• Keine Ergebnisse gefunden

Estimation Methodology for Multinomial Logit Model: Factors Influencing Occupa- Occupa-tional Choices

Chapter 8 provides findings from multinomial logit model on the factors affecting household activity participation, a procedure never applied in any study in the area. In the penultimate

3. METHODOLOGY AND FIELD RESEARCH INSTRUMENTS

3.9 Analytical Methodology Applied to the Research Questions

3.9.4 Estimation Methodology for Multinomial Logit Model: Factors Influencing Occupa- Occupa-tional Choices

The fourth question is aimed at highlighting not only the occupational activities of the respon-dents but also the factors influencing household decision559 to engage in these diverse occupa-tions. The previous research projects in the proposed six villages in Northwest Pakistan differen-tiated more than 200 jobs or occupations that were classified into different occupational categories (MANIG, 1991, p. 57). Hence, based on these studies, we considered seven broad occupations in the study area: private permanent employees; public sector employees; informal private wage labor; own enterprise activities; farming; and other remunerated activities.

Factors considered important in determining the occupational choice model are included as regressors, which in fact reflect the heterogeneity of incentives and constraints facing household occupational decisions. Some of these regressors merit comments. We expect individual charac-teristics (age, education) to strongly influence activity choices (although decision making may not be exclusively individual in this part of the world!). We expect the youngest to engage in non-farm sector due to its demand of muscles. We expect that after a certain age, individuals are more likely to retire, move out of the labour force. Therefore, to capture such possible non-linearity or life cycle for participation in different occupations we also introduced the square of age. Due to prevailing cultural-religious trends560, we expect engagement in various occupations to be clearly a male activity therefore gender was not included as regresser in the analysis.

Education as a variable will reflect the stock of skills at household level. We expect a positive and significant association between education levels and jobs in the organized sector at the household levels as shown by a number of studies in different developing country contexts (REARDON et al., 2000). Better educated individuals are likely to possess skills and aspirations to working outside agriculture which facilitate successful involvement in non-farm activities. In contrast, the usefulness of formal education for successful participation in the rural labor market is not always evident as the skills required to engage in many rural activities are either very simple or acquired outside the formal school system, through relatives and friends or on-the-job training. The formal educational system of Pakistan is having a traditional and narrow focus on academic disciplines, resulting in deficient technical and life skills for students.

59 SCHUTZ, 1964 as mentioned in CAMILLA et al., 2000 demonstrated that rural people in developing world consid-ered as traditional, illiterate and steeped in irrational practice are in fact conscious and reasoning decision makers, aiming to achieve certain goals in face of a set of constraints.

60 An excellent description of Pathan society is given in ALBRECHT, 1976 and MANIG, 1991.

The role of household size is vague as on one hand it may supply more economically active adults but what be the case when there is a high dependency ratio? To account for the labour resources at the household level, the variable on number of economically active members is also introduced. A high wage rate in other sectors makes the farming unattractive. However, due to possible endogeneity problems, we did not consider the effect of income on economic participa-tion. Similarly, higher landholding and livestock661 units may not only broaden the choices for households to engage in formal sectors but it may also encourage them to specialize in farming.

Finally, in order to capture the effect of the geographical variation, we introduce village dummies in the econometric model. It will take account of the availability of hard as well as soft infra-structure available to individuals living in these villages.

To identify the determinants behind rural household decision to engage in various occupations in Northwest Pakistan we assume that in a given period rational662 household choose among the seven63 mutually exclusive occupational alternatives that offers the maximum utility64. Follow-ing GREENE, 2003 and KENNEDY, 2003, suppose for the ith respondent faced with j choices, we assume the utility choice j as:

Uij = Zij β + εij (1)

If the respondent makes choice j in particular, then we assume that Uij is the maximum among the j utilities. Hence the statistical model is derived by the probability that choice j is made, which according to MADDALA, 1999 is:

Prob (Uij >Uik) for all other K ≠ j (2)

We further assume that the random utility error terms are distributed independently and identi-cally as log weibull distributions. Further, we have information on several characteristics of the employment decision makers therefore, we can make use of the multinomial logit model (LONG

61 Livestock can also act as a substitutable asset that can be sold in order to invest in land or small businesses, and vice versa (ELLIS & FREEMEN, 2004).

62 The behavioural notion of the model may be invoked here by considering rural households in the research villages as rational decision makers.

63 The research projects of 1967/68 and 1986/87 conducted by IRE, Goettingen University Germany in rural Northwest Pakistan identified these broad occupational categories. See for details MANIG, 1991. Implying the multinomial logit model will be inappropriate whenever two or more of the alternatives are close substitutes (KENNEDY, 2003) hence; we reduced the number of categories to six to avoid this situation.

64 Utility is maximised by expected earning gains from adopting an activity profile choice (BERHANU et al., 2007).

& FREESE, 2001; KENNEDY, 2003). We select this model not only because of the computational ease65 (MILLER & VOLKER, 1985) but also multinomial logit66 analysis exhibits a superior ability to predict occupational distribution67 (KEANE, 1992). While Probit versions of these models are theoretically possible, computation and identification issues limit their use (GREENE, 2003).

The multinomial logit model can allow us to estimate a set of coefficients βj corresponding to each occupational category as follows:

To identify the model, we impose the normalization by considering the parameter vector associ-ated with informal non-farm employment as zero (β1 = 0). So, the remaining coefficients βj

measures the change relative to the base group informal non-farm employment. The prob-abilities as, therefore,

65 Due to its computational ease, Multinomial logit is widely used in different areas of economics, social studies, and market research (mentioned in MADDALA, 1999; GREENE, 2003).

66 MNLT is a generalization of Logit Model based on the Random Utility Model (RUM) which is simply the likelihood of choosing a particular alternative (occupation) if the utility of that alternative is greater than utility of other alternatives for that respondent .

67 The study of occupational choice, by SCHMIDT & STRAUSS (1975) provides a well known application of Multino-mial Logit Model (GREENE, 2003).

Where Pr is the probability of an economic activity; i indexes the individuals; j represents the seven68 nominal unordered occupational categories; e is the natural log; and xi vector of exogenous variables affecting employment decision of household head. The model is estimated by maximum likelihood69, where the above probabilities enter the likelihood function. For further details see LONG & FREESE,2001; GREENE, 2003; and KENNEDY, 2003.

A simplification of the overall Multinomial Logit Model is as follow:

Formal = 1 if HH in private or government regulated sector Businessmen = 2 if HH in self employed in business and trade

Pure farmer = 3 if HH runs only own land

Multinomial Mixed farmer = 4 if HH works own land with non-farm job = f (Xi) Logit Analysis Pure tenant = 5 if HH is engaged in land tenancy

Mixed tenant = 6 if HH engaged in land tenancy with non-farm job Informal wage = 7 if HH rely on informal wage activities

Based upon the literature studied, the explanatory variables comprise a selection of individual, household and community characteristics. The informal private wage labor (casual workers including farm labor) as the base group70 is compared with the above mentioned occupations71 (polychotomous) and this indicate whether the other occupational categories can be regarded as systematically different in any way from the comparison group.