4.3  Robustness testing

As already mentioned this paper should be viewed as a microeconomic counterpart to the productivity work presented in Black, Guy and McLellan (2003) (hereafter BGM). Figure 5 compares aggregate labour productivity growth rates derived in both papers. There are some differences in industry coverage between the two series. Most notably BGM include agriculture while the current analysis does not. Another difference is that BGM are able to take account of changes in inventories of final goods.

Despite these differences the two series appear to be quite similar. For the period 1995 to 2002 average labour productivity growth as estimated by BGM was 1.1% as compared to 1.6%. Both series exhibit the same pattern. Current analysis gives a more volatile labour productivity growth series than does BGM. This is not surprising as inventories of final goods tend to be counter cyclical in New Zealand and act as a buffer to changes in demand (Buckle and Meads 1991).

Figure 5 – Labour productivity growth comparison

As explained in Appendix 1, when constructing the data set used for analysis thus far a number of firms were observed to have missing employment or GST sales information over part of their lifetime. There are a number of possible reasons for this but the most likely explanation is the 10% to 15% non-response rate to the Annual Business Frame Update (the survey used to update the Business Demography Statistics database). To avoid losing observations the approach taken was to impute missing information based on the most recent data available for those firms.

An alternative approach would be to simply discard any firm that has missing information at any point during its lifetime. This approach means that observations on labour productivity fall from approximately 2.4 million to less than 1 million.

Table 9 shows how the choice of data set affects aggregate entry, exit and turnover rates. Switching to the alternative data set does not alter the turnover rate by much, but does increase the entry rate and reduce the exit rate slightly.

Notes – All numbers are percentages and are the arithmetic averages of yearly observations between 1995 and 2003. The benchmark data set is the data set used throughout this paper where missing employment and GST sales information have been imputed. The alternative data set is the data set used only in this section where firms with missing employment or GST sales data have been discarded.

Table 10 shows how the choice of data set affects the GR aggregate decomposition. Switching to the alternative data set lowers average labour productivity growth for the period 1995-2003 from 1.8% to around 0.9% per annum. The signs of all decomposition components however remain unchanged.

Notes – All numbers are percentages and are the arithmetic averages of yearly observations between 1995 and 2003. The entering, exiting, within and between components sum to their respective totals for each data set. The benchmark data set is the data set used throughout this paper where missing employment and GST sales information have been imputed. The alternative data set is the data set used only in this section where firms with missing employment or GST sales data have been discarded.

Table 11 shows how the choice of data set affects the proportion of firms making positive contributions to their respective GR aggregate decomposition components. The overall pattern remains unchanged. However, a slightly higher proportion of continuing firms are able to make positive contributions to both the within and between components. A slightly lower proportion of entering firms are able to make positive contributions to labour productivity growth. A slightly higher proportion of exiting firms exit with below average labour productivity and hence make positive contributions to labour productivity growth.

Notes – All numbers are percentages and are the arithmetic averages of yearly observations between 1995 and 2003. The benchmark data set is the data set used throughout this paper where missing employment and GST sales information have been imputed. The alternative data set is the data set used only in this section where firms with missing employment or GST sales data have been discarded.