The Bonferroni curve and Generalised Entropy Inequality Measures Effat Dalir Mehdi Yaghoobi Awal Riyabi Ali reza Hosseinzadeh Department of Statistics Department of Statistics Department of Statistics Islamic Azad University, Mashhad Branch Islamic Azad University, Gonabad Branch Islamic Azad University, Gonabad Branch Mashhad, Iran Gonabad, Iran Gonabad, Iran e_dalir_e@yahoo.com m_yaghoobiawal@yahoo.com hoseinzade1@yahoo.com Abstract-Lorenz curve and the Generalised Entropy inequality measures are popular tools for the analysis of income inequality which are commonly used in Thus we have: π ππ = ∑ππ =1 π₯π(π) π π ππ = , ∑π=1 π₯π ππ π empirical studies in parallel to each other.Another Generalised Intropy measures size have the following income inequality analysis tools, is the Bonferroni general form: Curve that was introduced by Bonferroni (1930). πΊπΈ(πΌ) = This paper first introduces the Generalised Entropy 1 1 π πΌ [ ∑ππ=1 ( π ) − 1] (1) πΌ 2 −πΌ π ππ inequality measures and explains their relationship with Where πΌ is a nonnegative sensitivity parameter. Lorenz curves. Then by introducing the Bonferroni Taking the limit of equation (1), as πΌ → 0,1 gives curve and expressing its relationship with Lorenz curve, TheilΜs L and T inequality measures respectively we derive the Generalised Intropy index by using the Bonferroni curve. Finally, we calculate these indices for a number of Bonferroni curve function forms. Key words: Lorenz curve -Bonferroni Curve -Generalized Intropy measures- Theil index. π ππ πΏ = ∑ ππ ln ( ) ππ (2) π=1 π ππ π = ∑ ππ ln ( ) ππ (3) π=1 Half the square of the coefficient of variation is as the Ι. Introduction result of πΌ = 2, and the Atkinson measure is calculated as follows: π Consider a population which includes j individual and all has nonnegative income.Also, suppose to 1 1 ππ π΄ = 1 − [ ∑( )1−π ]1−π π ππ (4) π=1 show k- th individual income with π₯π . If the Where π is an inequality aversion parameter and lies population is formed arbitrarily, as have n groups, between zero and infinity. then i-th group that consisting of (ππ )individual and is a share of income(ππ ) and a population share(ππ ). II. LORENZ CURVES AND INCOME SHARES Lorenz curves are one of the important indicators of how wealth is distributed. These curves relate the cumulative percentage of income with the cumulative population percent and the first time was proposed by Max Lorenz (1905). function and the partial mean value,for π ≤ π₯ Definition 1.2. Suppose π is a nonnegative random ,respectively.By putting πΉ(π₯) = π, Bonferroni curve variable with density function π (π₯) and cumulative parametric form is as follows: distribution function πΉ(π₯). Then we show the related Lorenz curve by πΏ(π)and define it as follows: πΉ1 [πΉ −1 (π)] 1 π −1 = ∫ πΉ (π‘)ππ‘ , 0 ≤ π ≤ 1 π ππ 0 π΅(π) = (11) 0 As π → 0, π΅(π)Takes the form , Hence the 0 πΏ(π) = πΉ−1 (π) 1 ∫ πΈ(π) 0 π₯π(π₯)ππ₯ , π ∈ [0,1] (5) Bonferroni curve does not always start from the origin of coordinates, since Rohde (2008) using the definition of definite ππ΅ ππ > 0,the graph of π΅(π) is strictly increasing but nothing can be said about the integrals with total Riemann showed that: sign of its second derivative.Hence ,Bonferroni Curve π 1 1 ππ πΏ = lim − ∑ ln( ) = − ∫ πππΏ′ (π)ππ π→∞ π ππ 0 (6) π=1 could be convex in some parts and concave in some others(Olemedo,2009). Comparing the relations (5) π 1 ππ ππ π = lim ∑ ( ) ln( ) π→∞ π ππ ππ and (11) we have: π΅: [0,1] → [0,1] → π΅(π) π=1 1 = ∫ πΏ′ (π) ln(πΏ′ (π)) ππ πΏ(π) ={ π 0 (7) 0 π 1 2 1 1 ππ πΆπ = lim ∑ (( )2 − 1) π→∞ 2 2 π ππ 0<π<1 (12) π=0 For some Bonferroni curve properties and its π=1 1 1 = ∫ (πΏ′ (π)2 − 1)ππ 2 0 relationship with Lorenz curves we can refer to (8) Pundir (2005). 1 π 1 ππ 1−π 1−π π΄ = lim 1 − [∑ ( ) ] π→∞ π ππ IV. Bonferroni curves and Generalised Intropy measures π=1 1 = 1 − [∫ πΏ′ (π)1−π ππ] 1 1−π According to previous sections and the Bonferroni (9) 0 III. BONFERRONI CURVE AND ITS RELASHIONSHIP WITH THE LORENZ CURVE curves’ relationships and Lorenz curve with each other we can extend equations obtained for GE inequality indices by Lorenz curves to the Bonferroni For continuous and nonnegative random variableπ, curve. with the cumulative distribution function πΉ(π₯)that is According to the formula (12) we have: πΏ′ (π) = π΅(π) + ππ΅′ (π) derived from at least two times, Bonferroni curve as (13) follows coordinates on the page is defined in the unit Placement this formula in relations with (6) to (9), we square. have: π΅(πΉ(π₯)) = ππ₯ πΉ1 (π₯) = π πΉ(π₯) 1 (10) Where π, ππ₯ , and πΉ1 (π₯) are the first moment of πabout zero,i.e. ,the first moment distribution πΏ = − ∫ ln( π΅(π) + ππ΅′ (π))ππ 0 (14) 1 as follows: π = ∫ (π΅(π) + ππ΅′ (π)) ln(π΅(π) 0 πΉ(π₯) = + ππ΅′ (π)) ππ (15) 1 2 1 1 πΆπ = ∫ ((π΅(π) + ππ΅′ (π))2 − 1)ππ 2 2 0 π΄=1− 1 [∫0 (π΅(π) + 1 ππ΅′ (π))1−π ]1−π (16) 1 , π₯ > 0, π > 0,0 < π < 1 1 ππ 1−π ( ) π₯ With some mathematical calculations, we get: π₯ (17) 5. Calculation of GE inequality measures based on Bonferroni curves ππ₯ = 1 ∫0 π‘ 1−π ππ‘ 1 π₯ −1 ∫0 π‘ 1−π = ππ‘ 1 ππ1−π π₯ → π΅(π) = 2−π 2−π ππ πππ π‘βπ πππππ‘ππππ (14) π‘π (17) π€π βππ£π: πΏ = 1 − π − ππ(π) , π In this section,at first we calculate Bonferroni curve = π(−1 + π + πππ. for the Pareto distribution type 1, and then we deduce the size of GE inequality by the relations (14) to (17). Then for some major distributions, calculation results in a tabular summary are 1 2 1 1 πΆπ = π2 − 2 6 − 4π 2 type 1, the probability distribution function is defined 1 ) 2−π , 1 1 =1−( π 1−π )1−π 2 (1 − π) + 1 presented. If the π random variable has the Pareto distribution (18) For some other specific distribution, Generalised Intropy measures are calculated based on the Bonferroni Curves.The summary results are given in Table 1. Table 1: Generalised Intropy measures from Bonferroni curve for the particular multiple TABLE 1 Function L Rectangular π₯ − (1 − π)π 2ππ π΅(π) = (1 − π) + ππ πΉ(π₯) = (1 − π)1−π 1 (ln ( ) (1 + π)1+π 2π + π + 1) T ln(π + 1) ( (π + 1)2 ) 4π 1 + π2 −( ) 4π (1 − π)2 − (ln(1 − π) 4π − 1) Power πΉ(π₯) = π₯ πΌ πΌ > 1,0 ≤ π₯ ≤ 1 π΅(π) = ππΌ 1 − (ln(π) − 1) πΌ 1 − ln ( + 1) πΌ ln ( 1+πΌ 1 )− πΌ πΌ+1 1 2 πΆπ 2 1 6 π2 A 1 − 1 ((π + 1)2−π 2π(2 − π) 1 − (1 − π)2−π )1−π 1 (πΌ + 1)2 ( − 1) 2 πΌ(2 + πΌ) (1+πΌ)1−π πΌ 1 1-(πΌ1−π(1−π+πΌ))1−π Uniform π₯−π F(x)= π−π a<x<b 2π + (π − π)π π΅(π) = π+π Pareto (Ι) 1 ππ 1 ( π₯ )1−π π₯ > 0, π > 0,0 < π <1 ππ1−π π΅(π) = 2−π πΉ(π₯) = −ln( 2π ) π+π 1 − π − ln(π) 2π 2πππ(π + π) π+π π (−1 + π 1 + πππ. ) 2−π 2π2 1 − π+π 2 1 1 π2 − 6 − 4π 2 1 − (( 1−( 2π 1−π 1 ) )1−π π+π 1 1 π 1−π )1−π (1 − π)2 + 1 REFERENCES [1] C.E. Bonferroni, Elementi di statistica general, Libereria Seber, Firenze, 1930. [2] L.J. Imedio Olemedo, E. Barcena Martin, and E.M. Parrado Gallardo, A wide class of inequality measures based on the Bonferroni curve, 2009. [3] M.O. Lorenz, Methods of measuring the concentration of wealth, Publication of American Statistical Association, 1905. [4] S. Pundir, S. Arora, and K. Jain, Bonferroni curve and the related statistical inference,Statistics and Probability Letters, 75 (2005), 140-150. [5] N. Rohd, Lorenz Curves and Generalised Entropy Inequality Measures, Modeling Income Distributions and Lorenz Curves, Springer, 2008.