3. Data
All our data are from New Zealand's 2001 Business Practices Survey (BPS) conducted by Statistics New Zealand, the country's official statistical agency (Statistics New Zealand, 2002; Knuckey and Johnston, 2002; Fabling and Grimes, 2006). The survey included a wide range of business practices and firm characteristics. Being an official survey, its coverage was comprehensive (3,378 surveyed firms), its response rate excellent (82% response rate with 96% of respondents answering 90% or more of the questions); and its sampling approach meticulous.4 Questions in the survey were designed to test for the presence of "high performance" work practices posited by strategic management and economic literature (Knuckey and Johnston, 2002). The questions were either qualitative, or quantitative offering response ranges. We use both types of question in our empirical analysis, and test each category of response separately rather than use a restrictive representation of the responses (such as a Likert scale).
The questions contained under the category of "employee practices" in the survey are as follows:5
- Does this business systematically measure employee satisfaction?
- Are formal performance employee reviews used within this business (consistent methods that are recognised and regularly used)?
- How many employees are on "pay for performance" schemes (e.g. productivity based incentives, gain sharing, bonuses, etc)?
- In the last 12 months please estimate what proportion of this business's pre-tax payroll was related to employee education and training?
- Over the last 12 months please estimate the proportion of employees in this business who participated in in-house training?
- Over the last 12 months please estimate the proportion of employees in this business who participated in external training?
- Over the last 12 months please estimate the proportion of employees in this business who participated in job rotation/exchanges?
- Does this business have processes in place to manage health and safety (e.g. a training program, provision of information for employees)?
Using factor analysis, we take the first factor calculated across this list of practices as our measure of the suite of employee practices potentially relevant to firm performance. We denote this measure as SFEP where the EP relates to "employee practices" and the SF refers to the "special factor", being the factor derived solely from employee practice questions. (In subsequent analysis we differentiate this factor from general factors, GFj, j=1,.., n, derived from more general management practices and firm characteristics).
The BPS also surveys Business Results. Each firm is asked to record against a qualitative 3-point scale (plus "don't know") their firm's situation for three performance measures (our dependent variables):
- profitability relative to major competitors (denoted Pf);
- productivity relative to major competitors (Pd); and
- market share relative to three years' prior (Pm).6
We divide the responses to each of the business result questions into binary outcomes, grouping together the neutral and unfavourable responses as one outcome and the favourable responses as the other outcome for each variable. We undertake probit analysis on these data. Fabling and Grimes (2006) describes the reasons for grouping the responses into binary outcomes, rather than using all three response categories. The key reason is to compensate for potential respondent bias since (in each category) few respondents answered that their firm was doing relatively poorly.
In testing associations between business practices and firm performance, Fabling and Grimes (2006) found that ordered probit estimation gave the same signs on each variable as did the binary probit estimates, but the latter had preferred statistical properties (in keeping with prior expectations). That study also used an out-of-sample group to test whether the in-sample estimates had out-of-sample predictive power. For each of the performance measures, the in-sample results had strong out-of-sample predictive power, indicating that the performance measures do reflect underlying performance of the surveyed firms.7 This out-of-sample test is particularly important since the performance measures are self-reported. For these self-reported measures to have validity in the face of measurement issues raised by Bertrand and Mullainathan (2001), it was imperative that our estimated (in-sample) relationships determining high from low performance could also distinguish performance out-of-sample.
The survey data yield 2,147 observations for the Pf equation, 2,191 observations for Pd, and 2,529 observations for Pm (observations vary across the Pi since in each case we drop firms that answered "don't know"). For each performance measure, each of the binary result categories contains at least one-third of the observations. Our analysis seeks to determine whether choice of employee practices places an individual firm in the upper portion or lower portion of all firms for each performance measure.
Back to Top