The literature review comprises two parts, namely reliability and validity. Both reliability and validity are essential for effective measurement (American Educational Research Association et al. 1999; Cohen et al. 2013; Gregory 2011; Moerdyk 2015). The aim of the literature review was to explain the way reliability and validity are conceptualised and assessed.

To be included in the study, respondents needed to be employed at a large South African organisation, with a workforce of at least 60 employees. Several organisations were approached and 53 companies were eventually willing to participate. In total 60 respondents were randomly selected from each of these organisations. This presented a convenient sample of South African organisations, but a random sample of employees. More detail about the respondents is reported in the findings section.

The sampling of the respondents is discussed above. The data were generated from paper-and-pencil tests, completed in organisations where permission was granted by the appropriate authorities, and all respondents gave consent. These data were not primarily collected for this research and are archival data collected by the author as part of a larger research project. The ethical clearance obtained allows him to collect and use the data in further analysis and to publish academic articles based on the data. This use of the data was clearly stated in the permission letters as well as the consent forms. After cleaning up the data appropriate statistics were calculated. Cleaning up the data was limited to removing of limit data and replacing it with missing values. The statistical techniques used are described in the section ‘Data analysis’.

Eight instruments were administered.

A shortened version of the CEAI (Hornsby et al. 2002) was used. This instrument measures five constructs, namely the level of management support, work discretion and autonomy, rewards and reinforcement, time availability and organisational boundaries (Hornsby et al. 2002). Kuratko, Hornsby and Covin (2014:119) explain what is measured with each factor:

The shortened version proposed and used by Strydom (2013) was applied in this research. Where the original questionnaire consists of 48 items, the shortened version consists of 20 items, 4 items per construct. The items in the shortened version were selected from the 48 items based on their high loading on the particular factor, which represents the subscale. The four items with the highest item load per factor were selected. Substantial work on the factorial validity of the original instrument was done. Hornsby et al. (2002) report the results of an analysis of the five-factor CEAI solution, which showed Cronbach’s alpha of 0.92, 0.86, 0.75, 0.77 and 0.69, for the dimensions as listed above. Kamffer (2004) found similar alphas of 0.88, 0.80, 0.62, 0.71 and 0.77. Strydom (2013), using his shortened version of the CEAI, found alphas of 0.73, 0.82, 0.74, 0.68 and 0.57. The items of the CEAI were presented as statements, such as the following: ‘Individual risk takers are often recognised for their willingness to champion new projects, whether eventually successful or not’. Respondents were asked to respond to the statements by selecting one of five options, namely: strongly agree (5), agree (4), undecided (3), disagree (2) or strongly disagree (1). A high score on any particular factor of the CEAI would be indicative of a climate that is conducive to entrepreneurial activity, and a low score would suggest circumstances that hamper entrepreneurial activity. An overall high score would suggest the presence of a positive entrepreneurial climate.

The Utrecht Work Engagement Scale-9 (UWES-9; Schaufeli & Bakker 2004) is a summative assessment of vigour, dedication and absorption. The UWES is mentioned as the most often used self-report measure of engagement and has been validated in many countries around the world (Bakker et al. 2008). The questionnaire consists of nine items. Schaufeli and Bakker (2004:33) report that the ‘Cronbach’s α of all nine items varies from 0.85 to 0.94 (median = 0.91) across the nine national samples. The α-value for the total data base is 0.90’. With regard to validity, Schaufeli, Bakker and Salanova (2006) claim that the suggested three-factor structure of engagement is confirmed (across samples from different countries) and that the construct is related to other constructs in the expected manner. This suggests construct validity. The questionnaire consists of nine items. The following is a typical item from the scale: ‘At my work, I feel bursting with energy’. Respondents are requested to indicate their views on this statement on a scale ranging from 0 (never) to 6 (every day). The minimum total score is 0 and the maximum 54. A high score on the survey would indicate high levels of engagement and a low score would indicate that the respondents are not engaged.

The Organizational Commitment Scale (Allen & Meyer 1990) has been used to assess organisational commitment. They identify affective, continuance and normative commitment as components of commitment. The full scale consists of 24 questions. Allen and Meyer (1990) report an internal consistency of 0.86, 0.82 and 0.73 for the three subscales. Furthermore, Allen and Meyer (1990:13) report evidence of construct validity and also comment that the ‘relationship between commitment measures … and the antecedent variables … was, for the most part, consistent with prediction’. This points to convergent and discriminant validity. The first item of the scale reads as follows: ‘I would be very happy to spend the rest of my career with this organisation’. Respondents are requested to indicate their views on this statement on a scale ranging from 1 (strongly disagree) to 7 (strongly agree). A high score on the scale indicates high levels of commitment and low scores signify low commitment. For the purpose of this study, only eight items of the Affective Commitment Scale was used.

The Innovative Work Behaviour (IWB) questionnaire was developed by de Jong and den Hartog (2010) to assess the four dimensions they hypothesised to relate to workplace innovation, namely exploration, generation, championing and implementation of ideas. De Jong and den Hartog (2010) state that their analyses demonstrated sufficient reliability and criterion validity. However, they did not find proof of dimensionality in their questionnaire and suggest that it should be used as a one-dimensional construct. The questionnaire was used as presented in the article, with the exception that the stem of the questions was changed from ‘How often does this employee …’ to ‘As an employee how often do you …’ The questionnaire contains 10 items. The first reads as follows: ‘As an employee how often do you pay attention to issues that are not part of your daily work?’ Responses were on a seven-point scale, from 0 (never) to 6 (always). A high score on the scale indicates high levels of innovation in the workplace while low scores indicate low levels of innovation.

Kleysen and Street (2001) hypothesised that individual innovative behaviours (IIBs) consist of five dimensions, namely opportunity exploration, generativity, formative investigation, championing and application. They developed a 14-item questionnaire which assesses these dimensions, called the IIB. As with de Jong and den Hartog (2010), they were unable to confirm their hypothesised dimensionality, but suggest that including a variety of items contributes to a better understanding of the construct. The coefficient alphas for the subscales were 0.791, 0.791, 0.802, 0.893 and 0.796. They report an inter-correlation of 0.945 between the items. As such, they suggest that the items ‘can be combined into a single measure of innovation behaviour … with good construct validity’ (Kleysen & Street 2001:293). The first item of the scale reads as follows: ‘In your current job, how often do you … look for opportunities to improve an existing process, technology, product, service or work relationship?’ Respondents were asked to respond on a six-point scale, ranging from 1 (never) to 6 (always). A high score would then be indicative of high levels of innovative behaviour, whereas a low score would suggest the absence of innovative behaviour.

The Quality of Performance Appraisal Systems Questionnaire (QPASQ) was used to assess the perceived effectiveness of the (traditionally defined) performance appraisal systems in organisations. The QPASQ was developed by Steyn (2010) and is based on human resources management literature, which explains the characteristics of an effective performance appraisal system. Most items were borrowed from Grobler, Wärnick, Carrell, Elbert and Hatfield (2006), who provide a comprehensive list of requirements for an effective performance appraisal system. The items cover the following elements: relevance, reliability, freedom from contamination, discriminability or sensitivity, practicality, acceptedness, labour legislation requirements, specificity, (desired) outcomes, appropriate and contracting. Steyn (2010) reports internal consistency (Cronbach’s alpha) of 0.84 and significant correlations (in the expected direction) with other workplace attitudes. The questionnaire used in this research consisted of 18 items, with the first question reading as follows: ‘The performance appraisal system at my organisation is the primary mechanism used to assess the performance of the employees’. Respondents were requested to indicate their views on this statement on a scale ranging from 1 (Absolutely false – this is true in ±10% of all cases) to 5 (Absolutely true – this is true in ±90% of all cases. A high score would be indicative that a traditionally defined performance appraisal system is in place and effectively functioning while a low score would indicate that the respondents were not of the opinion that a traditionally defined effective performance appraisal system was functioning in their organisation.

The Human Resource Practices Scale (Nyawose 2009) was used to measure the perceived effectiveness of human resource practices, with three questions per practice. Seven HR practices were assessed in this study, namely training and development, compensation and rewards, performance management, supervisor support, staffing, diversity management, as well as communication and information sharing. Nyawose (2009) reports internal consistencies varying from 0.74 to 0.93 for these scales and significant correlations (in the expected direction) with outcomes such as occupational commitment and turnover intentions. Steyn (2012) reports alphas varying between 0.88 and 0.74 and significant correlations (in the expected direction) with outcomes such as job satisfaction, employee engagement, occupational commitment and turnover intentions. The following is the first question from the training and development part of the scale: ‘My company is committed to the training and development needs of its employees’. Respondents were requested to indicate their views on this statement on a scale ranging from 1 (disagree strongly) to 5 (agree strongly). For each individual HR practice, the minimum score would be 3 and the maximum 15. A high score on the survey would be indicative of a belief that HR practices were effective, whereas a low score would indicate that the respondents were not satisfied with the HR practices provided.

The Multifactor Leadership Questionnaire (Avolio, Bass & Jung 1995) was used in the study. The questionnaire measures aspects of transformational leadership (12 items), transactional leadership (6 items), as well as a laissez-faire leadership style (three items). Extensive research on the instrument indicates acceptable reliability as well as validity (Antokonis, Avolio & Sivasubraimania 2003; Avolio, Bass & Jung 1999; Bono & Judge 2004; Muenjohn & Armstrong 2008). Respondents were asked to indicate their levels of agreement with statements such as: ‘My manager makes others feel good to be around him/her’. Respondents were asked to indicate how often this behaviour was present in their managers, where (0) indicates ‘Not at all’, (1) ‘Once in a while’, (2) ‘Sometimes’, (3) ‘Fairly often’ or (4) ‘Frequently, if not always’. A high score on a specific scale would be indicative of a workplace where that type of leadership style is often displayed, while a low score would be indicative of the absence of such leadership.

Demographical information about the sample, as well as descriptive statistics on the instrument of interest, the shortened version of the CEAI, were calculated. The mean, standard error of mean, standard deviation, skewness and kurtosis for the CEAI² are presented. With regard to kurtosis, for a sample of 200, heavier tails (platykurtic shape) are indicated with values below -0.47 and a sharper peak (leptokurtic shape) is indicated with values higher than 0.62 (Doane & Seward 2009). For a sample of 200, the lower limit for skewness (skewed to the left) is -0.281 and the upper limit (skewed to the right) is 0.281. These cut-off scores will be used in making comments with regard to the normality of the distribution.

Next calculations were done concerning the reliability of the instrument. Cronbach’s alpha coefficient and the strict parallel method for calculating reliability, as generated through the SPSS-23 programme, were calculated. As many authors (Clark & Watson 1995; Field 2009; Guilford & Benjamin 1978; Hair et al. 2010; Nunnally & Bernstein 1994) note that coefficients between 0.6 and 0.8 are expectable, the margin in this research was set at 0.70. This norm was applied to all the calculations pertaining to reliability.

Several calculations were done with regard to gathering information on validity. The first related to criterion validity. To test for concurrent criterion-related validity, the correlation between CEAI, as independent variable, and IWB and IIB, as dependent variables, was calculated. A statistical significant correlation (p < 0.01) of a medium size (larger than 0.3, as defined by Cohen 1988) was set as a minimum indicator of concurrent validity.

The rest of the analyses concerned construct validity. Firstly, the homogeneity of the items was tested. The common inter-item correlation, for the entire as well as the five subscales of CEAI, was calculated. Guidelines on the size of an acceptable inter-item correlation vary, but for this research the range was set in line with Clark and Watson’s (1995) guidelines, which are 0.15 < R > 0.50. Should this correlation be too high, then it means the items are too similar. If the correlations were too low on the other hand, they are not related.

Factorial validity was assessed through testing if the different subscales of the CEAI loaded in different factors. This is a simple analysis to test whether the items of the subtest correlate more with the subtest’s own items than with items from another subtest (Nunnally & Bernstein 1994). Before performing this procedure, the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy was performed, as well as Bartlett’s test of sphericity. The standards of acceptability for KMO are above 0.7 (Field 2009; Hair et al. 2010). In the case of the Bartlett’s test the statistics generated should be statistically significant (p < 0.05) (Pallant 2010). Only if these results were acceptable, a principal component analysis was performed. The standard criteria of eigenvalues greater than 1 will be used for factor extraction. Ideally, five components, representing the five subscales of the CEAI, will be identified. Then the Varimax method with Kaiser normalisation was performed and values smaller than 0.4 were suppressed, to make interpretation easier. Ideally, 80% of the items will load on the appropriate factors (subscales).

Tests of known-group differences were conducted next. Firstly, it was tested whether government organisations (including parastatal entities) show less corporate entrepreneurship than private business entities, and secondly if managers show more corporate entrepreneurship than non-managers. One-way analysis of variance was performed in both cases. In the case of the multiple groups, the Scheffe post hoc test was performed to detect which groups differed from each other. Statistical significance of differences (p < 0.05) between groups was seen as sufficient evidence of known-group validity.

Information on convergent, and to a lesser extent divergent validity, was created by calculating the correlation between CEAI and several other measures. Firstly, it was hypothesised that corporate entrepreneurship would correlate significantly with innovative behaviour, more than generic organisational attitudes (employee engagement and organisational commitment). It was hypothesised that CEAI would correlate more with active forms of leadership (transformational and transactional) than with more passive forms of leadership (laissez-faire). Also, it was hypothesised that CEAI is related to performance, and as such, it would correlate more with a measure of effective performance management, rather than with a general measure of human resources management. Lastly, it was hypothesised that CEAI would correlate more with attitudes towards the job (employee engagement) than with attitudes towards the organisation (organisational commitment). A correlation of 0.5 would be seen as a clear sign of convergence, following Gregory’s (2011) example, and a non-significant correlation as evidence of divergent validity, relying on Cohen et al. (2013). However, in this case divergent validity was not the concern, and differences in correlation, as hypothesised, were used as indicative evidence of validity.

Factor analysis would also be performed to test for divergent validity. This was done through forcing CEAI and each of the other measures used in this study into a two-factor solution, and then performing a Varimax method rotation with Kaiser normalisation. After suppressing values smaller than 0.4, to make interpretation easier, the percentage of items that load (correctly) on the appropriate factor was calculated. Should 70% of the items with loadings of 0.4 load ‘correctly’, it would be interpreted as signalling divergent validity. Should this percentage not be achieved, it would point to a lack of divergent validity. The tolerance for items loading ‘incorrectly’, indicative of poor divergent validity, was set at 10%. Thus, should more than 10% of items load ‘incorrectly’, the instrument would be seen as invalid from a factorial point of view.

The total sample consisted of 3180 employees, representative of 53 companies. In total 57.1% reported that they were male employees, compared to 42.5% reporting that they were female employees (missing data = 0.4%). As far as race is concerned, 8.3% marked Asian, 58.4% black people, 8.4% mixed race and 24.6% white people (missing data = 0.3%). Their ages ranged between 20 and 72, with an average of 37.80 (standard deviation = 9.11). As far as tenure at their present company is concerned, it varied between 1 month and 42 years, with an average of 8.39 (standard deviation = 7.47).

As far as functions are concerned, the findings showed that 46.6% indicated that they were involved in the core business of the company, with 52.8% reporting that they fulfil supportive roles (missing data = 0.5%). With regard to position, 36.5% indicated that they hold some kind of managerial position, while 62.9% reported that they did not form part of management (missing data = 0.7%). In Table 1 the post levels of the respondents are presented (missing data = 1.8%).

Concerning formal schooling, 5.0% reported that they received less than 12 years of formal schooling and 25.5% said that they had completed 12 years of formal schooling. A further 40.2% reported that they had completed a degree or diploma, while 28.9% indicated that they had a higher degree or higher diploma (missing data = 0.4%).

The descriptive statistics for the total instrument, as well as for the five subscales, are presented in Table 2. The sample size was 3180. The maximum total score was 98 and the minimum total score was 30 (20 items). For the subscales, the maximum score was 20 and the minimum 4 (4 items).

Almost all subscales were skewed to the left, with the exception of time availability, which was within the boundaries of normality (-0.281 to 0.281). The total score was also skewed to the left, with a value of -0.291. With regard to kurtosis, the subscale time availability had a heavier tail and the subscale organisational boundaries had a sharper peak. However, the total CEAI score of 0.257 fell well within the boundaries of -0.47 to 0.62.

Reliability was reported as per the Cronbach’s alpha coefficient and the strict parallel method, as generated through the SPSS-23 programme. The coefficient for the total instrument was 0.758 (20 items). The unbiased reliability was 0.723 and the common inter-item correlation was 0.115. The coefficients for the subscales are presented in Table 3.

Interesting to note from the above is that common inter-item correlation related positively to reliability for the subscales (each with four items), though this parallel is not found for the total instrument (with 20 items).

The results pertaining to the correlation and CEAI and innovative behaviour are presented in Table 4.

The correlation coefficients presented in Table 4 reveal that the size of the correlations is small, with the highest coefficients just below the threshold of 0.03, which Cohen (1988) set for a medium effect.

Evidence on construct validity is presented under four subheadings.

Researchers and practitioners are urged to use the shortened version of the CEAI, using the 20 instead of 48 items. The shortened version of the CEAI showed acceptable reliability and validity and the use of the central statistics provided in Table 2 of this article is recommended, particularly in the South African or similar contexts. Researchers and practitioners can also now exploit the rich theory and empirical knowledge pertaining to CEAI within the South African context.

Though the sample size was relatively large, and sampling within organisations was done randomly, organisations were selected on the basis of convenience. The generalisation of the results is thus limited. Furthermore, a judgement had to be made regarding the overall validity of the instrument, as not all the indicators of validity were positive. Although subjectivity was uncomfortable, most authors refer to validity assessment as a judgement call. Further research on this is needed to demonstrate evidence of validity. Also, additional statistical techniques, such as structural equation modelling, could be used in future studies. This may provide additional insights about the topic. The research made use of only a single method of data collection, self-reporting. This limitation may be mediated by adding additional methods of reporting, and this is recommended for future studies.