income - in the year before retirement, and in the year of retirement - is larger for self-
employed than for unemployed. Furthermore, from those who were not poor, a larger
percentage of self-employed than of unemployed, fall into poverty. This probably
corresponds to that part of the population that works in low productivity activities, - small
agricultural workers, fishermen, or small artisanal workers for instance - that earn low and
unstable incomes and seldom discount to social security.
A Portuguese person that has retired before 65 years of age, and before that was an
employee working at least 15 hours per week, is relatively protected against falling into
poverty in the year of retirement. This may be explained by the mass of civil servants that
could retire before 65 in the considered period, and that had a well-built history of
contributions to social security.
6. Multivariate analysis: determinants of becoming poor
The analysis in section 5 enabled us to evaluate whether the transition to retirement
increases the probability of being poor and also for which groups this effect is stronger. But
being a bivariate analysis, the conclusions have to be considered carefully as compositional
effects may be present and then the marginal effects of each of the characteristics cannot be
properly identified.
We estimate in this section several probit models of the probability of becoming poor on
retirement, in order to try to evaluate which characteristics make some more likely to
became poor when they retire, overcoming the limitations of bivariate analysis.
In this analysis only the individuals that have retired during our sample period, that were at
least 50 years old in 1994, and that were not poor in the year before retirement are
considered.
Probit models are estimated considering the four definitions of poverty line that we have
been using and for each gender separately.
Among the explanatory variables we include those considered in previous section such as
gender, activity status in the year before retirement, and home ownership. We include also
other variables trying to account for differences by industry of activity in the year before
retirement, geographical region of residence, type of family and for being a civil servant or
15