Modelling the Effects of Public Support to Small Firms in the UK - paradise Gained?
Methodological Lessons Learnt
Three main methodological issues are relevant to these results. First, the use in the
models of survey rather than the indicators database data for 1996 and 2000. Secondly
the issue of contamination (i.e. situations where firms have had contact with BL post
March 1998), and thirdly other methodological uncertainties relating to the results.
Whilst survey data is never perfect, the value of the survey was demonstrated by rerunning
the analysis using the IDBR data collated over the period of the study. To test the effect of
using database rather than survey information for 1996 and 2000 the equations in
Table 4, as well as equations for turnover and employment growth, were re-run using
the Impact Indicators database information. Some notable differences were evident
between the signs and significance of the explanatory variables relating both to the
‘conditioning’ factors (i.e. business strategy, prior performance, firm characteristics,
the owner-manager) and the selection and assistance effects. For turnover (and
employment) growth the assistance and selection parameters estimated using the
indicators data were essentially similar to those estimated using the survey data. In
terms of productivity growth, however, the selection and assistance effects proved
insignificant in the equations run using the indicators database although the sign
pattern was consistent. That is, using the indicators database, the BL assistance effect
on productivity growth was positive but insignificant while the selection effect was
negative but insignificant.
The importance of this result depends crucially on our view of the reliability of the
impact indicators and survey information for turnover and employment growth over
the 1996 to 2000 period. Our general preference is to rely on the survey data and
results and to regard the consistency of the sign pattern of the selection and assistance
effects on productivity growth as reassuring.
To examine the potential impact of post-1998 BL contact it useful to realise that post-
1998 contact with BL might impact on turnover or employment growth over the
1996-2000 period but can have had no influence on whether a firm is an action or
control. The potential issue of contamination therefore relates solely to the second
element of the selection modelling and its impact can be assessed simply by including
in the equations an additional dummy variable for post-1998 contact with BL. One
practical difficulty here is the relatively large number of ‘don’t knows’ in the
responses to a question probing on the last contact with Business Links (before or
after March 1998), reducing further the number of observations in the sample
selection models. In both the productivity and turnover equations, however, the
contamination dummy variable proves insignificant suggesting that any contamination
effect is likely to be small.
One final methodological issue relates to other possible uncertainties in the analysis
which might be taken up in future studies. For example, the measures of sales growth
and productivity growth used here are both based on the total sales of valid Actions
and Controls. For manufacturing firms in the sample one clear possibility is that
increases in sales could be the result of increases in merchanting, factoring or sub-
contracting rather than increased sales of their own production. If the extent of such
developments differed between the assisted and non-assisted companies, inferences
about ‘productivity’ or growth improvements may be misleading. While there is no a
Stephen Roper and Mark Hart
20