Modelling the Effects of Public Support to Small Firms in the UK - paradise Gained?
MODELLING THE EFFECT OF BUSINESS LINK SUPPORT ON GROWTH
Introduction
This section outlines the results of our econometric analysis based primarily on the
interviewed assisted and non-assisted businesses. The analysis presented in this
section is in two main stages:
(a) An initial stage looking at those factors which influence the probability
that a firm received BL assistance during 1996, i.e. that a firm falls in
the assisted rather than the comparator group. This is modelled using a
series of probit equations based on the characteristics of firms in 1996,
their growth from 1994-96 and the sources of information through
which they may have accessed information about BL. These probit
models and their implications are described below.
(b) A second stage outlining the estimation of regression models for the
productivity (i.e., turnover per employee) growth of assisted businesses
and the non-assisted comparator businesses. Key interest here focuses
on the potential importance of BL support (the ‘assistance’ effect) and
whether there is any systematic ‘selection’ effect. That is whether the
selected assisted businesses were faster or slower growing than the
average.
The separation of the ‘assistance’ and ‘selection’ effects is important in getting a true
picture of whether the actual support provided by BL was a significant determinant of
productivity and sales growth. As indicated earlier, the econometric approach
adopted is intended to correct for any bias induced by the selection of firms to assist
by BL. Other forms of bias may occur, however, due to sample attrition, if over the
1996 to 2000 period firms in the assisted and comparator groups fail at different rates.
If failure is a random event then this bias will not be important. On the other hand, if
the rate of failure is affected by factors which are also linked either to the probability
of receiving assistance or business performance this sample attrition bias may be more
important. Our analysis suggests, however, that the characteristics of surviving
assisted and non-assisted businesses in the sample are essentially similar to those of
the firms in the original BL dataset. The implication is, as far as we can tell given the
limited information available in the BL dataset, that there is no significant sample
attrition bias.
What Factors Influenced the Probability of Receiving BL Assistance?
From the survey and the indicators database a wide range of firm characteristics,
owner-manager characteristics and market descriptors are available which might
impact on the probability of firms’ receipt of BL support. In statistical terms,
however, we would want to find a set of explanatory variables for the probit model
which are unlikely to influence the level of turnover growth post-1996.
Stephen Roper and Mark Hart
13