Measuring Task Discretion and Organizational Commitment
The measure of task discretion was derived from responses to five questions that began with the common stem: ‘In general, how much influence do you have over the following?’ The questions then referred to what tasks were done, the pace of work, how the work was done, the order in which tasks were done, and the timing of the start and finish of the working day. Against each of these domains of control, respondents replied on a four-point scale.
Their responses are shown in Table 1. It can be seen that a substantial majority of respondents perceived that they had at least some influence in four of the domains, but that only a half of respondents felt that they had at least ‘some’ control of when they started and finished work. A third were completely constrained in the latter regard, experiencing no control at all.
Table 1. Distribution of Domains of Task Discretion, 2004
| ||Domains of control|
|What tasks||Pace at which work is done||How tasks are done||Order of tasks||Time of start and finish of workday|
For the subsequent analysis I computed a single measure capturing the overall level of task discretion in the job. Assigning cardinal values 1–4 respectively to the responses ‘none’ to ‘a lot’, an additive scale is obtained, entitled the Task Discretion Index (TDI), by averaging the values of all five variables, yielding a range from 1 to 4 and a mean of 3.002. Cronbach's alpha statistic measuring scale reliability for this measure is 0.815, which implies a good level of reliability. Alternative indices can also be used, in order to test the robustness of the findings. One alternative is to generate scores from a factor analysis. The principal factor method was used, and this extracted only one factor. In another alternative, the fifth domain was excluded from the scale (for both the additive scale and the factor score), since its correlation with the other domains was the lowest. In what follows, a broadly similar pattern of findings emerges from using any of these alternatives, so only the findings from the additive scale are presented.
Complementing employees' estimates of their own task discretion, managers were also asked three questions about the individual task discretion involved in the jobs of employees. Managers were asked ‘to what extent would you say that individuals (in the largest non-managerial occupational group in the establishment) have discretion over how they do their work’. Subsequent questions asked about having ‘control over the pace at which they do their work’ and ‘involvement in decisions over how their work is organized’. Respondents could answer ‘a lot’, ‘some’, ‘little’ or ‘none’. The responses to these questions were averaged to generate a separate additive scale (Cronbach's alpha = 0.723), to be entitled the ‘Task Discretion Index, Managers’ Perception' (TDIMP), again ranging from 1 to 4. Earlier studies have found that there tends not to be a high correlation between managers' and employees' perceptions of task discretion; nevertheless, it is of interest to examine the extent to which the TDI and the TDIMP scales are correlated in the WERS04 data. For this purpose, I computed the mean TDI at establishment level, for only those employees who belonged to the largest occupational group. The mean establishment-level estimate of the employees' perception of discretion is measured imprecisely because of the limited numbers in each establishment who were issued with and responded to the self-completion questionnaire. In the event, the correlation coefficient between the mean establishment-level TDI and the TDIMP was 0.210, significantly positive with a p-value of 0.00. Restricting the sample to those few establishments (86) with at most 25 employees and where more than 50 per cent of employees responded on this question, the correlation coefficient is somewhat higher, at 0.315.
Table 2 shows the variation in task discretion across major occupational groups and across the education levels of the employee respondents. As Table 2 shows, the TDI and TDIMP are both broadly related as one would expect with the major occupational groups: managers and professionals and associate professionals, typically seen as the high-skilled groups, report above-average levels of discretion. Nevertheless, aside from these groups, there is less of a gradient of the TDI between traditional conceptions of occupational skill level and discretion. Table 2 also brings out that there is a positive association between employee discretion and education levels. Nevertheless, this association is shown only to apply within the upper levels of the education spectrum. At level 3 (two or more A-levels) and below there is essentially no relationship between education and task discretion, but there is a clear upward gradient between levels 3 and 5 (higher degree).
Table 2. Task Discretion Indices by Major Occupation Group
|Administrative and secretarial||2.99||2.98|
|Plant and machine operatives||2.73||2.43|
|Education level (equivalences)|
|0. No qualifications||2.95||—|
|1. GCSE grade D–G||2.92||—|
|2. GCSE grade A–C||2.88||—|
|3. Two or more A-levels||2.97||—|
|4. Batchelor's degree||3.11||—|
|5. Higher degree||3.21||—|
Particular cases at the 2-digit level also serve to make the point that the TDI is broadly in line with prior expectations. Marketing and sales managers, for example, have high levels of discretion (mean value 3.59) as do production, works and maintenance managers (3.46). By contrast, examples of occupations with low levels of discretion include call centre operators (2.31) and bus, van and coach drivers (2.57). One reason why elementary occupations do not all show especially low discretion levels, despite their low-skilled tag, is that this group embraces occupations that nevertheless require non-routine processes. Cleaners and domestics, for example, have slightly above-average discretion (3.05), despite being classed as low-skilled.
Task discretion has been found in detailed case studies and in earlier empirical work to be related strongly to job satisfaction (e.g. Green and Tsitsianis 2005). WERS04 asks employees about seven separate domains of job satisfaction, each measured against a five-point scale ranging from ‘very satisfied’ to ‘very dissatisfied’. Four of these domains pertain to intrinsic aspects of the job (sense of achievement, scope for initiative, amount of influence, the work itself) while the remaining three tap extrinsic aspects (pay, security and training). Assigning values 1 to 5 to the response points, I generated a simple additive index of intrinsic job satisfaction (Cronbach's alpha = 0.849). The individual-level correlation between this intrinsic job satisfaction index and the TDI was 0.371. Some validation of the discretion data is evident in this strong correlation.7
To measure organizational commitment, WERS04 asks three items drawn from the Organizational Commitment Questionnaire (Mowday et al. 1982). The questions asked respondents how far they agreed with the statements: ‘I share many of the values of my organization’, ‘I feel loyal to my organization’ and ‘I am proud to tell people who I work for’. The responses were against the scale: ‘strongly agree’, ‘agree’, ‘neither agree nor disagree’, ‘disagree’, ‘strongly disagree’. While the number of items is less than desirable, they form the core of the notion of affective commitment, essentially a measure of employee preferences concerning working for their employer.8 The responses from these three items were averaged to generate an additive scale of organizational commitment ranging from 1 to 5, with a Cronbach's scale reliability coefficient of 0.850.
Estimating the Model of Task Discretion
Tables 5 and 6 present estimates of the impact of organizational commitment and of other variables on task discretion, with Table 5 giving the results for the employee-level measure of discretion (TDI) and Table 6 for the establishment-level measure (TDIMP). In order to be able to compare better the findings from the two levels of analysis, the analysis in Table 5 is based only on employees in the non-managerial occupation groups.
Table 5. Determinants of Employee Task Discretion
| ||OLS||IV||Estab. FE|
|Occupation, ref: sci and tech profs|
|Teaching and research professionals||0.043||0.045||0.011|
|Business and public-service profs||−0.041||−0.039||0.073|
|Sci and tech associate professionals||−0.143||−0.143||−0.138|
|Health and soc. welfare assoc. profs||0.040||0.042||−0.053|
|Protective service occupations||−0.088||−0.086||−0.195|
|Business/public-service assoc. profs||0.067||0.068||0.115|
|Secretarial and related||−0.175||−0.173||−0.129|
|Skilled agricultural trades||−0.049||−0.048||−0.202|
|Skilled metal and electrical trades||−0.040||−0.040||−0.031|
|Skilled construction and building||0.051||0.052||−0.023|
|Caring personal service||−0.192||−0.190||−0.302|
|Leisure/other personal service||−0.213||−0.210||−0.020|
|Process, plant and machine operatives||−0.166||−0.167||−0.136|
|Technology and work organization variables|
|Number of uses of computer in job||0.029||0.029||0.033|
|‘Just-in-time’ production||−0.017||−0.017|| |
|Proportion working at home||0.175||0.178|| |
|Index of team use in largest occupational group||−0.066||−0.065|| |
|Index of self-led team in largest occupational group||0.025||0.025|| |
|Index of team discretion in largest occupational group||0.062||0.063|| |
|Flexible working time arrangements|
|Flexible shifts||−0.033||−0.033|| |
|Compressed-hours working||−0.001||−0.001|| |
|Annualized hours job||−0.043||−0.043|| |
|Zero hours work||−0.039||−0.040|| |
|Quality monitoring arrangements|
|Monitoring by manager/supervisor||−0.041||−0.042|| |
|Monitoring by inspector(s)||−0.002||−0.002|| |
|Monitoring by records of faults||−0.018||−0.018|| |
|Monitoring by customer surveys||−0.008||−0.007|| |
|No targets||0.023||0.023|| |
|25 or more employees||−0.000||−0.001|| |
|Trade union member||−0.090||−0.091||−0.047|
|Mean (s.d.) of dependent variable||2.928||2.928||2.928|
Table 6. Determinants of Managers' Estimates of Employee Task Discretion
|Employee commitment (managers' perception)||0.117||0.260|
|Occupation, ref: professional occupations|
|Administrative and secretarial||−0.031||−0.022|
|Plant and machine operatives||−0.285||−0.243|
|Technology and work organization variables|
|Arrangement to work from home in normal hours||0.246||0.231|
|Proportion working at home (almost) always||0.599||0.484|
|Index of team use in largest occupational group||−0.227||−0.225|
|Index of self-led team in largest occupational group||0.116||0.106|
|Index of team discretion in largest occupational group||0.450||0.426|
|Flexible working time arrangements|
|Annualized hours job||−0.041||−0.082|
|Zero hours work||−0.001||0.018|
|Quality monitoring arrangements|
|Monitoring by manager/supervisor||−0.075||−0.081|
|Monitoring by inspector(s)||−0.019||−0.014|
|Monitoring by records of faults||−0.105||−0.091|
|Monitoring by customer surveys||0.028||0.023|
|25 or more employees||−0.096||−0.077|
|Percent union members in establishment||−0.00124||−0.00102|
|Mean (s.d.) of dependent variable||2.894||2.894|
Column (1) of Table 5 gives the OLS estimates, while column (2) presents estimates using instruments for organizational commitment, and column (3) presents fixed-effects estimates which control for establishment-wide unobserved effects on job design.
Variables used as instruments for organizational commitment in column (2) are as follows. First, two variables are included which capture management's report on whether employees in the establishment are ‘led to expect long-term employment in this organization’. One dummy variable is included for ‘strongly agree’; another dummy represents ‘disagree’ or ‘strongly disagree’. Second, two variables are included which capture whether, in the management's view, ‘employees here are fully committed to the values of the organization’. Again, one dummy variable captures ‘strongly agree’, while another represents ‘disagree’ or ‘strongly disagree’. By using variables taken from the management questionnaire, one can avoid potential common method bias, that is, the bias due to unobserved heterogeneity associated with personal traits affecting both dependent and independent variables, since presumably judgments made by manager respondents are not correlated with those made by individual employees. Using these variables as instruments depends on the assumption that they do not themselves affect job design for individual employees in the establishment except via the effect that they may have on the organizational commitment of individuals. Moreover, in order to provide well-defined instrumental variable estimates, the instruments should also have a strong association with organizational commitment.
As usual in such cases these assumptions could be questioned. For example, even though the expectation of long-term employment is not obviously connected directly to autonomy in the workplace other than through commitment, it would not be hard to manufacture a possible explanation. Accordingly, diagnostics tests are needed to examine whether the instrumental variable assumptions are satisfied in practice. The Hansen J statistic for overidentification is computed to be 2.486, which implies that one cannot reject the null hypothesis that the instruments are uncorrelated with the error term in the equation estimating task discretion (; p-value = 0.478). In that sense, they are correctly excluded from the specification.
To test whether the task discretion is underidentified, the Anderson canonical correlation LR statistic is computed to be 209.21, which implies that the null hypothesis that the equation is not identified (the instruments not correlated with commitment) can be rejected (; p-value = 0.000). There is sufficient correlation between the instruments and employee organizational commitment. However, it could still be the case that instruments are ‘weak’, which would mean that the estimates are biased (usually downwards) in finite samples and that the significance level is higher than implied by the reported t-statistics (Murray 2006). The test for weak instruments is the first-stage F-statistic, which is computed to be 52.54; this implies that the true significance level is below 10 per cent when the nominal level is 5 per cent (critical value 19.93). Thus, the instruments are not weak.
Consider now the findings from Table 5. Do they support the hypotheses proposed in Section 2?
An initial striking finding is that task discretion, as predicted (Hypothesis 1), is positively and strongly affected by workers' organizational commitment. This conclusion emerges first in the OLS estimate shown in column (1), but it is supported by the IV estimate shown in column (2), the latter showing only a slightly lower coefficient, not significantly different. I conclude that jobs for workers with greater commitment are indeed afforded greater discretion. Moreover, the direct impact is substantial: using the IV estimate, a one standard deviation increase in commitment raises task discretion by 0.182, which is 24 per cent of the standard deviation of task discretion.
Put another way, if we compare the job designs of workers who on average ‘neither agree nor disagree’ with the three organizational commitment items with otherwise similar workers who ‘strongly agree’ with the items, the effect on discretion of the raised commitment would be 0.44, more than the equivalent of switching from a customer service occupation to a science and technology professional occupation which would normally be considered to be much more skilled.
The link between task discretion and skill (Hypothesis 2) is investigated first by including 24 2-digit occupational dummies, on the presumption that higher-level occupations require greater skills. The least discretion is generally afforded to some of the lowest-ranking occupations (e.g. customer services, comprising call centre operators and other customer care occupations). Yet, there are exceptions with, for example, health professionals also showing low task discretion. Closer inspection reveals that this low-discretion finding for health professionals is mainly driven by pharmacists/pharmacologists in large workplaces. In all lines of work those in supervisory positions have, as expected and noted above, substantively greater levels of discretion. Another way of investigating Hypothesis 2 is through the link with computer usage. Jobs with a greater range of computer usages can be seen as loosely linked with skill and, as can be seen, the estimated impact of this variable is positive and significant.
Thus, the link with skill is weak although broadly positive, in line with prior information about labour processes in specific occupations, and in line with the normal expectation in sociological literature. Nevertheless, there are groups of workers with relatively low discretion despite their high skill levels. The mixed picture is consistent with the ambiguous story implied by equation (4). For such high-skill/low-discretion occupations the explanation is that any extra productivity that might be obtained from giving them more discretion than less-skilled workers is more than outweighed by the loss of output from potentially lower effort levels that might accompany greater discretion.
Hypothesis 3 concerns the link between the person–job skills match and discretion. Respondents were asked ‘How well do the work skills you personally have match the skills you need to do your present job?’, and could answer on a five-point scale (Much higher/a bit higher/about the same/a bit lower/much lower). Table 5 shows that for a given job-skill level, workers who perceive that their skills match the required job skills are afforded more discretion than those who were under-skilled for the job. For this group of under-skilled (only about 5 per cent of the sample), employers have granted them less leeway because discretion for them would be less productive or even of negative value if it raised the frequency of inefficient actions. By contrast, those who thought that they had more skills than needed (55 per cent of the sample) were afforded neither more nor less than the discretion allowed for those whose skills were matched.
Consider now the additional variables included because they carry information about the technology or work organization that may independently affect the discretion that workers experience (thus affecting the production function). First I included an indicator of the use of a just-in-time (JIT) inventory control system on the grounds that this system requires less individual freedom to alter the pace and timing of work. Some 34 per cent of private-sector employees work in establishments deploying a JIT system in Britain (see Appendix). While the estimated effect on their discretion has a negative coefficient, it is not statistically different from zero in either the OLS or IV specifications. Thus, the one included establishment-level technical characteristic of production has a negligible association with discretion.
Aspects of work organization, however, are important. Homeworking, in particular, is expected to be associated with greater discretion, since for homeworkers managerial supervision of the labour process is restricted to problematic technologies of distance surveillance, uneasy home visits and the setting of behaviour-distorting targets (Felstead et al. 2003). The survey does not record whether individual workers are homeworkers, but it does report the proportion of employees who are working largely from home during working hours. Consistent with this expectation, workers in establishments with a larger proportion of homeworkers experience on average greater task discretion.
The effect of teamworking, as a distinct form of work organization, on task discretion has been investigated by Harley (2001) using data from WERS98. As Harley describes, two distinct hypotheses have been posed regarding the role of teamworking in shaping the design of individual jobs. Optimistic perspectives associated with HRM (Harley cites several) have expected teams to raise employee discretion. By contrast, critical accounts have held that teamworking's putative liberating effect on job design was illusory and that teams instead led to new forms of control, to work intensification and to limited, rather than enhanced, discretion. Harley's article constitutes an advance in our understanding of the effect of teams on employees. Using the nationally representative data available in WERS98 he finds that on average teams neither raise nor lower discretion significantly. Harley argues that the introduction of teams has had little or no effect because teams are managerially driven, and/or teams do not constitute a major change in hierarchical work organization. His analysis, however, is confined to establishments where either none or all of the establishment's largest occupational group (LOG) are in a team, and with further restrictions this means that only a minority of employees in WERS98 are included in the analysis. Moreover, Harley includes only one category of team in his analysis. Here I include both teams that appoint their own leaders and those that do not; and both teams which, according to managers, ‘jointly decide how work is done’ and those that do not. These two variables are each interacted with an index of the proportion of the largest group that is working in teams.
It can be seen from Table 5 that the impact of teamworking on discretion is significant but differentiated. Consider, first, teams where members do not jointly decide how work is done. Comparing establishments with no teamworking in the largest occupational group with establishments where there is 100 per cent teamworking, discretion is 0.65 lower, consistent with the critical accounts of teamwork. However, for those teams where team members jointly decide about work (covering establishments with 49 per cent of employees), the negative impact of teams is almost exactly neutralized: the joint impact is −0.002 and statistically insignificant. For these employees the essence of Harley's neutral finding is reproduced here. Finally, whether the team is self-led or otherwise appears to have no significant effect on whether the team enhances or diminishes employee discretion. These findings imply that, while the critical accounts of teamwork's effect on employees find support for about half of employees, there is a need to distinguish between team types in order to capture heterogeneity in their effects on work organization.
Also expected to have a positive association with individual discretion is where the firm introduces various flexible hours policies. One can sometimes distinguish between whether the policy is there to serve the flexibility needs of the employee (e.g. ‘flexitime’) and whether its function is mainly to provide flexibility for the employer (e.g. zero hours working). I expected the former to be associated with higher perceived discretion. The data, which is derived from the management questionnaire, allow us to identify whether each flexible working time arrangement is applied to some workers in the establishment, and not whether any given employee can access that arrangement. Nevertheless, it might be presumed that in many establishments the policies are generalized to all or most workers.
The pattern of coefficient estimates implies that task discretion for employees is raised where there is a ‘flexitime’ policy in place (no set starting and finishing times, although set overall hours). This finding is as expected, and serves if nothing else to confirm the reliability of workers' perceptions of discretion. Conversely, discretion is lowered in establishments where there are flexible shifts; and the coefficient estimates for a zero hours policy and for annualized work hours are negative although insignificant. These types of flexibility policies help employers to call on workers to work when employers want them to.
Where managers report having direct systems of quality, monitoring might also be expected to have a negative bearing on workers' task discretion. Managers were asked how they monitored the performance of employees, and allowed to state as many methods as they used, including direct supervisor/manager monitoring, monitoring by separate inspectors, self-monitoring, records of faults and complaints, customer surveys and other unspecified methods. Most establishments (82 per cent) use managers and supervisors to directly monitor quality, and this form of monitoring carries a negative coefficient. However, with a p-value of 0.16 the coefficient is not quite significant at conventional levels. The impacts on discretion of other forms of monitoring were negligible.
A further set of establishment characteristics concerns the use of targets. It was hypothesized by Gallie et al. (2004) that the growing use of targets to control production may have been one of the causes of the observed reductions in employee discretion during the 1990s. The idea is that where targets are in force, line managers might need to control work more closely to achieve them, but it is also possible that some targets could be imposed for employees precisely in situations where monitoring is costly. Responding managers were asked to state whether they had to meet any targets over a range of input and performance variables (profits, labour costs, sales, absenteeism and so on). A dummy variable was constructed to indicate whether or not any targets were used in the establishment. Only 12 per cent of employees worked in establishments with no targets. While the point estimate on the dummy variable for ‘no targets’ is positive it is not statistically significant. This finding suggests that a rising use of targets is unlikely to have been a major explanation for declining discretion during the 1990s, although it is conceivable the explanation would be more relevant in the public sector.
Person-level and establishment-level controls were also added, to account for otherwise unspecified factors that might influence job design. It is found that discretion is greater for older workers, and for non-whites. Discretion is set significantly lower for trade union members, a finding which has a straightforward interpretation. If employers fear that trade union members are more likely to behave in their own interests or those of the union, rather than the employer, they are likely to design jobs that afford workers less control over their actions. Alternatively, it could be that workers in low-discretion jobs are more easily organized.
While the estimates given in columns (1) and (2) have included standard errors adjusted for clustering within establishments, they do not allow for the possible unobserved effects of establishment characteristics on individual job design, some of which might be correlated with individual characteristics and hence generating biased estimates. By definition these establishment-specific characteristics are unobserved, but I take them to include both the effects of management culture and the particular production function of the establishment, both of which might be correlated with variables that are observed. The estimation shown in column (3) seeks to address this possibility. It shows the establishment-fixed effects estimates. As can be seen, there is little change from the magnitude of the coefficients given in columns (1) and (2), which implies that any unobserved fixed effects are largely orthogonal to the individual observed effects. Nevertheless, it is also the case that the R2 value is raised quite substantially from 0.18 to 0.32, suggesting that a notable amount of the variance of discretion can be accounted for by between-establishment variance. The test of the null hypothesis that the additions of establishment-fixed effects does not account for additional variance is rejected at the level p = 0.000, with F-statistic 1.882, critical value 1.000.
Some alternative specifications have been used in order to test the robustness of the findings.
One alternative was to utilize as independent variable the establishment-level index of task discretion derived from the reports of managers, TDIMP. It may be recalled that this variable applies to the discretion afforded, in the manager's view, to the largest occupational group in the establishment, which may not be the same as for other employees. Moreover, the variable to be explained here is the average discretion of employees in that group, rather than directly with the individual-level discretion in jobs. For these reasons, the analysis of TDI at the individual level has been preferred to the analysis of TDIMP at the establishment level. Nevertheless, it will be reassuring for the main findings if the same or similar relationships are shown at the establishment level, and with data from a different informant.
Table 6 presents the estimates of TDIMP across 1,554 establishments. I utilize the index of full employee commitment as perceived by the manager which ranges from 1 (‘strong disagreement’) to 5 (‘strong agreement’).11 As with the employee-level analysis, the managers' estimate of commitment may be endogenous, and for this analysis it is instrumented by the two variables capturing whether employees are led to expect long-term employment in the organization. The Hansen J statistic for overidentification was 0.092 (p = 0.762), suggesting that it is acceptable to reject the hypothesis that these instruments are correlated with the error term. The Anderson canonical correlation statistic is 105.56 (p = 0.00), which implies that the excluded instruments are correlated with organizational commitment; in other words, the equation is identified. Finally, the Cragg-Donald F-statistic was 53.6, which implies that the instruments are not weak.
Table 6 shows that discretion is enhanced in establishments with homeworking arrangements, and rises with the proportion of employees working at home. The index of team use is negatively associated with employee discretion (consistent with Table 5), but in establishments where teams are explicitly said to allow for teams to jointly decide how work is to be done, the teams are positively associated with individual discretion as perceived by managers: the coefficient for this group is calculated as 0.426 − 0.106 = 0.320, which is also found to be statistically different from zero (p = 0.007). In contrast to Table 5, then, this finding implies that there are some establishments (roughly half) where teams positively enhance discretion, in line with the story told by the more optimistic perspective on teamworking. The difference between this finding and the neutral finding using the individual-level data may be due either to the differing level of analysis or to the differing informants about discretion.12
Another distinctive finding from this establishment-level analysis is evidence that employees in establishments with no targets are here estimated to have substantially greater discretion than those in establishments where one or more targets are set. The difference is estimated as 0.172 in the IV estimates, which amounts to just under a quarter of one standard deviation in TDIMP. This finding contrasts with that for the individual-level analysis, which found only a small and insignificant effect. While one cannot be confident about the reasons for this difference in findings, one possibility is that managers in establishments that set targets feel at the same time that they are limiting employees' discretion, even if the employees do not experience it as any more restrictive than a no-target regime (and indeed the employees need not be aware of the targets).
Turning again to the central hypothesis of this article, this establishment-level analysis confirms that there is a strong association of organizational commitment with employee discretion. The IV estimate implies that moving from a state where employee commitment is neither agreed nor disagreed with (16 per cent of establishments), to a state where the manager strongly agrees that the employees are fully committed (19 per cent of establishments), is associated with a rise in TDIMP by 0.520, which is 72 per cent of the latter's standard deviation across establishments, and more than the average difference in discretion associated with moving from an elementary occupation to a professional occupation. The link with skill is also confirmed to be broadly positive, as implied in the occupational rankings (although with this establishment-level analysis there are no finer disaggregations of occupation than the 1-digit level).
Two further robustness checks were carried out. First, as an alternative to occupation as a measure of skill, in the individual-level analysis I entered the employees' achieved qualification level. This analysis showed that, after conditioning on all the other variables included hitherto in the analysis, the level of discretion increases between qualification levels 4 and 5 (as with the raw data shown in Table 2); however, discretion is also higher at levels 0 and 1 than it is at levels 2, 3 and 4. This finding reaffirms what the earlier analysis has shown — that the relationship between discretion and skill (here loosely measured by the job-holder's education level) is not necessarily unambiguously positive as is often assumed. However, the analysis also showed that the pattern of other findings was not substantially altered by the inclusion of education rather than occupation in the analysis.
Second, in a further estimation, the analysis was restricted to the employees who belonged to the LOG in the establishment. This sample restriction has the advantage that variables that were intended to apply to the LOG would be in principle more accurately measured; the disadvantage is that the sample size falls by more than half to 5,559. However, it is reassuring to confirm that the pattern of findings remains largely unchanged from those obtained with the full sample of 11,845 employees. The central finding of a substantial impact of commitment on discretion is again found, with a coefficient of 0.243 (0.014), which is not much different from the coefficient estimates shown in Table 5. The other conditioning variables follow the same pattern, but with one exception. For this restricted sample, consistent with expectations the presence of a just-in-time production system is negatively associated with discretion, and unlike for the full sample, this coefficient is statistically significant at the 10 per cent level. The estimated coefficient is −0.059 (0.031).13