1 vues

Transféré par biodunagbaje

Workplace for women

- Joan Scott- Genre
- Clio 113 14 La Fabrique Du Sexe Thomas Laqueur Et Aristote
- Eduquer contre l’homophobie dès l'école primaire
- PÊNIS DE TÊTE FILLES.pdf
- Trouble Dans Le Genre (Résumé Premier Chapitre)
- TD 2 Sexe et genre élève
- Untitled
- Le Sorbonnard Déchaîné n°39 (fev/mars 2014)
- Institution Mahefa
- Corrige Conc Gen 00
- Sondage Superstitions Francais
- biblio_hommes_femmes (1).pdf
- o Corpo Feminino No Cinema
- 1042812 Ar
- echantillonage
- Genre Et Inegalites Tema 18
- Mh0213845frc PDF.web
- Cp Collectif Education Lgbt - 16-09-2013
- Rapid Nutritional Assessment (Field Guide)
- Borne Superieure

Vous êtes sur la page 1sur 16

https://doi.org/10.1007/s11205-017-1582-8

Ó Springer Science+Business Media Dordrecht 2017

Abstract During the last two decades, interest towards sampling survey on gender roles

perceived by the population has significantly grown in developed as well as in developing

countries. The micro-perspective of research improves the knowledge of gender discrim-

ination dynamics even among subgroups. At the same time, it stands as a methodological

challenge, as this approach needs an adequate statistical method for the analysis of indi-

vidual data. This contribute aims firstly at offering the opportunity of increasing the

knowledge about gender disparities through individual opinions and perceptions. Secondly,

it aims at enlightening the pertinence of the poset methodology for the analysis of ordinal

variables and response profiles. To this purpose, we collected data about 16 African

countries included in Health Demographic Survey and we analysed a battery of questions

about decision-making dimension by means of a poset methodology. We summarized the

results of the poset analysis by means of descriptive indicators and we investigated their

relation with the Global Gender Gap index (GGGI), an official index released every year

by the World Economic Forum.

Ordering Partial order theory

1 Introduction

Gender statistics and gender gap indicators, commonly used to implement international

policies against discrimination of women, found their origin around three decades ago,

receiving the greatest support at the fifth World Conference on Women in Beijing (1995).

agnese.dibrisco@unimib.it

1

Department of Economics, Management and Statistics, University of Milano-Bicocca, Milan, Italy

2

Department of Sociology and Social Science, University of Milano-Bicocca, Milan, Italy

123

1110 A. M. Di Brisco, P. Farina

At that arena, discussion about women condition was replaced by the focus on gender that

implies a comparative approach between women and men (UN 1996). From then on, the

development and affirmation of gender statistics, on the basis of an internationally agreed

agenda, became a reality. The systematic investigation of whatever characteristic with the

aim of evaluating gaps and issues based on gender has been reaffirmed by the 3th goal of

the ‘‘Millennium Development Goals’’, as well as the 5th of the most recent ‘‘Sustainable

Development Goals’’ (UNDP 2015b).

Gender statistics released by International Agencies are simple indicators such as dif-

ferences, ratios, shares, proportions and rates as well as composite indicators (Joint

Research Centre-European Commission 2008), namely mono-dimensional quantities that

synthesize multiple variables purposively selected to evaluate national results and efforts

against discrimination. The latter group of indicators plays an important role in gender

studies and several world rankings based on gender composite indicators are annually

released by International Agencies, even if a controversy persists on the subjectivity

naturally connected with the construction of any composite indicator (Mecatti et al. 2012).

According to gender gap indicators, disparities between men and women decreased

almost everywhere in the last two decades (Schwab and Sala-i-Martin 2015; Selim 2015),

even if less than predicted, accordingly to the 3th goal of Millennium Development Goals

(UN 2015). The existing gaps about structural conditions of societies—education, work

and power—are the results of ideological as well as cultural traits shared by women and

men regarding their position within society. Women are now more educated than ever

before. By way of example, in many countries, the primary, secondary and even tertiary

education gaps have closed (UN 2015, pp. 28–30). Instead, the improvements in work

opportunities are not as relevant. Women still do most of the unpaid work, including

childcare and housework. They also work for pay, but most in low-skilled labors, and are

paid less for the same work (UNDP 2016). Finally, women are the minority in leadership

roles in business and especially in politics (UN 2015, pp. 28–30).

Besides the cultural traits that originate structural gaps, there exist individual traits, not

captured by gender gap indicators, that can ease (or worsen) the individual gap experienced

by each woman or man.

For this reason, in order to improve the understanding of the processes underlying the

still existing gaps, rather than only monitoring them at a macro-level, by mean of com-

posite indicators, it is of keen interest the adoption of a micro approach.

The growing interest in a micro perspective of research in the field of discrimination

among women and men goes together with the increase, during the last two decades, of

sampling survey focused on gender roles in developed as well as in developing countries

(UN 2015; Women 2014; Barker et al. 2011).

The shift from a macro perspective to individual-based approach allows for measuring

the widespread of inequalities between women and men in terms of their perception of

discrimination and their agreement on different social roles.

Shifting from the analysis of macro-dimensions of gender gaps to the analysis of

individual perception of discrimination and decision-making processes stands also as a

methodological challenge. Indeed, when dealing with the analysis of structural gaps at

macro-dimensions usually quantitative variables are used; consider, by way of example,

the proportion of illiterates about education, the proportion of employed women about

work or the number of women in parliament about power. As a counterpart, when focusing

on individual opinions and decision making processes usually qualitative ordinal variables

are available; consider, by way of example, the degree of agreement about an opinion or

the specification of who is the decision maker about a process.

123

Measuring Gender Gap from a Poset Perspective 1111

dynamics, but it needs an adequate statistical method for the analysis of ordinal data at

individual level.

This contribute aims at evaluating the gender gap at individual level, by taking

advantage from the poset approach, a statistical method for the analysis of ordinal variables

and response profiles. The poset approach, that has been at first applied to gender sensitive

data by Di Brisco, Bertarelli and Mecatti (VI European Congress of Methodology, Utrecht

2014), has the merit of permitting an individual-oriented analysis, respecting the ordinal

nature of data.

In the following section we illustrate the dataset (2.1), we briefly recall the poset

methodology (2.2) and we describe two indices, that synthesize the poset structure, about

respectively the degree of membership of women to the set of disadvantaged women and

the severity of the gender gap. In Sect. 3 we present some results about our application of

poset methodology on a battery of questions concerning the decision-making dimension

regarding the female population of 16 African countries. At last, in Sect. 4 we discuss these

results by country and we compare them with official Gender statistics released by

International Agencies.

household surveys that provide data for a wide range of monitoring and impact evaluation

indicators in the areas of population, health, and nutrition carried out between 2007 and

2012. They have large sample sizes (usually between 5000 and 30,000 households) and

typically, they are conducted about every 5 years. The DHS Program data and analysis

provides an in-depth look at the life courses of women and men through a standard

questionnaire. According to the needs of countries, the programme can include also a

gender empowerment module asking if women and men have equal access to information

and resources or they can make decisions for themselves and their children or their

agreement with gender norms that disempower women.

Within the DHS programme, the integrated Demographic and Health Series (IDHS) has

been developed to facilitate analysis of DHS data across countries and over time. The

IDHS includes more than 70 surveys conducted from 1988 to 2015 in 16 African countries.

The surveys interested only women respondents about many topics such as demographic

and socio-economic status, family planning, domestic violence and decision-making.

For the aim of this study on gender disparities at an individual level, we extracted the

most recent survey for each African country from http://idhsdata.org (Table 1).

The model questionnaire is similar across countries and variables are coded

consistently.

Gender disparities may arise in a variety of dimensions that often depend on the cultural

environment, economic availability or political empowerment. Conversely, the decision-

making dimension, which still is influenced by all these macro factors, also depends on an

individual’s propensity to gender empowerment. Within the main available topics, we

focused on decision-making.

123

1112 A. M. Di Brisco, P. Farina

Country Year Cases Universe

country we reported the year in

which the more recent survey has

been conducted (Year) Benin 2011 16.599 Women age 15–49

Burkina Faso 2010 17.087 Women age 15–49

Cote d’Ivoire 2011 10.060 Women age 15–49

Egypt 2008 16.527 Ever married women age 15–49

Ethiopia 2011 16.515 Women age 15–49

Ghana 2008 4.916 Women age 15–49

Guinea 2012 9.142 Women age 15–49

Kenya 2009 8.444 Women age 15–49

Malawi 2010 23.020 Women age 15–49

Mali 2012 10.424 Women age 15–49

Mozambique 2011 13.745 Women age 15–49

Nigeria 2013 38.948 Women age 15–49

Tanzania 2010 10.139 Women age 15–49

The total amount of respondent Uganda 2011 8.674 Women age 15–49

women (Cases) and the Zambia 2007 7.146 Women age 15–49

population responding to the Zimbabwe 2011 9.171 Women age 15–49

questionnaire (Universe)

ables. We dropped out four variables, about who has the final say on food to be cooked, on

spending husband’s earnings, on spending woman’s earnings and on household purchases

for daily needs, because of the great amount of missing data across countries. We kept

instead three variables that ask ‘‘Who in your family usually has the final say on the

following decisions:’’ (1) ‘‘own health care’’, (2) ‘‘visits to family’’ and (3) ‘‘large

household purchases’’.

Each variable concerns a specific dimension of gender disparity: more precisely, the

first variable quantifies the level of emancipation of women with respect to their own

health care; the second variable evaluates the level of women’s social isolation while the

third variable deals with the possibility for women to administrate money.

Each variable provides for the same response structure: indeed, a woman respondent,

asked to answer who has the final say on a decision-making, can answers one of the

following: ‘‘husband/partner’’ (1), ‘‘respondent and husband/partner jointly’’ (2) or ‘‘re-

spondent’’ (3).

The possible answers induce a natural ordering with respect to the dimension of

emancipation and freedom on women. The response item (1) suggests a low level of

emancipation and corresponds to a condition of disparity against women; the response item

(2) corresponds to a balanced condition while the response item (3) suggests a high level of

emancipation and corresponds to a condition of disparity in favour of women. Hence, the

dataset is characterized by three variables that are orderable with respect to the dimension

of gender empowerment and freedom.

In Table 2 we reported the entire aggregated dataset consisting of a matrix of 27 rows

(one for each possible combination of responses to the three questions) and 16 columns

(one for each African country).

123

Table 2 Percentage frequencies observed in the sample by profiles (27 rows) and by countries (16 columns)

Benin Burkina Cote Egypt Ethiopia Ghana Guinea Kenya Malawi Mali Mozambique Nigeria Tanzania Uganda Zambia Zimbabwe

Faso d’Ivoire

111 22.606 42.341 43.340 8.518 12.262 9.720 40.090 12.200 22.018 68.309 11.801 46.031 31.405 17.885 14.404 3.755

211 0.703 1.322 3.302 0.180 1.695 0.865 4.438 1.823 0.740 1.305 0.700 1.131 1.364 2.924 2.074 0.824

311 0.593 0.321 1.699 0.159 0.459 1.072 3.898 0.801 0.535 1.682 0.689 1.135 0.345 1.405 1.509 0.879

121 6.807 8.019 6.283 2.867 3.958 8.820 6.372 4.888 10.118 5.162 7.588 6.889 4.108 3.342 3.937 1.172

221 4.360 1.338 3.254 0.533 2.941 3.217 6.162 2.704 2.516 2.752 2.675 2.230 1.808 3.038 2.994 3.352

321 0.584 0.069 0.529 0.152 0.229 1.591 0.990 0.601 0.528 0.376 0.930 0.196 0.099 0.475 1.037 1.008

131 1.228 20.264 6.636 0.506 3.230 4.531 2.009 3.986 7.390 3.046 1.642 1.620 1.019 3.418 4.102 0.824

231 0.152 1.292 1.282 0.111 0.698 1.453 0.825 0.761 0.555 0.129 0.298 0.414 0.230 1.101 1.462 0.788

331 0.440 0.451 1.427 0.125 0.389 1.141 1.274 0.601 1.057 1.058 0.838 0.422 0.148 1.576 1.155 1.062

112 2.887 0.917 1.378 2.860 2.413 1.176 0.825 2.103 5.382 0.811 1.263 1.013 9.318 2.791 1.792 0.751

Measuring Gender Gap from a Poset Perspective

212 1.744 2.568 1.779 0.658 4.466 1.107 2.699 3.626 0.997 1.211 1.814 0.921 1.791 4.196 2.405 1.319

312 0.119 0.084 0.144 0.076 0.239 0.484 1.934 0.521 0.244 0.106 0.379 0.104 0.197 0.532 0.613 0.605

122 3.573 2.652 2.292 13.899 5.902 6.918 1.709 2.284 11.300 0.576 4.064 3.265 7.214 2.468 3.230 1.612

222 35.204 4.992 13.896 40.422 33.785 25.631 16.042 28.546 12.621 4.939 34.680 25.402 20.822 23.106 19.896 49.588

322 0.694 0.153 0.465 0.769 1.127 3.251 1.724 1.402 0.720 0.647 2.640 0.581 0.740 0.892 1.462 3.059

132 0.508 1.536 0.561 0.699 1.446 1.314 0.225 1.022 4.306 0.153 0.390 0.303 0.740 1.291 1.226 0.678

232 1.092 3.379 1.427 1.420 6.919 2.421 0.975 5.148 1.294 0.223 2.606 1.372 0.641 2.791 4.762 2.711

332 0.279 0.161 0.497 0.173 0.738 1.453 0.435 0.581 0.786 0.106 1.079 0.447 0.378 1.139 0.872 1.374

113 1.998 0.925 1.058 5.928 1.426 1.522 0.525 2.845 2.311 1.705 5.097 0.422 5.144 3.361 5.587 0.898

213 0.161 0.245 0.272 0.506 0.718 0.346 0.165 0.982 0.192 0.118 0.448 0.111 0.509 1.709 2.027 0.531

313 0.754 0.329 0.769 0.159 0.389 0.934 1.184 1.142 0.462 0.753 0.804 0.329 0.493 1.576 1.155 0.843

123 2.379 0.375 0.465 7.389 1.316 3.251 0.180 2.404 2.846 0.423 1.355 0.370 2.383 1.614 4.173 0.879

223 3.116 0.275 1.154 4.744 3.409 3.632 0.600 6.350 1.301 0.376 3.249 0.902 2.251 2.601 6.742 7.291

323 0.982 0.092 0.705 1.004 0.748 3.182 0.690 1.522 0.634 0.317 2.491 0.606 0.493 1.709 1.603 2.290

133 0.872 3.081 1.314 2.299 2.173 2.490 0.255 2.784 4.385 0.623 1.688 0.632 2.087 2.411 3.678 1.319

233 0.525 1.063 0.833 1.530 3.529 1.937 0.270 2.244 1.037 0.118 2.904 0.592 0.657 1.956 2.216 2.876

1113

123

Table 2 continued

1114

Benin Burkina Cote Egypt Ethiopia Ghana Guinea Kenya Malawi Mali Mozambique Nigeria Tanzania Uganda Zambia Zimbabwe

Faso d’Ivoire

123

333 5.639 1.758 3.238 2.313 3.399 6.538 3.508 6.130 3.725 2.975 5.889 2.562 3.615 8.696 3.890 7.712

We recall that a woman respondent, asked to answer who has the final say on a decision-making, can answers one of the following: ‘‘husband/partner’’ (1), ‘‘respondent

and husband/partner jointly’’ (2) or ‘‘respondent’’ (3)

A. M. Di Brisco, P. Farina

Measuring Gender Gap from a Poset Perspective 1115

Dealing with ordinal variables achieving the goal of ordering profiles (as illustrated in this

Section) and quantifying the severity of gender disparities (as illustrated in Sect. 2.3)

involves a methodological challenge.

When dealing with multidimensional ordinal data, the standard statistical methods that

provide for score aggregation, namely methods that summarize quantitative observations

into aggregate statistics, or scaling, namely methods that create a continuum upon which

measured objects are located, are not adequate (Fattore et al. 2011). In addition, standard

statistical techniques measure gender equality or gender gap only at an aggregated level,

i.e. at a country level.

For the purpose of our study on gender empowerment at an individual level, we take

advantage of the poset approach, a statistical method for the analysis of ordered sets.

In what follows, we give some basic definitions and elements of the partial order theory;

those interested can find an exhaustive dissertation in Neggers and Kim (1998).

Let us first define a finite partially ordered set (poset), synthetically indicated as

PðX; Þ, as a finite set X with a partial order relation that is a binary relation satisfying

the following properties:

1. Reflexivity: x x; 8x 2 X ;

2. Anti-symmetry: if x y and y x then x ¼ y for x; y 2 X;

3. Transitivity: if x y and y z then x z for x; y; z 2 X.

We define that two elements of the set X are comparable if x y or y x, otherwise they

are not comparable. When any two elements of X are comparable, then the poset P is said a

chain or a linear order; on the other hand, if any two elements of X are not comparable then

the poset P is said an anti-chain.

Given a poset, it is of interest to identify some subsets enjoying special properties. In

this regard, an up-set U of a poset is a subset of P such that if x 2 U and x z then z 2 U;

in a similar way a down-set D of a poset is defined as a subset of P such that if x 2 D and

y x then y 2 D. An extension of a poset P is also a poset defined on the same finite set X,

whose set of order relations comprises that of P. An extension is said to be linear if it is a

linear order or a chain.

To the purpose of this study, we consider a battery of K variables with ordinal nature,

each one with its own set of jk possible response items or levels, with k ¼ 1; . . .; K.

T

We define a response profile as a vector p ¼ ðp1 ; . . .; pK Þ such that each pk element is

equal to level c 2 f1; 2; . . .; jk g of the k th variable. The number of all possible profiles

QK

results equal to: #p ¼ jk .

k¼1

Generally speaking, given a set of objects (profiles in our study), we may be interested

in ordering them, recalling that order is not a characteristic of each single object but

concerns comparisons between objects (Davey and Priestley 2002).

To evaluate if two profiles can be ordered and what is the order between them, we can

T

formally define the following rule. Given two generic profiles, say pm ¼ p1m ; . . .; pKm and

T

pn ¼ p1n ; . . .; pKn , over K ordinal variables, we assess that pm dominates pn , and we write

pm pn , if and only if pim pin 8i ¼ 1; . . .; K. The set of response profiles with the ordering

rule generate a finite partially ordered set (poset).

123

1116 A. M. Di Brisco, P. Farina

In the light of this ordering rule, giving at least two ordinal variables, the issue of partial

ordering necessarily arises, meaning that, by comparing any possible couple of profiles

some couple of profiles cannot be ordered. To better clarify this issue, let us consider two

dichotomous variables and let us list the entire set of profiles: ({1,1}, {1,2}, {2,1}, {2,2}).

By way of example, it is evident that f1; 1g f2; 2g while does not exist a ‘‘natural order’’

among {1,2} and f2; 1g.

Starting from this partial ordering among profiles, the purpose of our analysis becomes

the identification of disadvantaged profiles, with respect to some reference profiles, in each

linear extension of the poset (Fattore 2015).

In this regard, the mathematical theory of partial order provides a useful formalization

of such a methodological problem. Neggers and Kim (1998) demonstrated a preposition

that relates the definition of an up-set to the existence of an anti-chain that generates that

up-set. More formally, they show that given a finite poset P and an up-set U, then there

exists an anti-chain u such that u P. Consequently, z 2 U if and only if 9u 2 u such that

u z and the up-set U is generated by an anti-chain. In an analogous way, the theorem can

be stated also for down-sets. An anti-chain, because of its property of generating down-sets

and up-sets, provides a partial clustering of profiles and therefore plays the role of

threshold in classifying profiles with respect to the latent dimension of interest.

A Theorem, demonstrated by Neggers and Kim (1998), states that the set of linear

extensions of a finite poset P uniquely identifies P: as a result, a poset is uniquely identified

by the set of all its linear extensions. Since in all linear extensions the profiles are com-

pletely ordered, a profile that does not belong to the down-set generated by a chosen

threshold can take a position above the threshold in some linear extensions and below it in

others.

Fattore (2015) defines the evaluation (or identification) function,g : T ! ½0; 1 where T

is the finite collection of all possible profiles, as a fuzzy membership function from the

finite space of all possible profiles to the [0,1] interval that counts how many times a profile

belongs to the down-set D, in the set of all linear extensions:

8

> 1 if p 2 D

< #fl 2 EðPÞ : 9d 2 d : d p 2 lg

gð pÞ ¼ otherwise ð1Þ

>

: jEðPÞj

0 if p 2 U

where EðPÞ is the set of all possible linear extensions of the poset, while d is the anti-chain

selected as a threshold. The evaluation function assumes value 1 for profiles certainly

under the threshold, namely belonging to the down-set generated by the threshold and

value 0 for profiles certainly upper the threshold, namely belonging to the up-set generated

by the threshold. For profiles that are not certainly ordered with respect to the chosen

threshold, the evaluation function can assume values in the real interval (0-1).

Since the selected threshold identifies the set of disadvantaged profiles, with respect to

the dimension of women empowerment, the value assumed by the evaluation function does

not quantify the intensity of the gender disadvantage but the degree of membership of a

profile to the set of disadvantaged profiles, accordingly to the interpretation provided by

Fattore et al. (2012). Indeed, the evaluation function results equal to the proportion

between favorable outcomes (the number of times a profile is below the threshold) and the

number of possible cases (the total number of linear extensions).

The partial order approach can be advantageous in this field of analysis and when

dealing with ordinal variables collected at an individual level. Firstly, it respects the

ordinal nature of the data (Fattore et al. 2011b): no transformation procedure on the data,

123

Measuring Gender Gap from a Poset Perspective 1117

such as aggregation or scaling, is required and the ordinal variables are not treated

improperly as continuous variables. Secondly, it is truly multidimensional, meaning that it

provides for a conjoint analysis on the set of variables related, in our study, to the latent

dimension of gender empowerment.

Starting from a set of qualitative ordinal variables, by mean of the poset approach we get to

a continuous function that assumes values in the interval [0,1] and whose distribution can

be synthesize by standard descriptive indicators.

As a first synthetic indicator, one among many, we can compute the mean of the

evaluation function: such a measure quantifies the average degree of membership of

profiles belonging to the disadvantaged group, characterized by gender gap against women.

This synthetic indicator can be interpreted as the theoretic average degree of membership,

depending only on the structure of the poset and on the selected threshold. Nevertheless,

we are interested in quantifying a similar average with respect to the observed dataset,

across countries. In this regard, by simply weighting the evaluation function with respect to

the relative frequency of the profiles in the sample, we get a second synthetic indicator:

X

l¼ gð pÞwp ð2Þ

p

where wp is the weight assigned to profile p and usually corresponds to the relative

frequency of the profiles in the sample. This second indicator quantifies the observed

degree of membership of a woman (and not a profile as before) to the disadvantaged group.

Given the evaluation function, that depends on the selected threshold, l takes as

minimum and maximum values (lower and upper bounds respectively) the following

X

lLB ¼ wp Ifgð pÞ¼1g

p

X

lUB ¼ 1 wp Ifgð pÞ¼0g ð3Þ

p

We observe that lLB is equal to 0 if wp ¼ 0, for all profiles in the down-set, while lUB is

equal to 1 if wp ¼ 0 for all profiles in the up-set. Given its extremes, the index l can be

normalized by getting

l lLB

lN ¼ ð4Þ

lUB lLB

We observe that lLB and lUB are rough measures that only depend on profiles belonging

to the down-set and to the up-set of the poset, while l, and hence lN , synthesises the entire

evaluation function distribution.

Finally, we define a third synthetic indicator, the gender gap indicator G, that quantifies

the severity of the gap, for the subgroup of disadvantaged women. Such a gender gap

measure is formally defined as follows (Fattore et al. 2011b),

X

#p

G¼ wp distð pÞ ð5Þ

p¼1

123

1118 A. M. Di Brisco, P. Farina

M . We underlie that wp is a weight assigned to profile p and it is

equal to the relative frequency of subjects sharing the same profile; p is the first profile

greater than the threshold and, at last, M is the absolute distance of the first profile greater

T

than the threshold from the minimal element, that is defined as pmin ¼ p1min ; . . .; pKmin

such that pimin pin 8i ¼ 1; . . .; K; 8n 2 T. We observe that the absolute distance distð pÞ is

set equal to 0 when a profile is greater than the threshold. By construction, the gender gap

index is a weighted mean of normalized distances and hence it assumes values on (0,1): the

greater the measure of this index, the greater the gender gap.

The gender gap measure G quantifies the severity of the discrimination against gender

suffered by disadvantaged women by evaluating the distance between each profile and the

first not disadvantaged profile, namely the first profile greater than the threshold in the

linear extension. More specifically, we can interpret G as a measure of the the height of the

step a woman has to overtake to exit her condition of disadvantage.

3 Results

The collection of three variables, each one with three possible item responses, determines a

total amount of 3 3 3 ¼ 27 possible response profiles. Because of the ordinal nature of

the variables in exam, a partial order among profiles is immediately available.

The Hasse diagram, illustrated in Fig. 1, provides a useful graphical visualization of the

entire set of profiles and of its partial ordering.

To better interpret the information collected in the Hasse diagram, we recall that such a

diagram is a special kind of graph in which a node corresponds to a response profile and

profiles linked by an edge are comparable, namely, there exists a natural order between

them, otherwise, the profiles are not comparable. The two profiles at the extremes of the

diagram have an immediate and clear interpretation: profile ‘‘111’’ is associated with the

most severe condition of gender gap against women while profile ‘‘333’’ is associated with

the maximum condition of empowerment and freedom in favour of women.

Fig. 1 Hasse diagram of the 27 possible response profiles: each profile corresponds to a node of the graph.

If pm pn then the node corresponding to profile pm is placed above the node corresponding to profile pn .

Furthermore, if pm pn and 9= pl : pm pl pn then an edge is inserted between pm and pn

123

Measuring Gender Gap from a Poset Perspective 1119

Conversely, profile ‘‘222’’ seems to represent a point of balance among the set of

profiles and it takes a ‘‘central’’ position in the Hasse diagram. The poset methodology is

completely data-driven and does not provide for any distributional assumption on the data,

still it requires the choice of a threshold that, from a mathematical point of view, has to be

an anti-chain in order to generate the down-set of disadvantaged profiles. For the purpose

of our study on gender disparities, hence, we have to determine the set of certainly

disadvantaged profiles by fixing an a priori threshold. In this regard, if we think in terms of

down-set rather than in term of the threshold, it seems reasonable to classify a woman as

certainly disadvantaged if she answers ‘‘1’’ to at least one question but never answer ‘‘3’’.

Indeed, by selecting profile ‘‘222’’ as the threshold profile, the down-set we get contains

exactly the profiles we intuitively classify as deprived profiles.

The same down-set would have been obtained by selecting as a threshold the anti-chain

made up by profiles (‘‘122’’, ‘‘212’’, ‘‘221’’); nevertheless, profile ‘‘222’’ would no longer

be in the down-set. The latter profile corresponds to women that are allowed to participate

in decision-making but are not free to decide on their own in all the three dimensions

considered.

We decided to provide results regarding two different scenarios, namely when the

threshold is set equal to ‘‘222’’ (scenario I) and when the threshold is equal to the anti-

chain (‘‘122’’, ‘‘212’’, ‘‘221’’) (scenario II).

In both scenarios, all profiles containing, at least, one answer ‘‘3’’ are excluded from the

down-set: indeed, it seems reasonable not to classify a woman as certainly disadvantaged if

she has expressed her independence in decision-making for at least one dimension.

In a study on gender disparities, it is of sure interest to identify the proportion of

disadvantaged women and to evaluate the gap that separates the group of disadvantaged

women from the group of empowered women. In this regard, the second and the third

synthetic indicators, namely the observed degree of membership lN and the gender gap

index G, illustrated in the previous Section, address this requirement.

Once selected the proper anti-chain as a threshold (‘‘222’’ for scenario I and (‘‘122’’,

‘‘212’’, ‘‘221’’) for scenario II) we analyzed separately each country and we constructed the

evaluation functions, using the R package PARSEC (Arcagni and Fattore 2014).

In the following Figure, we reported the evaluation functions in the two scenarios:

The evaluation function gð pÞ takes values on ð0; 1Þ: the profiles are ordered in

ascending order with respect to the value of the evaluation function and we observe four

steps in the graph. Hence, it is evident that in each scenario, given the chosen threshold and

given the starting poset, the profiles can be classified, in the first place, into five different

groups, from a level of complete disadvantage against woman, to a level of complete

empowerment in favor of women. The mean value of the evaluation function is equal to

0.519 in scenario I and so slightly over half of the profiles are a posteriori classified as

disadvantaged profiles against women. Conversely in scenario II, the mean value of the

evaluation function is lower, equal to 0.371.

The absolute average values of the evaluation function depend only upon the structure

of the poset and the chosen threshold, as a consequence both the graphical representation

of the evaluation function and its mean value are the same across countries.

To evaluate the average degree of membership of women to the disadvantaged set of

profiles we have to weigh the evaluation function with respect to the observed relative

frequencies and construct the synthetic measure lN for each country.

We emphasize, at this stage, the added value that the poset methodology offers with

respect to other aggregative approaches. In this regard, standard methods usually allow

only classifying profiles dichotomously as disadvantaged or not disadvantaged while the

123

1120 A. M. Di Brisco, P. Farina

of women to the set of disadvantaged profiles.

Focusing the attention on the subgroup of disadvantaged women, as a next step, we

quantify the gap that distances disadvantaged woman from empowered women. In this

regard, we computed the gender gap index G for each country and we ranked them in

ascending order. The results about the gender indicators are reported in Tables 3 and 4:

The empirical results observed in both scenarios suggest a polarized situation: in

countries where the proportion of disadvantaged women is high, also the disadvantaged

group is well distanced from the group of empowered women; on the contrary in countries

where the proportion of disadvantaged women is low, the disadvantaged group is not so far

from the group of empowered women. By way of example, we may observe that in

Zimbabwe, in both scenarios, the normalized mean degree of membership and the gender

gap index assume the lowest values, with respect to the other countries. On the contrary in

both scenarios, Mali is the country with the highest level of G and the second highest level

of lN .

By simply analysing the linear relation between the two synthetic indicators constructed

from the poset, we observe that, in both scenarios, as the degree of membership lN

increases also the gender gap index G increases, supporting the evidence of polarization.

At last, we compare the results obtained with the poset analysis with some official

statistics indicators. Usually, official indicators on gender gap synthesize aggregate

statistics and offer a photograph of the cultural environment, of the economic situation and

of the political empowerment of countries. Conversely, the analysis we have performed

emphasizes the gender gap, at an individual level, within decision-making.

Table 3 For each country we reported the degree of membership of women to the disadvantaged group of

profiles and the gender gap index

Country l lLB lUB lN (ranking) Gender gap index (ranking)

Burkina Faso 0.862 0.641 0.931 0.759 (16) 0.709 (15)

Cote d’Ivoire 0.861 0.755 0.917 0.653 (14) 0.681 (14)

Egypt 0.807 0.699 0.880 0.594 (9) 0.325 (2)

Ethiopia 0.747 0.674 0.801 0.572 (5) 0.392 (4)

Ghana 0.686 0.575 0.776 0.553 (2) 0.402 (5)

Guinea 0.866 0.783 0.918 0.617 (10) 0.656 (13)

Kenya 0.687 0.582 0.766 0.570 (4) 0.402 (6)

Malawi 0.799 0.657 0.905 0.574 (6) 0.557 (10)

Mali 0.918 0.851 0.952 0.666 (15) 0.845 (16)

Mozambique 0.736 0.646 0.791 0.618 (11) 0.391 (3)

Nigeria 0.907 0.869 0.929 0.625 (13) 0.651 (12)

Tanzania 0.861 0.778 0.912 0.621 (12) 0.599 (11)

Uganda 0.711 0.597 0.802 0.554 (3) 0.464 (8)

Zambia 0.668 0.507 0.785 0.579 (8) 0.433 (7)

Zimbabwe 0.673 0.624 0.727 0.474 (1) 0.204 (1)

Given the threshold ‘‘222’’ (scenario I). We recall that l is the mean of the evaluation function and lLB and

lUB are its lower and upper bounds, respectively

123

Measuring Gender Gap from a Poset Perspective 1121

Table 4 For each country we reported the probability that women belong to the disadvantaged group of

profiles and the gender gap index given the threshold (‘‘221’’.’’212’’.’’122’’) (scenario II)

Country l lLB lUB lN (ranking) Gender gap index (ranking)

Burkina Faso 0.754 0.592 0.881 0.562 (16) 0.680 (11)

Cote d’Ivoire 0.690 0.616 0.778 0.454 (14) 0.738 (14)

Egypt 0.355 0.295 0.476 0.332 (3) 0.429 (2)

Ethiopia 0.380 0.336 0.463 0.345 (6) 0.531 (6)

Ghana 0.382 0.318 0.520 0.316 (2) 0.469 (4)

Guinea 0.676 0.623 0.758 0.394 (10) 0.723 (13)

Kenya 0.361 0.296 0.481 0.348 (8) 0.510 (5)

Malawi 0.618 0.531 0.779 0.350 (9) 0.552 (9)

Mali 0.850 0.801 0.903 0.481 (15) 0.859 (16)

Mozambique 0.359 0.299 0.445 0.409 (12) 0.552 (8)

Nigeria 0.640 0.615 0.675 0.419 (13) 0.831 (15)

Tanzania 0.624 0.570 0.704 0.399 (11) 0.690 (12)

Uganda 0.436 0.366 0.571 0.339 (4) 0.543 (7)

Zambia 0.405 0.308 0.586 0.348 (7) 0.468 (3)

Zimbabwe 0.153 0.128 0.231 0.245 (1) 0.386 (1)

empowerment and on individual propensity but is also influenced by the social, economic

and political background.

The official index we selected is the Global Gender Gap index (GGGI) which can be

decomposed into four sub-indexes referred to four latent dimensions: the economic par-

ticipation (ECON), the health condition and the survival probability (HEALTH), the

political empowerment (POLITIC) and the educational attainment (EDUC).

For each scenario, we define two separate linear models for lN and G, adopting the four

sub-indexes of GGGI as covariates. Since both the dependent variables lN and G lie on the

restricted interval (0,1), we implemented a Beta regression model (Ferrari and Cribari-Neto

2004). We specify the following model for the mean of the dependent variable

gðmi Þ ¼ b0 þ b1 EDUCi þ b2 HEALTHi þ b3 POLITICi þ b4 ECONi ð6Þ

where mi is the mean of the dependent variable (lN or G respectively) and the link function

gðÞ is the logit function which is strictly monotone and twice differentiable. In Table 5 we

reported the estimates of the regression parameters (b0 ; b1 ; b2 ; b3 and b4 Þ):

Due to lack of updated data, Benin was removed from the GGGI report and so we

excluded it from this analysis. The GGGI sub-index about health and survival has a

significant influence on both the synthetic indicators lN and G. In particular, since the

regression parameter b1 has a negative sign, the highest the condition of health of a

country, the lowest the degree of membership lN and the gender gap index G. The

education is one of the fundamental means of empowerment of women’s freedom and

reduction of the gender gap, and indeed the parameter b2 related to the GGGI sub-index

about educational attainment has a negative sign in all models. Also the political

empowerment has a negative linear relation with respect to lN and G but the regression b3

parameter is not statistically significant in any scenario. The parameter b4 related to

123

1122 A. M. Di Brisco, P. Farina

Table 5 Beta regression models for the normalized mean lN and for the gender gap index, G for each

scenario

b0 b1 (EDUC) b2 (HEALTH) b3 (POLITIC) b4 (ECON)

Dependent variable: l

Scenario I

Estimate 18.704 20.792 218.285 20.511 0.265

SD 11.423 0.756 12.07 0.742 0.726

Scenario II

Estimate 25.139 21.135 225.729 20.422 0.494

SD 11.903 0.801 12.582 0.801 0.791

Dependent variable: G

Scenario I

Estimate 81.666 22.377 282.414 21.248 0.695

SD 21.69 1.325 22.793 1.284 1.294

Scenario II

Estimate 55.361 23.071 254.251 20.3498 0.383

SD 20.912 1.327 22.029 1.26 1.247

In bold the coefficients significant at p \ 0.05, in italics at p \ 0.1

Fig. 2 Plot of the evaluation function for the poset structure represented in Fig. 2 given the threshold

profile ‘‘222’’ (scenario I, left side) and given the threshold (‘‘122’’, ‘‘212’’, ‘‘221’’) (scenario II, right side).

The profiles are listed in ascending order

economic participation, which has a positive sign, lacks of statistical significance in all

models; indeed, a simple scatterplot shows that no relation exists between the ECON

covariate and the dependent variables lN and G.

4 Conclusions

national results and efforts against discrimination. The existing gaps about structural

conditions of societies—equal access to education, work, and power above all—do not

depend only on cultural traits at macro level, but are the results also of individual traits

shared by women and men regarding their position and their social roles within society. For

123

Measuring Gender Gap from a Poset Perspective 1123

this reason, the adoption of a micro approach can improve the understanding of the

processes beneath the still existing gaps. Our contribute addresses the decision-making

process, in the form of the final say ‘‘on own health care’’ as representative of Human Right

violation; ‘‘visit family’’ as representative of the freedom, and ‘‘large household pur-

chases’’ as representative of the economic power (Kishor and Subaiya 2008). The

implementation of poset procedure to these dimensions made it possible to rank the

position of each country and compare it to the one resulting from use of macro indicators.

As expected, countries belonging to group with the highest level of membership lN and

the highest level of Gender Gap Index as Burkina Faso, Mali, Ivory Coast, and Nigeria

share also the lowest position in the Gender inequality index (UNDP 2015a). It is due

mainly to the high proportion of early motherhood (fertility rates under 20 years old), and

the lowest rate of leadership (proportion of seats in parliament). Both indicators are

explanatory of social norms, and traditions that limit women options of life, broadening

therefore the gender gap. It is no coincidence that the same countries share also higher

illiterate rate of girls that have a negative effect on empowerment by virtually terminating a

woman’s access to some sources of empowerment (Unicef 2007). On the contrary,

countries at the top of the ranking show better position in the ranking of Gender Equality

Index because of low adolescent birth rate and better education. This is the case of

Zimbabwe and Ghana where women and girls postpone maternity and are more often

literate (UNDP 2015).

It is worth noting however that while the groups at the top and at the bottom of ranking

of membership lN and Gender Gap Index established by poset procedure are congruent

with the level of some macro indicators of gender gap, the profile of countries in the middle

of the list needs to be consider according the two different perspectives concurrently. To

give an example Egypt is near to the top of the poset ranking, but its position is much lower

in the Gender Inequality index. A possible interpretation of this discrepancy lies in the

separation of the public and private sphere as a cultural trait of that country. The decision

making process in the family is balanced between men and women, while their presence in

the public arena is negligible since the proportion of women in parliament and the female

presence on the job market are very low. The comparison between macro and micro

indicators give at least two positive indications on poset methodology. First, the results of

this approach are congruent comparing different sources of analysis like decision making

process and macro indicators of the gender gap. Last, it offers the opportunity of increase

the knowledge through individual opinions and perceptions, a unique source of data.

Compliance with Ethical Standards

Conflict of interest The authors declare that they have no conflict of interest.

References

Arcagni, A., & Fattore, M. (2014). PARSEC: An R package for poset-based evaluation of multidimensional

poverty. In R. Bruggemann, L. Carlsen, & J. Wittmann (Eds.), Multi-indicator systems and modelling

in partial order (pp. 317–330). New York: Springer.

Barker, G., Contreras, J. M., Heilman, B., Singh, A. K., Verma, R. K., & Nascimento, M. (2011). Evolving

men: initial results from the international men and gender equality survey (IMAGES). Washington DC,

Rio de Janeiro: International Center for Research on Women (ICRW), Instituto Promundo.

Davey, B. A., & Priestley, B. H. (2002). Introduction to lattices and order. Cambridge: CUP.

Fattore, M. (2015). Partially ordered sets and the measurement of multidimensional ordinal deprivation.

Social Indicators Research. doi:10.1007/s11205-015-1059-6.

123

1124 A. M. Di Brisco, P. Farina

Fattore, M., Brüggemann, R., & Owsiński, J. (2011a). Using poset theory to compare fuzzy multidimen-

sional material deprivation across regions. In S. Ingrassia, R. Rocci, & M. Vichi (Eds.), New per-

spectives in statistical modeling and data analysis (pp. 49–56). Berlin, Heidelberg: Springer.

Fattore, M., Maggino, F., & Colombo, E. (2012). From composite indicators to partial orders: Evaluating

socio-economic phenomena through ordinal data. In F. Maggino & G. Nuvolati (Eds.), Quality of life

in Italy (pp. 41–68). Dordrecht: Springer.

Fattore, M., Maggino, F., & Greselin, F. (2011b). Socio-economic evaluation with ordinal variables:

Integrating counting and poset approaches. Statistica & Applicazioni, 1, 31–42.

Ferrari, S., & Cribari-Neto, F. (2004). Beta regression for modelling rates and proportions. Journal of

Applied Statistics, 31(7), 799–815.

Joint Research Centre-European Commission. (2008). Handbook on constructing composite indicators:

Methodology and user guide. Paris: OECD Publishing.

Kishor, S., & Subaiya, L. (2008). Understanding women’s empowerment: A comparative analysis of

Demographic and Health Surveys (DHS) data. DHS Comparative Reports No. 20. Calverton, Mary-

land: Macro International Inc.

Mecatti, F., Crippa, F., & Farina, P. (2012). A special gen (d) re of statistics: Roots, development and

methodological prospects of gender statistics. International Statistical Review, 80(3), 452–467.

Neggers, J., & Kim, H. S. (1998). Basic posets. Singapore: World Scientific.

Schwab, K., & Sala-i-Martin, X. (2015). World Economic Forum’s Global Competitiveness Report,

2014–2015. Switzerland: World Economic Forum. Retrived from http://reports.weforum.org/global-

competitiveness-report-2014-2015/. Accessed 25 Jan 2016.

Selim, J. (2015). Human Development Report: Rethinking work for human development. http://hdr.undp.

org/en/rethinking-work-for-human-development. Accessed 16 Jan 2016.

UN (1996). Report of the Fourth World Conference on Women, 4–15 September 1995—Beijing declaration

and platform for action, United Nations, New York. http://www.un.org/womenwatch/daw/beijing/pdf/

Beijing%20full%20report%20E.pdf. Accessed 26 Jan 2016.

UN. (2015). The millennium development goals report 2015. New York: United Nations Publications.

UNDP (2015a). Gender Inequality Index, Human development Index Report http://hdr.undp.org/en/content/

gender-inequality-index-gii. Accessed 25 Jan 2016.

UNDP (2015b). Sustainable development goals Goal 5: Gender equality. http://www.undp.org/content/

undp/en/home/sdgoverview/post-2015-development-agenda.html. Accessed 25 Jan 2016.

UNDP (2016). Imbalances in paid and unpaid work, in Human Development Report 2015 (pp 107–130).

http://hdr.undp.org/sites/default/files/2015_human_development_report.pdf. Accessed 25 Dec 2016.

Unicef (2007). Catalyst for education and gender equality, Issue 1 September 2007. http://www.unicef.org/

education/files/Catalyst_1_Sept07_Web.pdf. Accessed 25 Jan 2016.

Women, U. N. (2014). The World Survey on the role of women in development 2014: Gender equality and

sustainable development. A report signed by the Secretary General and Executive Director, UN

Women. New York, NY: UN. http://www2.unwomen.org/-/media/headquarters/attachments/sections/

library/publications/2014/unwomen_surveyreport_advance_16oct.pdf?v=1&d=20150303T172710.

Accessed 26 Jan 2016.

123

- Joan Scott- GenreTransféré paralexaalex
- Clio 113 14 La Fabrique Du Sexe Thomas Laqueur Et AristoteTransféré parRadu Toma
- Eduquer contre l’homophobie dès l'école primaireTransféré parsavoisien88
- PÊNIS DE TÊTE FILLES.pdfTransféré parAnie Dumont
- Trouble Dans Le Genre (Résumé Premier Chapitre)Transféré parwared76
- TD 2 Sexe et genre élèveTransféré parMme et Mr Lafon
- UntitledTransféré parapi-110755137
- Le Sorbonnard Déchaîné n°39 (fev/mars 2014)Transféré parAxelle Gamsi
- Institution MahefaTransféré parGender Links
- Corrige Conc Gen 00Transféré parMamadou Thioye
- Sondage Superstitions FrancaisTransféré parfablejoueur
- biblio_hommes_femmes (1).pdfTransféré parAline Melo
- o Corpo Feminino No CinemaTransféré parLucas Nunes
- 1042812 ArTransféré parJean Jerry
- echantillonageTransféré partarnima
- Genre Et Inegalites Tema 18Transféré parflorin251
- Mh0213845frc PDF.webTransféré parEuropean Institute for Gender Equality
- Cp Collectif Education Lgbt - 16-09-2013Transféré parDominique
- Rapid Nutritional Assessment (Field Guide)Transféré parAdri De Udok
- Borne SuperieureTransféré parAmina Da
- Genre Cooperation SuremainTransféré parSarah Trichet-Allaire
- visibilité et lisibilité du travail fémininTransféré parsophie_bsk
- Sondage YouGov Pour Le HuffPost CLVTransféré parLeHuffPost01
- EETransféré parAyisse Anliane
- Chapitre v Relations d'Equivalence Et Relations d'OrdresTransféré parBabacar Ngom
- Les+étude..Transféré parJaouad Elouali
- TDTransféré parsimao_sabrosa7794
- Borne SuperieureTransféré parKhalid Mak
- 6Transféré parNadine Reda
- zorn.pdfTransféré parYoussef Er Dosh

- Processus Élémentaires de Dessin Et de Mise en FormeTransféré parbarakupara
- These Rami HAMAMTransféré parNicolas Richard
- Cours PL.pdfTransféré parKalo Mendelssohn
- 49461242 La Compression JPEG d Images FixesTransféré parBacary Sène
- DeterminantsTransféré parKatarina Siročić
- Holomorphe045.pdfTransféré pargazzamnounou
- Sec Minesponts 2006 Sic MPTransféré parBrahim Achayfad
- Rapport Du Projet PisteTransféré parbouhamida
- Performance GlobalTransféré parDana N
- Ecrivez Un Programme en Langage CTransféré parkaidi chaimaa
- programmation-systeme-ghbnvc765Transféré parTrop Freshloïc
- Devoir 1Transféré parsoufianeeslaouy
- Cours Math - Chap 4 Algèbre Arithmétique - 2ème Sciences (2009-2010) Mr Abdelbasset Laataoui Www.espacemaths.comTransféré parSami Ketata
- 0914-cours3-principe2 (1)Transféré parHichem Maalaoui
- qcm-chap6Transféré parClaude Badr
- BertrandTransféré parAbdellatif Elouarrate
- Guide Présentation Memoire IngenieurTransféré parmohamadou
- Cours_334_Serie_4.pdfTransféré parloutrage
- Probabilités et StatistiquesTransféré parHamid Faracha
- SQSTransféré parpichelour2002
- examenNantes2008.pdfTransféré parMed Amine Talhaoui
- Longueur d Une BobineTransféré parNghap
- Cyclic Properties of Sand Dynamic for Seismic ApplicationsTransféré parjimmy Andres
- Lire Un Diagramme de Bode Pour en Extraire Les Informations RecherchéesTransféré parlfadli
- torseurTransféré parhamakouka
- Romain Basson, Corps FinisTransféré parAnonymous va7umdWyh
- exercices corrigées beton_et_acier.docxTransféré parNassimSpencer