**3. Methodology and data source**

conclusions. Fougère [24] highlights the individual characteristics on which employers base their hiring selection strategies, particularly in periods of high unemployment in the case of France. His study is devoted to the influence of the economic situation on the individual durations of unemployment. The individual characteristics enumerated in his study are demographic (sex, age, nationality and family status) and socio-economic (initial training, qualification and previous career path). In the same vein, Bonnal and Fougère [25] reached the same conclusions with the integration of a new socio-economic variable (socioprofessional category of the individual). The work of Joutard and Werquin [26] enriches the understanding by including in the analysis of the individual determinants of duration the distinction between the exit of unemployment on a precarious job and that on a stable job. The results highlight the difference between individual factors and work experience according to the type of job found (stable or precarious). The work of Bourdet and Persson [27], based on a comparative analysis between France and Sweden, emphasizes the need to put in place appropriate policies to absorb youth unemployment fairly quickly. According to this study, the employment policy for young people became permanent in France while it was able to return to Sweden at its level of fifteen years ago (affecting about 4% of the young working population). According to him, this situation is due to the fact that in France the measures were

The growth and persistence of unemployment are worrying because of the change in its physiognomy, which highlights the duality between graduates and non-graduates [28] and the issue of the exit from unemployment with unemployed seniority [29]. The work of Lê et al. [30] leads to an increase in the long-term unemployment rate between 2008 and 2013 among the most fragile asset categories: workers, employees, young people, people without diplomas, single parents, inhabitants of sensitive urban areas and immigrants. In 2013, for example, people without a diploma have a long-term unemployment risk that is twice as high as

On the other hand, other studies have pointed out that the effect of time spent on unemployment on the risk of leaving unemployment is explained by the heterogeneous nature of the cohorts of entries into unemployment. According to Di Paola and Moullet [31], the non-taking into account of the heterogeneity existing between individuals leads to a negative temporal dependence of the risk of exit from unemployment. This phenomenon of negative time dependence of risk also underlined by Heckman and Singer [14] is better known under the name of "movers-stayers." It shows the importance of controlling individual heterogeneity in understanding

Some studies on long-term unemployment have been observed in Africa. The first study is done by Lachaud [32]. According to this author, the mode of access to the labor market seems to be a major determinant of situations of social exclusion, and the influence of this factor goes far beyond what could be called the inherent "logical exclusion" of the destabilization of African economies. Guessan [33] indicates that Ivorian unemployed people in Côte d'Ivoire are less active in job search than non-Ivorian unemployed people and are more competitive. In addition, the study points out that the unemployment rate and the density of the municipality of residence of the unemployed influence the intensity of the job search and the choice of job search methods. Chort et al. [34] conclude for Senegal that apprenticeship is a decisive factor in entry into the labor market, based on the competing risk-entry duration models. By comparing the trajectories of former apprentices with those of non-apprentices, they conclude that the role of entry to apprenticeship is important for subsequent social and occupa-

taken late.

*Regional Development in Africa*

the phenomenon.

tional integration.

**200**

those with a level of 2 years or more.

Our econometric study of the links between the individual characteristics of young people and the duration of unemployment joins the paradigm of the demand for labor [35].<sup>3</sup> The heterogeneity of individual characteristics makes it possible to assess the correlation existing between jobseekers and the difficulty of getting out of unemployment. From a methodological point of view, our analysis is enriched by the integration of a certain number of explanatory variables.

#### **3.1 Econometric model of duration**

The objective being to estimate the probability of a young person finding a job according to the number of months spent unemployed, several econometric models could be useful: for example, a probit or a Tobit. The use of a duration model essentially resides in the fact that it allows to keep in the sample individuals still unemployed at the time of the survey. This type of model is necessary, especially to take into account the censored unemployment episodes. One of the difficulties encountered with the estimation by a duration model is to choose, from all the possible options, a particular description of the probability distribution of the duration variable, having implications on the form of the risk function. A common prediction of empirically observed unemployment durations is that they are characterized by a negative time dependence,<sup>4</sup> leading to the rejection of constant random functions, such as that associated with the exponential model [36]. For this type of analysis, the Weibull model is the most appropriate. This parametric estimation method is widely used [37, 38] because of, on the one hand, the relatively simple form of its survival function and, on the other hand, it belongs both to the family of so-called accelerated life models and so-called proportional hazards models. In addition, the Weibull model can estimate a monotonous chance increasing, monotonous decreasing or constant. A first, primarily descriptive, approach to the durations of our sample using the non-parametric Kaplan-Meier estimator encourages

<sup>2</sup> The distribution of jobs in Côte d'Ivoire is divided into three sectors: the modern sector characterized by modern production technology, a skilled and salaried workforce and compliance with labor regulations. The informal rural sector is identified by microenterprises often confused with the family unit, a rudimentary production technique and the non-respect of labor regulations.

<sup>3</sup> Mériaux [35] calls the "labor demand paradigm" as a guiding principle of an observational program. Rather than focusing on the exchange of facts that occur on the market, priority is given to the different characteristics of each individual that could be an asset or a constraint to the job search.

<sup>4</sup> Allowing to estimate the chances that an individual leaves unemployment, or any other state of the labor market, at a given moment, knowing that he was unemployed until the previous moment (function of risk).

are at least equal to c. If we consider c as a random variable, then c must be independent of T\* after taking into account the other factors explaining the duration of unemployment. The distribution of probabilities of unemployment duration

*A Gender Analysis of the Determinants of Youth Unemployment in Côte d'Ivoire*

*F(t)* then represents the probability that an unemployment duration T lasts less

*f t*ðÞ¼ *dF t*ð Þ

By *f(t)dt*, we then have the probability that the unemployment duration T will end between t and (*t+dt*) periods. These functions (distribution and density) make it possible to determine the survival function S (t) and the function (t) with:

*F t*ð Þ� þ *Δt F t*ð Þ

Generally, the identification of the factors that affect the probability of exit from unemployment to employment is analyzed using the risk functions. We subscribe to this logic. Moreover, Steiner [39] emphasizes that risk functions can be interpreted as reduced forms of the basic job search model as elaborated by McCall [40].

The duration models lend themselves to various types of estimation. The approach chosen for the estimates is the parametric method and, more specifically, proportional hazard models [41]. Suppose t follows a Weibull distribution, noted:

where t is an embodiment of T; λ is the risk function and p is a scale parameter. These parameters (λ and p) can be estimated by the maximum likelihood method. The formulated likelihood function is the sum of the likelihood functions of

*=*dt S tð Þ <sup>¼</sup> �dS tð Þ

*Prob t*ð Þ ≤*T* ≤ *t* þ *Δt*j*T* ≥*t Δt*

*<sup>Δ</sup>tS t*ð Þ <sup>¼</sup> *f t*ð Þ

*=*dt S tð Þ ¼ � dlnS tð Þ

*S t*ð Þ

*f t*ðÞ¼ *S t*ð Þ*λ*ð Þ*t* (6)

*f t*ðÞ¼ *<sup>λ</sup>p*ð Þ *<sup>λ</sup><sup>t</sup> <sup>p</sup>*�<sup>1</sup> (7)

*ln ti* � *X*<sup>0</sup> *i β*

*σ*

� *exp*

� � �� (8)

*F t*ðÞ¼ Probð Þ *T* <*t* (2)

*dt* (3)

(4)

dt (5)

T can be specified by the following distribution function:

*DOI: http://dx.doi.org/10.5772/intechopen.85287*

than t periods. The corresponding density function is:

*λ*ðÞ¼ *t* lim *Δt*!0

λðÞ¼ t

uncensored and censored observations such that:

*i*¼1

*δi*

*ln ti* � *X*<sup>0</sup> *i β <sup>σ</sup>* � *ln<sup>σ</sup>* � �

ln *<sup>L</sup>*ð Þ¼ *<sup>β</sup>*, *<sup>α</sup>*j*data* <sup>X</sup>*<sup>n</sup>*

**203**

Or

And so that:

**3.2 Estimation method**

¼ lim *Δt*!0

f tð Þ S tð Þ <sup>¼</sup> dF tð Þ

#### **Figure 1.**

*Survival function by age group (labor force) obtained by non-parametric estimate of Kaplan-Meier. Source: Authors' calculations based on 2012 AGEPE data.*

#### **Figure 2.**

us to reject the exponential model (constant randomness) since it results in a decreasing monotonous distribution of durations, that is to say, a growing chance (see **Figures 1** and **2**). Anything that reinforces our choice of a Weibull model.

Let us note: T is the duration of unemployment; T\* is the random unemployment duration of an uncensored individual; c is the censored unemployment duration of an individual who has still not found a job at the time of the survey and censored, an indicator variable equal to 1 if the observation is censored and 0 otherwise.

$$\begin{cases} T = \mathfrak{c}, \text{if } \mathit{crsored} = \mathfrak{1}. \\\ T = T^\*, \text{otherwise.} \end{cases} \tag{1}$$

In addition, the model assumes that individuals whose unemployment durations are censored are representatives of all individuals whose unemployment durations

*A Gender Analysis of the Determinants of Youth Unemployment in Côte d'Ivoire DOI: http://dx.doi.org/10.5772/intechopen.85287*

are at least equal to c. If we consider c as a random variable, then c must be independent of T\* after taking into account the other factors explaining the duration of unemployment. The distribution of probabilities of unemployment duration T can be specified by the following distribution function:

$$F(t) = \text{Prob}(T < t) \tag{2}$$

*F(t)* then represents the probability that an unemployment duration T lasts less than t periods. The corresponding density function is:

$$f(t) = \frac{dF(t)}{dt} \tag{3}$$

By *f(t)dt*, we then have the probability that the unemployment duration T will end between t and (*t+dt*) periods. These functions (distribution and density) make it possible to determine the survival function S (t) and the function (t) with:

$$\begin{split} \lambda(t) &= \lim\_{\Delta t \to 0} \frac{\text{Prob } (t \le T \le t + \Delta t | T \ge t)}{\Delta t} \\ &= \lim\_{\Delta t \to 0} \frac{F(t + \Delta t) - F(t)}{\Delta t \text{ S}(t)} = \frac{f(t)}{\text{S}(t)} \end{split} \tag{4}$$

Or

$$\mathbf{\hat{x}(t)} = \frac{\mathbf{f(t)}}{\mathbf{S(t)}} = \frac{\mathrm{d}\mathbf{F(t)}\acute{\bigtriangleup}\_{\mathrm{dt}}}{\mathbf{S(t)}} = \frac{-\mathrm{d}\mathbf{S(t)}\acute{\bigtriangleup}\_{\mathrm{dt}}}{\mathbf{S(t)}} = -\frac{\mathrm{d}\ln\mathbf{S(t)}}{\mathbf{d}t} \tag{5}$$

And so that:

$$f(t) = \mathbf{S}(t)\boldsymbol{\lambda}(t) \tag{6}$$

Generally, the identification of the factors that affect the probability of exit from unemployment to employment is analyzed using the risk functions. We subscribe to this logic. Moreover, Steiner [39] emphasizes that risk functions can be interpreted as reduced forms of the basic job search model as elaborated by McCall [40].

#### **3.2 Estimation method**

The duration models lend themselves to various types of estimation. The approach chosen for the estimates is the parametric method and, more specifically, proportional hazard models [41]. Suppose t follows a Weibull distribution, noted:

$$f(t) = \lambda p \left(\lambda t\right)^{p-1} \tag{7}$$

where t is an embodiment of T; λ is the risk function and p is a scale parameter. These parameters (λ and p) can be estimated by the maximum likelihood method.

The formulated likelihood function is the sum of the likelihood functions of uncensored and censored observations such that:

$$\ln L(\beta, a|data) = \sum\_{i=1}^{n} \left[ \delta\_i \left( \frac{\ln t\_i - X\_i^\prime \beta}{\sigma} - \ln \sigma \right) - \exp \left( \frac{\ln t\_i - X\_i^\prime \beta}{\sigma} \right) \right] \tag{8}$$

us to reject the exponential model (constant randomness) since it results in a

*Youth survival function by sex obtained from non-parametric Kaplan-Meier estimation. Source: Authors'*

**Figures 1** and **2**). Anything that reinforces our choice of a Weibull model.

(

otherwise.

**202**

**Figure 2.**

*calculations based on 2012 AGEPE data.*

**Figure 1.**

*Authors' calculations based on 2012 AGEPE data.*

*Regional Development in Africa*

decreasing monotonous distribution of durations, that is to say, a growing chance (see

*Survival function by age group (labor force) obtained by non-parametric estimate of Kaplan-Meier. Source:*

Let us note: T is the duration of unemployment; T\* is the random unemployment duration of an uncensored individual; c is the censored unemployment duration of an individual who has still not found a job at the time of the survey and censored, an indicator variable equal to 1 if the observation is censored and 0

> *T* ¼ *c*, *if censored* ¼ 1*: <sup>T</sup>* <sup>¼</sup> *<sup>T</sup>*<sup>∗</sup> , *otherwise:*

In addition, the model assumes that individuals whose unemployment durations are censored are representatives of all individuals whose unemployment durations

(1)

where *σ=1/p*; *δi=1* for individuals having completed their period of unemployment; *δi=0* for those still unemployed and *Xi*' is the vector of the explanatory variables

EEMCI provides information on the socio-economic and demographic characteristics of 11,600 households, or 49,590 individuals. Among them, there are 28,875 individuals of working age or 58.23%. The base sample comprises 19,115 individuals aged 14–35 years. This young population represents 66.20% of people of working age and is composed as follows: 61.56% of employed persons (11,767 individuals); 7.30% (1394 young people) unemployed and 31.14% inactive (5954 young people).

*A Gender Analysis of the Determinants of Youth Unemployment in Côte d'Ivoire*

*DOI: http://dx.doi.org/10.5772/intechopen.85287*

The dependent variable of the model is the duration of unemployment, contin-

It is expected that a married woman would tend to stay in unemployment longer than an unmarried woman by the presence of a spouse who is able to support the family. This situation is accentuated when there are dependent children. In addition, young women living in urban areas should experience longer unemployment durations. In addition, more educated individuals would be less likely to be in a

uous variable expressed in months. From the various works cited above and according to the variables available in the database, we retain these sociodemographic variables (age, sex, place of residence, marital status and level of education) as explanatory variables. Each of the variables has several modalities that are supposed to have different effects on the duration of unemployment (see

At the level of the non-parametric approach, a first analysis is made by distinguishing according to the age group in order to capture the specificities that could exist between young people and adults. The parametric approach is discussed

Kaplan-Meier estimators of the non-parametric approach for survival functions by age group show that adult unemployment durations are generally shorter than those of youth (see **Figure 1**). It also appears that the unemployment durations of young women are longer than those of young men (see **Figure 2**). Their survival function above that of young men confirms their difficulties in entering the labor

If we combine the age criterion with that of gender, we can, in relation to the level of the rates, constitute two groups of young people. The youngest (14–24 years old), and particularly young women, can be classified as the most group. Young men over the age of 24 are significantly less exposed to unemployment than those in

**Table 3** (annexed) presents the results of risk function estimates taking into account the heterogeneity between individuals. It traces the explanatory factors of

Among the factors that expose workers to long-term unemployment, the analyses tend to highlight individual characteristics. Among these characteristics, it is common to distinguish "demographic" individual characteristics (sex, age and

the duration of unemployment among young people by sex.

**3.3 Model variables**

**Table 2** in the appendix).

**4. Econometric results**

**4.1 Non-parametric approach**

later.

market.

**205**

the first group.

**4.2 Parametric approach**

state of prolonged unemployment.

As specified, the model may suffer from a heterogeneity problem. This problem can be considered in the duration models as the main result of an incomplete specification. It is usually due to the fact that the observed process may have started at different points in the calendar for the different individuals in the sample. To take into account this heterogeneity, a random element *υ<sup>i</sup>* summarizing heterogeneity not taken into account in the model is introduced in the parametric model<sup>5</sup> and the law of duration is rewritten conditionally to this term.

If there is apparently no general theorem in the choice of the distribution of this term, the most recurrent laws are the Gamma law and the inverse-Gaussian law (Inverse Gaussian).<sup>6</sup> The Gamma distribution is frequently used in this type of analysis because the value of θ is higher with this specification. Thus, in the case where *υ<sup>i</sup>* follows a Gamma distribution of mean 1 and variance θ, the heterogeneity test is to check if the parameter θ is statistically different from zero.

Data from the study came from the Employment Survey of Households in Côte d'Ivoire (EEMCI) conducted in 2012 by the Agency for Studies and Promotion of Employment (AGEPE), a structure under the supervision of the Ministry of Employment. The purpose of the survey was to update the employment indicators, with a view to constitute a reference for the evaluation of current initiatives in the field of employment promotion.


#### **Table 2.**

*Descriptive statistics of model varia.*

<sup>5</sup> The strictly non-parametric approach of the Kaplan-Meier estimator is robust to this problem.

<sup>6</sup> Finding done by Jenkins [42].

*A Gender Analysis of the Determinants of Youth Unemployment in Côte d'Ivoire DOI: http://dx.doi.org/10.5772/intechopen.85287*

EEMCI provides information on the socio-economic and demographic characteristics of 11,600 households, or 49,590 individuals. Among them, there are 28,875 individuals of working age or 58.23%. The base sample comprises 19,115 individuals aged 14–35 years. This young population represents 66.20% of people of working age and is composed as follows: 61.56% of employed persons (11,767 individuals); 7.30% (1394 young people) unemployed and 31.14% inactive (5954 young people).

#### **3.3 Model variables**

where *σ=1/p*; *δi=1* for individuals having completed their period of unemployment; *δi=0* for those still unemployed and *Xi*' is the vector of the explanatory variables As specified, the model may suffer from a heterogeneity problem. This problem

If there is apparently no general theorem in the choice of the distribution of this term, the most recurrent laws are the Gamma law and the inverse-Gaussian law (Inverse Gaussian).<sup>6</sup> The Gamma distribution is frequently used in this type of analysis because the value of θ is higher with this specification. Thus, in the case where *υ<sup>i</sup>* follows a Gamma distribution of mean 1 and variance θ, the heterogeneity

Data from the study came from the Employment Survey of Households in Côte d'Ivoire (EEMCI) conducted in 2012 by the Agency for Studies and Promotion of Employment (AGEPE), a structure under the supervision of the Ministry of Employment. The purpose of the survey was to update the employment indicators, with a view to constitute a reference for the evaluation of current initiatives in the

**Average Standard deviation Average Standard deviation**

can be considered in the duration models as the main result of an incomplete specification. It is usually due to the fact that the observed process may have started at different points in the calendar for the different individuals in the sample. To take into account this heterogeneity, a random element *υ<sup>i</sup>* summarizing heterogeneity not taken into account in the model is introduced in the parametric model<sup>5</sup>

and the law of duration is rewritten conditionally to this term.

test is to check if the parameter θ is statistically different from zero.

**Variables Young man Young woman**

Single 0.519 0.492 0.387 0.487 Married 0.403 0.491 0.598 0.490 Widower/divorced 0.005 0.750 0.151 0.122

Abidjan 0.190 0.392 0.199 0.400 Other urban 0.226 0.419 0.229 0.420 Rural 0.583 0.493 0.571 0.494

No 0.440 0.496 0.574 0.494 Primary 0.276 0.447 0.226 0.442 Secondary 0.223 0.416 0.130 0.335 Superior 0.061 0.240 0.300 0.170

Young 0.350 0.477 0.420 0.493 Young adult 0.651 0.477 0.579 0.493

<sup>5</sup> The strictly non-parametric approach of the Kaplan-Meier estimator is robust to this problem.

field of employment promotion.

*Regional Development in Africa*

*Marital status*

*Middle of residence*

*Level of education*

*Groupe d'âges*

**Table 2.**

**204**

*Source: Authors' calculations based on 2012 AGEPE data.*

*Descriptive statistics of model varia.*

<sup>6</sup> Finding done by Jenkins [42].

The dependent variable of the model is the duration of unemployment, continuous variable expressed in months. From the various works cited above and according to the variables available in the database, we retain these sociodemographic variables (age, sex, place of residence, marital status and level of education) as explanatory variables. Each of the variables has several modalities that are supposed to have different effects on the duration of unemployment (see **Table 2** in the appendix).

It is expected that a married woman would tend to stay in unemployment longer than an unmarried woman by the presence of a spouse who is able to support the family. This situation is accentuated when there are dependent children. In addition, young women living in urban areas should experience longer unemployment durations. In addition, more educated individuals would be less likely to be in a state of prolonged unemployment.
