Abstract
The Winsorized mean is a well-known robust estimator of the population mean. It can also be seen as a symmetric aggregation function (in fact, it is an ordered weighted averaging operator), which means that the information sources (for instance, criteria or experts’ opinions) have the same importance. However, in many practical applications (for instance, in many multiattribute decision making problems) it is necessary to consider that the information sources have different importance. For this reason, in this paper we propose a natural generalization of the Winsorized means so that the sources of information can be weighted differently. The new functions, which we will call Winsorized weighted means, are a specific case of the Choquet integral and they are analyzed through several indices for which we give closed-form expressions: the orness degree, k-conjunctiveness and k-disjunctiveness indices, veto and favor indices, Shapley values and interaction indices. We also provide a closed-form expression for the Möbius transform and we show how we can aggregate data so that each information source has the desired weighting and outliers have no influence in the aggregated value.
Similar content being viewed by others
Notes
In this framework, outliers may be due to the fact that the same subject may have been taught by different teachers, students may have been ill, or they may have copied answers, etc.
In general, the pair (r, s) is obtained by taking \(r=\max r_i\) and \(s=\max s_i\), where \((r_i,s_i)\) is the pair used for removing the outlier of the ith alternative.
Notice that we use the convention \(\sum _{t=p}^{q} f(t) = 0\) when \(p> q\).
References
Aggarwal CC (2017) Outlier analysis, 2nd edn. Springer, Cham
Bai C, Zhang R, Song C, Wu Y (2017) A new ordered weighted averaging operator to obtain the associated weights based on the principle of least mean square errors. Int J Intell Syst 32(3):213–226
Barnett V, Lewis T (1994) Outliers in statistical data, 3rd edn. Wiley, Chichester
Beliakov G (2018) Comparing apples and oranges: the weighted OWA function. Int J Intell Syst 33(5):1089–1108
Beliakov G, Dujmović J (2016) Extension of bivariate means to weighted means of several arguments by using binary trees. Inf Sci 331:137–147
Choquet G (1953) Theory of capacities. Ann Inst Fourier 5:131–295
Denneberg D (1994) Non-additive measures and integral. Kluwer Academic Publisher, Dordrecht
Dixon WJ (1960) Simplified estimation from censored normal samples. Ann Math Stat 31(2):385–391
Dubois D, Koning JL (1991) Social choice axioms for fuzzy set aggregation. Fuzzy Sets Syst 43(3):257–274
Grabisch M (1995) Fuzzy integral in multicriteria decision making. Fuzzy Sets Syst 69(3):279–298
Grabisch M (1997) \(k\)-order additive discrete fuzzy measures and their representation. Fuzzy Sets Syst 92(2):167–189
Grabisch M (2016) Set functions, games and capacities in decision making, theory and decision library, series C, vol 46. Springer, Berlin
Grabisch M, Labreuche C (2010) A decade of application of the Choquet and Sugeno integrals in multi-criteria decision aid. Ann Oper Res 175(1):247–286
Grabisch M, Labreuche C (2016) Fuzzy measures and integrals in MCDA. In: Greco S, Ehrgott M, Figueira RJ (eds) Multiple criteria decision analysis: state of the art surveys, international series in operations research and management science, vol 233, 2nd edn. Springer, New York, pp 553–603
Grabisch M, Marichal J, Mesiar R, Pap E (2009) Aggregation functions. Cambridge University Press, Cambridge
Harsanyi JC (1959) A bargaining model for cooperative \(n\)-person games. In: Tucker AW, Luce RD (eds) Contributions to the theory of games. Vol. 4, Annals of Mathematics Studies, vol 40. Princeton University Press, Princeton, pp 325–355
Heilpern S (2002) Using Choquet integral in economics. Stat Pap 43(1):53–73
Hoitash U, Hoitash R (2009) Conflicting objectives within the board: evidence from overlapping audit and compensation committee members. Group Decis Negot 18(1):57–73
Huber PJ, Ronchetti EM (2009) Robust statistics, 2nd edn. Wiley, Hoboken
Iglewicz B, Hoaglin D (1993) How to detect and handle outliers. ASQC Quality Press, Milwaukee
Komorníková M, Mesiar R (2011) Aggregation functions on bounded partially ordered sets and their classification. Fuzzy Sets Syst 175(1):48–56
Lenormand M (2018) Generating OWA weights using truncated distributions. Int J Intell Syst 33(4):791–801
Leys C, Ley C, Klein O, Bernard P, Licata L (2013) Detecting outliers: do not use standard deviation around the mean, use absolute deviation around the median. J Exp Soc Psychol 49(4):764–766
Liu X (2011) A review of the OWA determination methods: classification and some extensions. In: Yager RR, Kacprzyk J, Beliakov G (eds) Recent developments in the ordered weighted averaging operators: theory and practice. Springer, Berlin, pp 49–90
Llamazares B (2007) Choosing OWA operator weights in the field of social choice. Inf Sci 177(21):4745–4756
Llamazares B (2013) An analysis of some functions that generalizes weighted means and OWA operators. Int J Intell Syst 28(4):380–393
Llamazares B (2015a) Constructing Choquet integral-based operators that generalize weighted means and OWA operators. Inf Fus 23:131–138
Llamazares B (2015b) A study of SUOWA operators in two dimensions. Math Probl Eng 2015: Article ID 271,491
Llamazares B (2016a) A behavioral analysis of WOWA and SUOWA operators. Int J Intell Syst 31(8):827–851
Llamazares B (2016b) SUOWA operators: constructing semi-uninorms and analyzing specific cases. Fuzzy Sets Syst 287:119–136
Llamazares B (2018a) Closed-form expressions for some indices of SUOWA operators. Inf Fus 41:80–90
Llamazares B (2018b) Construction of Choquet integrals through unimodal weighting vectors. Int J Intell Syst 33(4):771–790
Llamazares B (2019a) SUOWA operators: a review of the state of the art. Int J Intell Syst 34(5):790–818
Llamazares B (2019b) SUOWA operators: an analysis of their conjunctive/disjunctive character. Fuzzy Sets Syst 357:117–134
Marichal JL (1998) Aggregation operators for multicriteria decision aid. Ph.D. thesis, University of Liège
Marichal JL (2004) Tolerant or intolerant character of interacting criteria in aggregation by the Choquet integral. Eur J Oper Res 155(3):771–791
Marichal JL (2007) \(k\)-intolerant capacities and Choquet integrals. Eur J Oper Res 177(3):1453–1468
Murofushi T, Soneda S (1993) Techniques for reading fuzzy measures (iii): interaction index. In: Proceedings of the 9th fuzzy systems symposium, Sapporo (Japan), pp 693–696
Murofushi T, Sugeno M (1991) A theory of fuzzy measures. Representation, the Choquet integral and null sets. J Math Anal Appl 159(2):532–549
Murofushi T, Sugeno M (1993) Some quantities represented by the Choquet integral. Fuzzy Sets Syst 56(2):229–235
Owen G (1972) Multilinear extensions of games. Manag Sci 18(5–part–2):64–79
Riordan J (1968) Combinatorial identities. Wiley, New York
Rota GC (1964) On the foundations of combinatorial theory I. Theory of Möbius functions. Z Wahrscheinlichkeitstheorie Verwandte Geb 2(4):340–368
Seo S (2006) A review and comparison of methods for detecting outliers in univariate data sets. Master’s thesis, University of Pittsburgh
Shapley LS (1953) A value for \(n\)-person games. In: Kuhn H, Tucker AW (eds) Contributions to the theory of games, vol 2. Princeton University Press, Princeton, pp 307–317
Sugeno M (1974) Theory of fuzzy integrals and its applications. Ph.D. thesis, Tokyo Institute of Technology
Torra V (1997) The weighted OWA operator. Int J Intell Syst 12(2):153–166
Tukey JW (1977) Exploratory data analysis. Addison-Wesley, Reading
Wainer H (1976) Robust statistics: a survey and some prescriptions. J Educ Stat 1(4):285–312
Wilcox RR (2012) Modern statistics for the social and behavioral sciences: a practical introduction. CRC Press, Boca Raton
Wilcox RR, Keselman HJ (2003) Modern robust data analysis methods: measures of central tendency. Psychol Methods 8(3):254–274
Yager RR (1988) On ordered weighted averaging operators in multicriteria decision making. IEEE Trans Syst Man Cybern 18(1):183–190
Zhang Z, Xu Z (2014) Analysis on aggregation function spaces. Int J Uncertain Fuzziness Knowl Based Syst 22(05):737–747
Acknowledgements
The author is grateful to two anonymous referees for valuable suggestions and comments. This work is partially supported by the Spanish Ministry of Economy and Competitiveness (Project ECO2016-77900-P) and ERDF.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Proofs
Proofs
We first recall the definition and some properties of binomial coefficients (see, for instance, Riordan (1968, pp. 1–3) and Grabisch (2016, p. 3).
Remark 2
Let \(m\in \mathbb {N}\) and \(k \in \mathbb {Z}\). Then:
-
1.
\(\displaystyle \left( {\begin{array}{c}m\\ k\end{array}}\right) ={\left\{ \begin{array}{ll} \displaystyle \frac{m!}{k!(m-k)!} &{} \text {if } 0\le k \le m,\\ 0 &{} \text {otherwise}. \end{array}\right. }\)
-
2.
\(\displaystyle \left( {\begin{array}{c}m\\ k\end{array}}\right) = \left( {\begin{array}{c}m\\ m-k\end{array}}\right) \).
-
3.
If \(0 \le k \le m\), then \(\displaystyle \sum _{j=0}^{k} (-1)^{j} \left( {\begin{array}{c}m\\ j\end{array}}\right) = (-1)^{k} \left( {\begin{array}{c}m-1\\ k\end{array}}\right) \).
The following simple remarks on summation will be useful in some of the proofs.
Remark 3
Let \(p,q\in \mathbb {N}\), with \(p\le q+1\).Footnote 5 Then:
Remark 4
Let \(\varvec{p}\) be a weighting vector. If \(\varnothing \subsetneq A \subseteq N\) and \(t\ge 1\), then
In particular, when \(A = N\) we have
Remark 5
Let \(\varvec{p}\) be a weighting vector and \(j\in N\). If \(t\ge 1\), then
Remark 6
Let \(\varvec{p}\) be a weighting vector and \(j,k\in N\). If \(t\ge 1\), then
Proof of Proposition 1
Let \(\varvec{p}\) be a weighting vector and \((r,s)\in \mathcal {R}\). Since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish two cases:
-
1.
If \(r+s=n-1\), then
$$\begin{aligned} {{\,\mathrm{orness}\,}}\big (M_{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-1}\sum _{t=s+1}^{n-1} 1 = \frac{n-s-1}{n-1} = \frac{r}{n-1}. \end{aligned}$$ -
2.
If \(r+s<n-1\), then, by Remarks 4 and 3, we have
$$\begin{aligned}&{{\,\mathrm{orness}\,}}\big (M_{\varvec{p}}^{(r,s)}\big )\\&\quad = \frac{1}{n-1} \left( \sum _{t=s+1}^{n-r-1}\frac{1}{\left( {\begin{array}{c}n\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N \\ |T|=t \end{array}} \sum _{i\in T} p_i + r \right) \\&\quad = \frac{1}{n-1}\left( \frac{1}{n}\sum _{t=s+1}^{n-r-1} t + r \right) \\&\quad = \frac{1}{n-1}\left( \frac{(n-r)(n-r-1) - s(s+1)}{2n} + r \right) \\&\quad = \frac{1}{n-1} \frac{n(n-1) + r(r+1) - s(s+1)}{2n}\\&\quad = \frac{1}{2} + \frac{r(r+1) - s(s+1)}{2n(n-1)}. \end{aligned}$$Notice that, when \(r+s=n-1\), the previous expression returns \(r/(n-1)\). So, it is also valid in the case \(r+s=n-1\).
\(\square \)
Proof of Proposition 2
Let \(\varvec{p}\) be a weighting vector, \((r,s)\in \mathcal {R}\), and \(k\in N{\setminus } \{n\}\). Since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish the following cases:
-
1.
If \(n-k \le s\) (or, equivalently, \(k\ge n-s\)), then
$$\begin{aligned} {{\,\mathrm{conj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big ) = 1. \end{aligned}$$ -
2.
If \(s< n-k < n-r\) (or, equivalently, \(r< k < n-s\)), by Remarks 4 and 3 we have
$$\begin{aligned} {{\,\mathrm{conj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big )&= 1 - \frac{1}{n-k} \sum _{t=s+1}^{n-k}\frac{1}{\left( {\begin{array}{c}n\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N \\ |T|=t \end{array}} \sum _{i\in T} p_{i} = 1 - \frac{1}{n-k} \sum _{t=s+1}^{n-k} \frac{t}{n} \\&= 1 - \frac{1}{n-k} \frac{(n-k)(n-k+1) - s(s+1)}{2n}\\&= 1 - \frac{n-k+1}{2n} + \frac{s(s+1)}{2n(n-k)} \\&= \frac{n+k-1}{2n} + \frac{s(s+1)}{2n(n-k)}\\&= \frac{n(n-1) + s(s+1) - k(k-1)}{2n(n-k)}. \end{aligned}$$ -
3.
If \(n-k \ge n-r\) (or, equivalently, \(k \le r\)), we distinguish two cases:
-
(a)
If \(r+s=n-1\), then
$$\begin{aligned} {{\,\mathrm{conj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big ) = 1 - \frac{1}{n-k} \sum _{t=n-r}^{n-k} 1 = 1 - \frac{r-k+1}{n-k} = \frac{n-(r+1)}{n-k}. \end{aligned}$$ -
(b)
If \(r+s<n-1\), then
$$\begin{aligned}&{{\,\mathrm{conj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big )\\&= 1 - \frac{1}{n-k} \left( \sum _{t=s+1}^{n-r-1}\frac{1}{\left( {\begin{array}{c}n\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N \\ |T|=t \end{array}} \sum _{i\in T} p_{i} + r-k+1 \right) \\&= 1 - \frac{1}{n-k} \left( \sum _{t=s+1}^{n-r-1} \frac{t}{n} + r-k+1 \right) \\&= 1 - \frac{1}{n-k} \left( \frac{(n-r)(n-r-1)-s(s+1)}{2n} + r-k+1 \right) \\&= 1 - \frac{1}{n-k} \left( \frac{n+1-2k}{2} + \frac{r(r+1)-s(s+1)}{2n} \right) \\&= \frac{n(n-1) + s(s+1) - r(r+1)}{2n(n-k)} . \end{aligned}$$Notice also that, when \(r+s=n-1\), the previous expression returns \((n-r-1)/(n-k)\). So, it is also valid in the case \(r+s=n-1\).
-
(a)
\(\square \)
Proof of Proposition 3
Let \(\varvec{p}\) be a weighting vector, \((r,s)\in \mathcal {R}\), and \(k\in N{\setminus } \{n\}\). Since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish the following cases:
-
1.
If \(k \ge n-r\), then
$$\begin{aligned} {{\,\mathrm{disj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-k}\sum _{t=k}^{n-1} 1 = 1. \end{aligned}$$ -
2.
If \(s< k < n-r\), by Remarks 4 and 3 we have
$$\begin{aligned} {{\,\mathrm{disj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big )&= \frac{1}{n-k} \left( \sum _{t=k}^{n-r-1}\frac{1}{\left( {\begin{array}{c}n\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N \\ |T|=t \end{array}} \sum _{i\in T} p_{i} + r \right) \\&= \frac{1}{n-k} \left( \sum _{t=k}^{n-r-1} \frac{t}{n} + r \right) \\&= \frac{1}{n-k} \left( \frac{(n-r)(n-r-1) - k(k-1)}{2n} + r \right) \\&= \frac{n(n-1)+r(r+1)-k(k-1)}{2n(n-k)}. \end{aligned}$$ -
3.
If \(k\le s\), we distinguish two cases:
-
(a)
If \(r+s=n-1\), then
$$\begin{aligned} {{\,\mathrm{disj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-k}\sum _{t=n-r}^{n-1} 1 = \frac{r}{n-k}. \end{aligned}$$ -
(b)
If \(r+s<n-1\), then
$$\begin{aligned} {{\,\mathrm{disj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-k} \left( \sum _{t=s+1}^{n-r-1}\frac{1}{\left( {\begin{array}{c}n\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N \\ |T|=t \end{array}} \sum _{i\in T} p_{i} + r \right) . \end{aligned}$$Notice that the above expression coincides with that of the second item when \(k=s+1\). Therefore,
$$\begin{aligned} {{\,\mathrm{disj}\,}}_{k}\!\big (M_{\varvec{p}}^{(r,s)}\big ) = \frac{n(n-1) + r(r+1) - s(s+1)}{2n(n-k)}. \end{aligned}$$Notice also that, when \(r+s=n-1\), the previous expression returns \(r/(n-k)\). So, it is also valid in the case \(r+s=n-1\).
-
(a)
\(\square \)
Proof of Proposition 4
Let \(\varvec{p}\) be a weighting vector, \((r,s)\in \mathcal {R}\) and \(j\in N\). Since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish two cases:
-
1.
If \(r+s=n-1\), then
$$\begin{aligned} {{\,\mathrm{veto}\,}}\!\big (M_{\varvec{p}}^{(r,s)},j\big ) = 1 - \frac{1}{n-1}\sum _{t=s+1}^{n-1} 1 = 1 - \frac{r}{n-1} = \frac{s}{n-1}. \end{aligned}$$ -
2.
If \(r+s<n-1\), then, by Remarks 5 and 3, we get
$$\begin{aligned} {{\,\mathrm{veto}\,}}\!\big (M_{\varvec{p}}^{(r,s)},j\big )&= 1 - \frac{1}{n-1}\left( \sum _{t=s+1}^{n-r-1} \frac{1}{\left( {\begin{array}{c}n-1\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N{\setminus } \{j\} \\ |T|=t \end{array}} \sum _{i\in T} p_{i} + r \right) \\&= 1 - \frac{1}{n-1}\left( \frac{1-p_{j}}{n-1} \sum _{t=s+1}^{n-r-1} t + r \right) \\&= 1 - \frac{r}{n-1} - \frac{(1-p_{j})\big ((n-r)(n-r-1)-s(s+1)\big )}{2(n-1)^2}. \end{aligned}$$Notice that, when \(r+s=n-1\), the previous expression returns \(s/(n-1)\). So, it is also valid in the case \(r+s=n-1\).
\(\square \)
Proof of Proposition 5
Let \(\varvec{p}\) be a weighting vector, \((r,s)\in \mathcal {R}\) and \(j\in N\). Since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish two cases:
-
1.
If \(r+s=n-1\), then
$$\begin{aligned} {{\,\mathrm{favor}\,}}\!\big (M_{\varvec{p}}^{(r,s)},j\big ) = \frac{1}{n-1}\sum _{t=s}^{n-1} 1 - \frac{1}{n-1} = \frac{n-s-1}{n-1}= \frac{r}{n-1}. \end{aligned}$$ -
2.
If \(r+s<n-1\), then, by Remarks 5 and 3, we get
$$\begin{aligned}&{{\,\mathrm{favor}\,}}\!\big (M_{\varvec{p}}^{(r,s)},j\big )\\&\quad = \frac{1}{n-1} \left( \sum _{t=s}^{n-r-2} \frac{1}{\left( {\begin{array}{c}n-1\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N{\setminus } \{j\} \\ |T|=t \end{array}} \left( \sum _{i\in T} p_{i} + p_{j}\right) + r + 1 \right) - \frac{1}{n-1}\\&\quad = \frac{1}{n-1} \left( \frac{1-p_{j}}{n-1} \sum _{t=s}^{n-r-2} t + p_{j}(n-r-s-1) + r \right) \\&\quad = \frac{(1-p_{j})\big ((n-r-2)(n-r-1)-s(s-1) + 2r(n-1) \big )}{2(n-1)^2}\\&\qquad + \frac{p_{j}(n-s-1)}{n-1}\\&\quad = (1-p_{j}) \frac{(n-1)(n-2) + r(r+1) - s(s-1)}{2(n-1)^2} + p_{j}\left( 1 - \frac{s}{n-1}\right) . \end{aligned}$$Notice that, when \(r+s=n-1\), the previous expression returns \(r/(n-1)\). So, it is also valid in the case \(r+s=n-1\).
\(\square \)
Proof of Proposition 6
Let \(\varvec{p}\) be a weighting vector, \((r,s)\in \mathcal {R}\) and \(j\in N\). Since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish two cases:
-
1.
If \(r+s=n-1\), then
$$\begin{aligned} \phi _{j}\big (\mu _{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n}\sum _{t=s}^{n-1} 1 - \frac{1}{n}\sum _{t=s+1}^{n-1} 1 = \frac{1}{n}. \end{aligned}$$ -
2.
If \(r+s<n-1\), then, by Remark 5, we have
$$\begin{aligned} \phi _{j}\big (\mu _{\varvec{p}}^{(r,s)}\big )&= \frac{1}{n} \left( \sum _{t=s}^{n-r-2}\frac{1}{\left( {\begin{array}{c}n-1\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N{\setminus } \{j\} \\ |T|=t \end{array}} \left( p_{j} + \sum _{i\in T} p_i\right) + r + 1 \right) \\&\quad - \frac{1}{n} \left( \sum _{t=s+1}^{n-r-1}\frac{1}{\left( {\begin{array}{c}n-1\\ t\end{array}}\right) } \sum _{\begin{array}{c} T\subseteq N{\setminus } \{j\} \\ |T|=t \end{array}} \sum _{i\in T} p_i + r \right) \\&= \frac{1}{n} \left( (n-r-s-1)p_{j} + \frac{1-p_{j}}{n-1} (s-(n-r-1)) + 1 \right) \\&= \frac{1}{n}\left( \frac{s+r + n(n-r-s-1)p_j}{n-1}\right) \\&= \frac{r+s}{n-1}\,\frac{1}{n} + \frac{n-1-r-s}{n-1} p_{j}\\&= \frac{r+s}{n-1}\,\frac{1}{n} + \left( 1 - \frac{r+s}{n-1}\right) p_{j}. \end{aligned}$$Notice that, when \(r+s=n-1\), the previous expression returns 1 / n. So, it is also valid in the case \(r+s=n-1\).
\(\square \)
Proof of Proposition 7
Let \(\varvec{p}\) be a weighting vector, \((r,s)\in \mathcal {R}\) and \(j,k\in N\). Notice that \(I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big )\) can be written as
where \(I_{K}\), \(K \subseteq N\), is defined by
Since \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish two cases:
-
1.
If \(r+s=n-1\), then it is easy to check that \(I_{\{j,k\}}, I_{\{j\}}, I_{\{k\}}\), and \(I_{\varnothing }\) take the following values:
$$\begin{aligned} I_{\{j,k\}}= & {} {\left\{ \begin{array}{ll} \sum \nolimits _{t=0}^{n-2} 1 = n-1 &{} \text {if } s=0, \\ \sum \nolimits _{t=s-1}^{n-2} 1 = n - s&{} \text {otherwise}, \end{array}\right. } \\ I_{\{j\}}= & {} I_{\{k\}} = \sum \limits _{t=s}^{n-2} 1 = n - 1 - s, \\ I_{\varnothing }= & {} {\left\{ \begin{array}{ll} \sum \nolimits _{t=s+1}^{n-2} 1 = n-2-s &{} \text {if } s< n-1, \\ 0 &{} \text {otherwise}. \end{array}\right. } \end{aligned}$$We now calculate \(I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big )\) taking into account the different values of \(I_{\{j,k\}}\), \(I_{\{j\}}\), \(I_{\{k\}}\), and \(I_{\varnothing }\). We distinguish three cases:
-
(a)
If \(s=0\), which is equivalent to \(r=n-1\), we get
$$\begin{aligned} I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-1} (n-1 -2(n-1)+n-2) = - \frac{1}{n-1}. \end{aligned}$$ -
(b)
If \(0< s < n-1\) we have
$$\begin{aligned} I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-1} (n-s -2(n-1-s) + n-2-s) = 0. \end{aligned}$$ -
(c)
If \(s=n-1\), which is equivalent to \(r=0\), we obtain
$$\begin{aligned} I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big ) = \frac{1}{n-1}. \end{aligned}$$Therefore,
$$\begin{aligned} I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big ) = {\left\{ \begin{array}{ll} \frac{1}{n-1} &{} \text {if } r=0, \\ -\frac{1}{n-1} &{} \text {if } r=n-1, \\ 0 &{} \text {otherwise}. \end{array}\right. } \end{aligned}$$(4) -
2.
If \(r+s<n-1\), then, by Remark 6, we can see that \(I_{\{j,k\}}\), \(I_{\{j\}}\), \(I_{\{k\}}\), and \(I_{\varnothing }\) take the following values:
$$\begin{aligned} I_{\{j,k\}}= & {} {\left\{ \begin{array}{ll} \sum \nolimits _{t=0}^{n-2} 1 = n-1 &{} \text {if } s=0 \text { and } r=n-2,\\ \sum \nolimits _{t=0}^{n-r-3} \left( p_{j} + p_{k} + \frac{t(1-p_{j}-p_{k})}{n-2}\right) + r + 1 &{} \text {if } s=0 \text { and } r< n-2,\\ \sum \nolimits _{t=s-1}^{n-r-3} \left( p_{j} + p_{k} + \frac{t(1-p_{j}-p_{k})}{n-2}\right) + r + 1&{} \text {otherwise}, \end{array}\right. } \\ I_{\{j\}}= & {} \sum _{t=s}^{n-r-2} \left( p_{j} + \frac{t(1-p_{j}-p_{k})}{n-2}\right) + r, \\ I_{\{k\}}= & {} \sum _{t=s}^{n-r-2} \left( p_{k} + \frac{t(1-p_{j}-p_{k})}{n-2}\right) + r, \\ I_{\varnothing }= & {} {\left\{ \begin{array}{ll} 0 &{} \text {if } r=0 \text { and } s=n-2,\\ \sum \nolimits _{t=s+1}^{n-2} \frac{t(1-p_{j}-p_{k})}{n-2} &{} \text {if } r=0 \text { and } s < n-2,\\ \sum \nolimits _{t=s+1}^{n-r-1} \frac{t(1-p_{j}-p_{k})}{n-2} + r - 1&{} \text {otherwise}, \end{array}\right. } \end{aligned}$$We now calculate \(I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big )\) taking into account the different values of \(I_{\{j,k\}}\), \(I_{\{j\}}\), \(I_{\{k\}}\), and \(I_{\varnothing }\). For instance, when \(s=0\) and \(r=n-2\) we get
$$\begin{aligned}&I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big )\\&\quad = \frac{1}{n-1}\left( n-1 - p_{j} - (n-2) - p_{k} - (n-2) + \frac{1-p_{j}-p_{k}}{n-2} + n-3\right) \\&\quad = \frac{1}{n-1}\frac{1-(n-1)(p_{j}+p_{k})}{n-2} = -\frac{1}{n-2}\left( p_{j}+p_{k} -\frac{1}{n-1}\right) . \end{aligned}$$Once all the cases have been analyzed (to avoid tedious calculations, the remaining cases are left to the reader), we have
$$\begin{aligned} I_{jk}\big (\mu _{\varvec{p}}^{(r,s)}\big ) = {\left\{ \begin{array}{ll} \frac{1}{n-2}\left( p_{j}+p_{k} -\frac{1}{n-1}\right) &{} \text {if } r = 0 \text { and } 0< s< n-1,\\ -\frac{1}{n-2}\left( p_{j}+p_{k} -\frac{1}{n-1}\right) &{} \text {if } s = 0 \text { and } 0< r < n-1,\\ 0 &{} \text {otherwise}. \end{array}\right. } \end{aligned}$$(5)
-
(a)
Expressions (4) and (5) together establish the truth of Proposition 7. \(\square \)
Proof of Proposition 8
Let \(\varvec{p}\) be a weighting vector and \((r,s)\in \mathcal {R}\). Given \(A\subseteq N\), since
and \(\mu _{\varvec{p}}^{(r,s)}\) is given by expression (1) (or expression (2) when \(r+s=n-1\)), we distinguish the following cases:
-
1.
If \(|A| \le s\), then
$$\begin{aligned} m^{\mu _{\varvec{p}}^{(r,s)}}(A) = 0. \end{aligned}$$ -
2.
If \(s< |A| < n-r\), by Remark 4 and the third item of Remark 2, we get
$$\begin{aligned}&m^{\mu _{\varvec{p}}^{(r,s)}}(A) \\&\quad = \sum _{t=s+1}^{|A|} (-1)^{|A|-t} \sum _{\begin{array}{c} T\subseteq A \\ |T|=t \end{array}} \sum _{i\in T} p_{i} = \sum _{t=s+1}^{|A|} (-1)^{|A|-t} \left( {\begin{array}{c}|A|-1\\ t-1\end{array}}\right) \sum _{i\in A} p_{i}\\&\quad =\! \left( \sum _{i\in A} p_{i} \!\!\right) \!\sum _{t=s+1}^{|A|} \!(-1)^{|A|-t} \left( {\begin{array}{c}|A|-1\\ |A|-t\end{array}}\right) = \left( \sum _{i\in A} p_{i} \!\right) \!\sum _{j=0}^{|A|-s-1} \!(-1)^{j} \left( {\begin{array}{c}|A|-1\\ j\end{array}}\right) \\&\quad =\! \left( \sum _{i\in A} p_{i}\!\! \right) \! (-1)^{|A|-s-1} \!\left( {\begin{array}{c}|A|-2\\ |A|-s-1\end{array}}\right) = (-1)^{|A|-s-1} \!\left( {\begin{array}{c}|A|-2\\ s-1\end{array}}\right) \!\! \left( \sum _{i\in A} p_{i} \!\!\right) \!\!. \end{aligned}$$Notice that when \(s=0\) we have \(m^{\mu _{\varvec{p}}^{(r,s)}}(A) = 0\).
-
3
If \(|A| \ge n-r\), we distinguish two cases:
-
(a)
If \(r+s=n-1\), then, by the third item of Remark 2 we have
$$\begin{aligned} m^{\mu _{\varvec{p}}^{(r,s)}}(A)&= \sum _{t=s+1}^{|A|} (-1)^{|A|-t} \left( {\begin{array}{c}|A|\\ t\end{array}}\right) = \sum _{t=s+1}^{|A|} (-1)^{|A|-t} \left( {\begin{array}{c}|A|\\ |A|-t\end{array}}\right) \\&= \sum _{j=0}^{|A|-s-1} (-1)^{j} \left( {\begin{array}{c}|A|\\ j\end{array}}\right) = (-1)^{|A|-s-1} \left( {\begin{array}{c}|A|-1\\ |A|-s-1\end{array}}\right) \\&= (-1)^{|A|-s-1} \left( {\begin{array}{c}|A|-1\\ s\end{array}}\right) . \end{aligned}$$ -
(b)
If \(r+s<n-1\), by Remark 4 and the third item of Remark 2, we have
$$\begin{aligned}&m^{\mu _{\varvec{p}}^{(r,s)}}(A)\\&\quad = \sum _{t=s+1}^{n-r-1} (-1)^{|A|-t} \left( {\begin{array}{c}|A|-1\\ t-1\end{array}}\right) \sum _{i\in A} p_{i} + \sum _{t=n-r}^{|A|} (-1)^{|A|-t} \left( {\begin{array}{c}|A|\\ t\end{array}}\right) \\&\quad = \left( \sum _{i\in A} p_{i} \right) \sum _{t=s+1}^{n-r-1} (-1)^{|A|-t} \left( {\begin{array}{c}|A|-1\\ |A|-t\end{array}}\right) + \sum _{t=n-r}^{|A|} (-1)^{|A|-t} \left( {\begin{array}{c}|A|\\ |A|-t\end{array}}\right) \\&\quad = \left( \sum _{i\in A} p_{i} \right) \sum _{j=|A|-n+r+1}^{|A|-s-1} (-1)^{j} \left( {\begin{array}{c}|A|-1\\ j\end{array}}\right) + \sum _{j=0}^{|A|-n+r} (-1)^{j} \left( {\begin{array}{c}|A|\\ j\end{array}}\right) \\&\quad = \left( \sum _{i\in A} p_{i} \right) \left( (-1)^{|A|-s-1} \left( {\begin{array}{c}|A|-2\\ |A|-s-1\end{array}}\right) \right. \\&\qquad \left. - (-1)^{|A|-n+r} \left( {\begin{array}{c}|A|-2\\ |A|-n+r\end{array}}\right) \right) + (-1)^{|A|-n+r} \left( {\begin{array}{c}|A|-1\\ |A|-n+r\end{array}}\right) \\&\quad = \left( \sum _{i\in A} p_{i} \right) \left( (-1)^{|A|-s-1} \left( {\begin{array}{c}|A|-2\\ s-1\end{array}}\right) - (-1)^{|A|-n+r} \left( {\begin{array}{c}|A|-2\\ n-r-2\end{array}}\right) \right) \\&\qquad + (-1)^{|A|-n+r} \left( {\begin{array}{c}|A|-1\\ n-r-1\end{array}}\right) . \end{aligned}$$Notice also that, when \(r+s=n-1\), the previous expression returns \((-1)^{|A|-s-1} \left( {\begin{array}{c}|A|-1\\ s\end{array}}\right) \). So, it is also valid in the case \(r+s=n-1\).
-
(a)
\(\square \)
Rights and permissions
About this article
Cite this article
Llamazares, B. An Analysis of Winsorized Weighted Means. Group Decis Negot 28, 907–933 (2019). https://doi.org/10.1007/s10726-019-09623-8
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10726-019-09623-8