And (EC_{t – 1}^{t}) represents the devotion to LHG raise of DMU’s efficiency improvement from (t – 1) to (t). The higher the value is, the larger the devotion is. The (MML) index is recorded as (MI). The value of (MI) is the LHG. The green total factor productivity index of laying hens breeding under the common frontier and scale frontier are as Eqs. (15)–(16):

$$ metaMI_{t – 1}^{t} = sqrt {frac{{1 – D_{{_{t – 1} }}^{m} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{{_{t – 1} }}^{m} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} times frac{{1 – D_{{_{t} }}^{m} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{{_{t} }}^{m} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}}} $$

(15)

$$ groupMI_{t – 1}^{t} = sqrt {frac{{1 – D_{{_{t – 1} }}^{g} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{{_{t – 1} }}^{g} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}} times frac{{1 – D_{{_{t} }}^{g} left( {x^{t} ,y^{t} ,f^{t} ;y^{t} , – f^{t} } right)}}{{1 – D_{{_{t} }}^{g} left( {x^{t – 1} ,y^{t – 1} ,f^{t – 1} ;y^{t – 1} , – f^{t – 1} } right)}}} $$

(16)

For the DMUs with scale heterogeneity, we can measure the technology gap between the group frontier and the common frontier, which is caused by the specific group structure.

Data and variables

Based on the research of the existing literature36, this paper selects five indexes to build the input–output indicator system. Details are as below:

  1. 1.

    Input variables:

    1. (1)

      Quantity of concentrated forage. Mainly includes seeds of crops and their by-products.

    2. (2)

      Quantity of grain consumption. Quantity of grain consumed is the quantity of grain consumed by laying hens when they are raised. For example: corn, sorghum, broken rice, wheat, barley, wheat bran, etc.

    3. (3)

      Material expenses. The sum of water and fuel power costs, labor costs, and medical epidemic prevention fees. Water and fuel power costs include water, electricity, coal and other fuel power costs; labor costs mean the human management cost of each laying hen from the brood stage to the laying stage; medical and epidemic prevention costs include the cost of disease prevention and control.

  2. 2.

    Positive output Main product production, which is the egg production per layer.

  3. 3.

    Negative output Total discharge. According to the calculation method of The Manual of Pollutant Discharge Coefficient, Eq. (17) is used to calculate the COD, TN, and the TP of each layer. Then, according to the calculation method of class GB3838-2002 water quality standard in V, Eq. (18) is used to calculate the total discharge.

$$ POLLUTANTS = FP(FD) times Days $$

(17)

$$ TOTAL , POLLUTANTS = frac{COD}{{40}} + frac{TN}{2} + frac{TP}{{0.4}} $$

(18)

where, (FP(FD)) is the pollution discharge coefficient and the (Days) is the average raising days. Descriptive statistics of input and output indicators are shown in Table 1.

Table 1 Descriptive statistics of input and output indicators.
Full size table

The quantity of concentrate, the quantity of food consumed, the cost of labor, the cost of medical treatment all come from “National Agricultural Product Cost and Benefit Data Compilation”. The pollutant discharge coefficient of laying hens is derived from “The Manual of Pollutant Discharge Coefficient”. According to the definition of scale in above two materials, a small scale 300–1000 laying hens, a medium scale 1000–10,000 laying hens, and a large scale greater than 10,000 laying hens are grouped to calculate cost efficiency.

From 2004 to 2018, this paper selects 24 major egg-producing provinces (municipalities) in China as samples, after eliminating singular data in the three scales and averaging the missing data, the final small-scale group is left with 7 provinces including Liaoning, Shandong, Henan, Heilongjiang, Jilin, Shanxi, and Shaanxi; the medium-scale group is the remaining 21 provinces of Beijing, Hebei, Jiangsu, Liaoning, Shandong, Tianjin, Zhejiang, Anhui, Henan, Heilongjiang, Jilin, Hubei, Inner Mongolia, Shanxi, Yunnan, Gansu, Ningxia, Shaanxi, Sichuan, Xinjiang, Chongqing; the large-scale group has 18 provinces, including Beijing, Fujian, Guangdong, Henan, Jiangsu, Liaoning, Shandong, Tianjin, Anhui, Henan, Heilongjiang, Hubei, Jilin, Shanxi, Yunnan, Gansu, Sichuan and Chongqing.

As is shown in Table 2, after dividing the provinces by region, the eastern region has 10 provinces (municipalities): Liaoning, Shandong, Beijing, Hebei, Jiangsu, Tianjin, Zhejiang, Fujian, Guangdong, Henan. The central region has 7 provinces (autonomous region): Henan, Heilongjiang, Jilin, Shanxi, Anhui, Hubei, Inner Mongolia. The western region has 7 provinces (municipalities): Shaanxi, Gansu, Ningxia, Sichuan, Xinjiang, Chongqing, Yunnan.

Table 2 Samples selected from 2004–2018.
Full size table


Source: Ecology - nature.com

Using excess heat to improve electrolyzers and fuel cells

Machine learning, harnessed to extreme computing, aids fusion energy development