4.3 Estimation results
4.3.1 Ownership of small businesses and upward mobility
Let us first address the effect of being an owner of a small business on upward mobility. As
estimations are based on a random effect probit model and as the key explanatory variable is a dummy
variable, we need to describe the procedures by which we identify the effects of the explanatory
variables. In the tables we report marginal effects, calculated for predicted probabilities of positive
outcomes.11 The effect of the discrete variable representing business ownership or not is calculated for
a discrete change from 0 to 1 in this dummy variable, holding the other covariates at their means. The
estimations also involve time dummies, the coefficients of which are not reported.12
Table 10 shows results for two specifications of conditional probabilities or transitions rates:
one based on quintiles, transition into the upper quintile (quintile 5) in year t if located in quintile 1-4
in year t-1, and one based on deciles, transition into decile 10 in year t if placed in deciles 1-9 in year
t-1. Independent of specification, and as expected, there is a positive relationship between business
ownership and income advancement; using the wider definition of business owners enhances this
positive relationship. However, the effect of business ownership and of the wider definition are
modest: being a business owner increases the probability for moving into quintile 5 by 3.3 percent on
the narrow definition and 3.8 percent on the wide definition. The corresponding estimates for moving
upward to decile 10 are 1.3 percent and 1.6 percent on the narrow and wide definition, respectively.
As noted when discussing the mobility tables of Section 3, we may expect to find a stronger
relationship between business ownership and upward mobility later in the period 1993-2003. Indeed,
estimations restricted to transitions in 2001, 2002 and 2003 do show somewhat stronger effects, 4.8
percent according to the quintile specification for the wide definition of business owners.
11 The xtprobit routine of Stata is used.
12 All tables report estimates for the proportion of the total variance contributed by the person level variance, rho. When rho
is zero, a panel model, such as the random effects model, is not a significant improvement.
23