Business organisations are constantly looking for ways to gain an advantage over their competitors (Beyleveld & Schurink, 2005; Castaneda & Toulson, 2013). Historically, their focus was on producing as much as possible without considering exact demand (Turner & Chung, 2005). Recently, businesses have embarked upon finding more efficient ways to deal with large turnovers (Umble, Haft & Umble, 2003). One way of achieving this is by employing an Enterprise Resource Planning (ERP) system. An ERP system is a mandatory, integrated, customised, packaged software-based system that handles most of the system requirements in all business operational functions such as finance, human resources, manufacturing, sales and marketing (Wua, Onga & Hsub, 2008). Although expectations from ERP systems are high, these systems have not always led to significant organisational enhancement (Soh, Kien & Tay-Yap, 2000) and most ERP projects turn out to be over budget, not on time and unsuccessful (Abugabah & Sanzogni, 2010; Hong & Kim, 2002; Kumar, Maheshwari & Kumar, 2003).

Previous studies indicate that ERP project failures result because of poor project communication (Somers & Nelson, 2004), lack of senior management support (Al-Mashari, Al-Mudimigh & Zairi, 2003), presence of ethnic differences (Yusuf, Gunasekaran & Abthorpe, 2004), low user acceptance (Amoako-Gyampah & Salam, 2004; Shih & Huang, 2009), ineffective integration of systems (Al-Mashari et al., 2003), user dissatisfaction (Beyleveld & Schurink, 2005; Yu, Li & Gagnonc, 2009) and inadequate system training (Amoako-Gyampah & Salam, 2004). The continued increase in ERP system failures led researchers to seek solutions (Gumussoy, Calisir & Bayram, 2007), resulting in a growing body of research examining the predictors of information technology acceptance and utilisation amongst ERP users (Chau & Hu, 2002; Esteves-Sousa & Pastor-Collado, 2000; Liang, Saraf, Hu & Xue, 2007; Nah & Lau, 2001; Venkatesh, Morris, Davis & Davis, 2003). Adam, Kotzé and Van der Merwe (2011) report that the most important factors within a South African ERP environment are (1) a satisfying user experience, (2) simplicity and ease of use, (3) fast system response, (4) meeting user work output expectation and (5) self-learning.

The interaction between human factors and technological design factors such as system complexity, ergonomic layout and interface responsiveness, both during the pre-implementation and post-implementation phases of ERP systems, are regarded as important focus areas of research and practice for industrial and organisational (I/O) psychologists (Schultz & Schultz, 2014). More specifically, however, Cable and O’Driscoll (2010) found that 39% of I/O psychologists in their research engaged in activities related to human factors and technology, although they only spent between 1% to 9% of their time on these activities. Although this may be attributed to limited work opportunities, a more thorough analysis revealed that it is more likely that the work activities of I/O psychology practitioners are seen as more philosophical about the application of the principles of the behavioural sciences, embracing the humanistic model of organisational functioning. They conclude that I/O psychology practitioners could differentiate themselves from other practitioners ‘by extending their scope of practice into what may be perceived as opportunities for the further application of the behavioural sciences’ (Cable & O’Driscoll, 2010, p. 18). Furthermore, this scope extension should be aimed at full exploitation of the capabilities of complex technologies such as ERP, by deliberately promoting technology acceptance amongst system users (Youngberg, Olsen & Hauser, 2009).

Information technology (IT) acceptance research has yielded many competing models for explaining the relation between user attitudes, perceptions and beliefs and eventual system use (Meade & Islam, 2006; Venkatesh et al., 2003). The most commonly used models include the diffusion of innovations (Rogers, 1995), the theory of reasoned action (Ajzen & Fishbein, 1975), the theory of planned behaviour (Ajzen & Madden, 1986) and the technology acceptance model (Davis, 1986).

From these models, the technology acceptance model, an adaptation of the theory of reasoned action (Koh, Prybutok, Ryan & Wu, 2010), proved to be the most extensively used by information system researchers in their attempts to determine the predictors of system usage behaviour, perhaps because of its meticulousness (Etsebeth, 2012), parsimony and its magnitude of empirical support (Taylor & Todd, 1995). It is a powerful, robust and commonly employed model for predicting and explaining user behaviour towards information system usage (Hong, Thong, Wong & Tam, 2002; Legris, Ingham & Collerette, 2003; Surendran, 2012; Tome, Johnston, Meadows & Nyemba-Mudenda, 2014). Additionally, an increased need has developed for studies that examine and extend the technology acceptance model in complex (Wu & Wang, 2005), mandatory (Lee et al., 2003) IT settings. Therefore, the technology acceptance model (TAM) has been extended to address specific research objectives, namely (2) TAM2, which aims at assisting in the designing workplace interventions to enhance user acceptance of new systems, (2) the unified technology acceptance model (UTAUT), which focuses on measuring the likelihood of new systems, and (3) TAM3, assisting managers and decision-makers to make informed decisions about workplace interventions (Etsebeth, 2012).

The technology acceptance model has been successfully tested, across a wide variety of applications, by several previous studies in countries such as the United States of America, Japan, Switzerland, New Zealand, Hong Kong, Singapore, Malaysia, the Arab world and the United Kingdom. However, from available literature, there is little evidence that the technology acceptance model has been investigated in developing countries and specifically in South Africa (Averweg, 2008). Therefore, the aim of this study was to test the technology acceptance model in a South African manufacturing organisation. This study will contribute to scientific knowledge regarding relations amongst individuals’ perceptions of information system usage, their attitudes, behavioural intentions and actual use of such a system within a South African context.

The technology acceptance model consists of six distinct yet causally related constructs, namely external variables, perceived ease of use, perceived usefulness, attitude towards using, behavioural intention to use and actual system use (Davis, Bagozzi & Warshaw, 1989; Koh et al., 2010). In the technology acceptance model, perceived ease of use and perceived usefulness determine an individual's information systems acceptance (Lee, Kozar & Larsen, 2003; Surendran, 2012) by determining their attitude towards using and subsequent behavioural intention to use, which culminates in actual system use (Wu & Wang, 2005). Perceived usefulness is used as both a dependent and an independent variable since it is predicted by perceived ease of use and in turn predicts attitude towards using and behavioural intention to use simultaneously (Davis et al., 1989; Koh et al., 2010; Lee et al., 2003). The perceived ease of use, attitude towards using and behavioural intention to use components represent the core functions of the technology acceptance model, whereas external variables and actual system use serve merely as input to and output from the model respectively. Figure 1 demonstrates the technology acceptance model as a theoretical framework and its internal construct associations.

The rationale of the technology acceptance model is that the influence of external variables on technology acceptance behaviour is mediated through user beliefs and attitudes, in which beliefs represent a degree of instrumentality tied to action and attitudes are purely affective. Beliefs relate to an individual's subjective assessment that performing some behaviour will result in a specific consequence, whereas attitudes relate to an individual's positive or negative affective feelings about performing the behaviour (Lee et al., 2003).

Both perceived ease of use and perceived usefulness are belief constructs that are indirectly influenced by external variables in reinforcing a user's belief that using a particular system could improve their performance and through their belief that using a particular system (by implication, ERP) will be free of effort (Brown, 2002; Saade & Bahli, 2005; Surendran, 2012). External variables are therefore a bridge between internal beliefs, attitudes and intentions represented in the technology acceptance model and various individual differences, situational restrictions and organisational interventions imposing on behaviour (Guritno & Siringoringo, 2013). Identifying external variables and recognising their impact enables system developers to manipulate these variables and in so doing have better control over user attitude towards using and behavioural intention to use and the subsequent enhanced actual system use (Hong et al., 2002). Previous researchers have identified two main categories of external variables, namely individual differences (e.g. self-efficacy, innovativeness and computer attitudes) and system characteristics (e.g. voluntariness, relative advantage and complexity), as identified by Lee et al. (2003).

Davis, Bagozzi and Warshaw (as cited in Guritno & Siringoringo, 2013) found that although perceived usefulness and perceived ease of use are distinct dimensions, an enhanced perceived ease of use contributes to better performance, thus having a direct influence on perceived usefulness.

Research has further revealed that a relation exists between perceived usefulness and attitude towards using, and between perceived ease of use and attitude towards using new technology (Legris et al., 2003; Gefen & Straub, 2000; Gumussoy et al., 2007; Guritno & Siringoringo, 2013; Moon & Kim, 2001). Attitude towards using involves judgment on whether a behaviour is good or bad and whether the user is in favour of or against performing it (Leonard, Cronan & Kreie, 2004) and has a direct effect on the intention to use IS or ERP systems in the future (Guritno & Siringoringo, 2013; Liao, Shao, Wang & Chen, 1999). According to the technology acceptance model, attitude towards using is jointly determined by perceived ease of use and perceived usefulness (Guritno & Siringoringo, 2013).

The technology acceptance model further proposes that computer usage is ensured by behavioural intention to use, which predicts a user's intention to perform an intentional act such as deciding to accept and use an information system. Behavioural intention to use has also been found to accurately predict the actual use of a computer information system (Guritno & Siringoringo, 2013; Recker, Rosemann, Green & Indulska, 2006; Yu et al., 2009). Behavioural intention to use is influenced by attitude towards using and perceived usefulness (Chau & Hu, 2002; Guritno & Siringoringo, 2013).

Historically, technology acceptance model research has been done, covering traditional and relatively simple yet important environments, such as personal computing, email systems, word processing and spreadsheet software (Hong et al., 2002). However, with the introduction and acceptance of complex information systems such as ERP that cut across functional and organisational boundaries and require business process reengineering during implementation, it was clear that there is an increased need to study and extend the technology acceptance model in complex information systems settings (King & He, 2006). Several researchers have responded to this need and conducted research on the more complex information systems environment of ERP (see Table 1).

Apart from the study done by Seymour, Makanya and Berrangé (2007) within an academic environment using a hybrid version of the technology acceptance model, namely the UTAUT, no South African research study could be traced that used the technology acceptance model as a foundation to examine the mediating influence of shared beliefs in the benefits of the ERP system leading towards end user acceptance and usage within a corporate environment.

The purpose of this study was to determine, within a South African corporate ERP environment, the internal relations between technology acceptance model constructs and their mediating influence towards actual system use. Based on this objective, eight hypotheses were proposed as depicted in Figure 2.

Perceived ease of use is defined as the extent to which a person believes that using a particular system will be free of effort whereas perceived usefulness is defined as the extent to which a person believes that using a particular system will enhance their job performance (Davis, 1989). Previous research by Amoako-Gyampah and Salam (2004) as well as Davis (1989) shows that perceived ease of use is positively related to perceived usefulness. For this study, we hypothesise that if users consider an ERP system as being easy to use they will also find it more useful.

The Technology Acceptance Model Questionnaire (TAMQ) with 23 items was used for gathering data regarding the various constructs depicted in the core of the technology acceptance model, namely perceived usefulness, perceived ease of use, behavioural intention to use and attitude towards using and actual system use; the questionnaire was compiled by Malhotra and Galletta (1999). This scale was adapted from prior studies, many of which had already established reliability and validity. The Cronbach's alpha reliability scores for the core technology acceptance model constructs were found to range from 0.68 to 0.97 for perceived ease of use and from 0.71 to 0.96 for perceived usefulness (Amoako-Gyampah & Salam, 2004; Chau & Hu, 2002; Guritno & Siringoringo, 2013; Jamaluddin, 2013; Malhotra & Galletta, 1999; Surendran, 2012; Taylor & Todd, 1995; Venkatesh et al., 2003), from 0.69 to 0.92 for attitude towards using (Amoako-Gyampah & Salam, 2004; Chau & Hu, 2002; Taylor & Todd, 1995; Venkatesh et al., 2003) and from 0.62 to 0.96 for behavioural intention to use (Amoako-Gyampah & Salam, 2004; Chau & Hu, 2002; Venkatesh et al., 2003).

Malhotra and Galletta (1999) established convergent validity and discriminant validity of all pre-existing technology acceptance model measures. Factor analysis provided evidence of distinct loadings of various factors and convergent and discriminant validity based on inter-item correlations. All constructs used in the technology acceptance model met the criteria of validity and reliability.

The data was statistically analysed by means of the SPSS, version 22, and the AMOS software programs. Descriptive statistics (e.g. means and standard deviations) were used to analyse the data. Pearson correlation coefficients were computed to determine the relations between variables. A cut-off point of 0.05 was set for the statistical significance (p) of the results. Effect sizes (Cohen, 1988) were used to decide on the practical significance of the findings. A cut-off point of 0.30 (medium effect; Cohen, 1988) was set for the practical significance of correlation coefficients.

Structural equation modelling was used to assess the factorial validity of the measuring instruments’ perceived usefulness, perceived ease of use, attitude towards using, behavioural intention to use and actual system use. Amongst the fit indices produced by the AMOS program is the Chi-square statistic (χ²), which is the test of absolute fit of the model. However, the χ² value is sensitive to sample size. Therefore, additional best-fit indices, such as the root means square error of approximation (RMSEA), standardised root mean residual (SRMR), degrees of freedom (df), Tucker-Lewis index (TLI), comparative fit index (CFI), Akaike information criterion (AIC), Bayes information criterion (BIC) and composite reliability (Raykov, 2009), were computed for each scale. Composite reliability ( ρ) is superior to Cronbach's alpha coefficients when latent variable modelling is used.

This study was conducted at a South African steel manufacturer as part of their SAP ERP Consolidation Project, with the intent to determine user acceptance of the new consolidated system. The research sample was drawn from the SAP ERP users group, consisting of system users and technical specialists, over a 16-month period from September 2010 to December 2011. The battery of questionnaires was placed on the organisation's information portal by their systems administrator, so that participants could complete and submit it electronically online. Prior to placing the questions online, some hard copies of questionnaires were distributed and collected by the SAP basis technical department. The system's administrator extracted the completed questionnaire data from the information portal database and made them available to the researchers all data together with the hard copies. This data were verified and collated into a single database for statistical analysis.

Table 4 shows the descriptive statistics, reliability coefficients and inter-item correlation coefficients of the technology acceptance model measuring instruments.

Scores on three of the dimensions, perceived ease of use, attitude towards using and behavioural intention to use, seem to be leptokurtic by peaking high with a kurtosis greater than 1, confirming their frequency distributions to be concentrated around the centre. The remaining two, perceived usefulness and actual system use, seem to be platykurtic by peaking low with a kurtosis of less than 1, with their frequencies distribution concentrated widely around the distribution tail ends. All five dimensions are negatively skewed, which is indicative of most values being distributed to the right of the mean. A multivariate kurtosis score for the data of 280.55 with a critical ratio score of 64.22 was reported by AMOS, which is indicative of non-normality.

Composite reliability obtained using the method as proposed by Raykov (2009) rendered reliability coefficients ranging from 0.96 to 0.99 for the scales used in this study. This score surpassed the 0.70 cut-off point Nunnally and Bernstein (1994) propose and is therefore acceptable.

The correlation coefficients between perceived ease of use, perceived usefulness, attitude towards using, behavioural intention to use and actual system use (ASU) for employees working in a SAP ERP environment are reported in Table 5.

All correlation coefficients obtained were significant. Practically significant positive correlations of medium effect were obtained between perceived ease of use and perceived usefulness (r = 0.48) as well as between perceived ease of use and attitude towards using (r = 0.49), demonstrating higher user perceived usefulness and higher attitude towards using as a result of an increase in user perceived ease of use. Practically significant positive correlations of large effect were obtained between perceived usefulness and attitude towards using (r = 0.75) as well as between perceived usefulness and behavioural intention to use (r = 0.55), confirming that higher levels of perceived usefulness are associated with higher levels of end-user attitude towards using as well as end-user behavioural intention to use the SAP ERP system. Practically significantly positive correlations of medium effect were obtained between attitude towards using and behavioural intention to use (r = 0.48) as well as between behavioural intention to use and actual system use (r = 0.37), demonstrating that a higher attitude towards using positively enhances behavioural intention to use, which in turn leads to higher actual system use.

The structural model (model 8) was tested based on the measurement model (model 7). Model 8, which emerged as an over-identified structural model, having 276 unique covariance terms and 53 parameters, was tested rendering the following fit results: χ² of 562.53 (df = 223; p = 0.00), RMSEA of 0.08, SRMR of 0.06, TLI of 0.91, CFI of 0.92, AIC of 668.53 and BIC of 853.22. Figure 3 shows the standardised path coefficients estimated by AMOS.

Given the cross-sectional nature of the data, two competing models, models 8a and 8b, were also tested to determine possible redundancies amongst path coefficients. In model 8a, the relation between perceived ease of use and perceived usefulness was annulled by constraining the path coefficient to zero. Results indicated a poor fit to the data (χ² (224; N = 241) = 625.56; p < 0.001; TLI = 0.89; CFI = 0.91, RMSEA = 0.09, SRMR = 0.19, AIC = 729.56; BIC = 910.77). A path coefficient in the remaining competing model 8b was also constrained to zero: between perceived usefulness and attitude towards using. Model 8b rendered the following fit statistics: χ² (224; N = 241) of 608.64, TLI of 0.90, CFI of 0.91, RMSEA of 0.09, SRMR of 0.11, AIC of 712.64 and BIC of 893.85. Table 6 presents fit statistics for the various structural models.

Upon inspection of Table 6, model 8 emerged as the best fitting and most parsimonious structural model revealing the lowest χ² (562.53), df (223) and AIC (668.53) values. Furthermore, all other fit statistics, RMSEA (0.08), SRMR (0.06), TLI (0.91) and CFI (0.92, were within acceptable bounds. The following changes in chi-square (Δχ²) were found: model 8 and model 8a had a Δχ² value of 63.03 (Δ df = 1; p < 0.01), whilst model 8 and model 8b showed a Δχ² value of 46.11 (Δ df = 1, p < 0.01). The AIC and BIC values of model 8 were also substantially lower than the values for model 8a and model 8b. Therefore, model 8 will be discussed with reference to the hypotheses of this study by referring to the relevant standardised regression weights as per Figure 3.

The aim of this study was to test the technology acceptance model within a South African SAP ERP user environment. Results acquired through structural equation modelling confirmed that positive inter-construct relation exists between all technology acceptance model constructs. Actual system use, which is the measurement for technology acceptance, was directly affected by behavioural intention to use and indirectly by perceived usefulness. In turn, behavioural intention to use was directly affected by perceived usefulness and attitude towards using and indirectly affected by perceived ease of use. Attitude towards using was directly affected by perceived ease of use and perceived usefulness. Perceived usefulness was directly affected by perceived ease of use.

Structural equation modelling rendered a best fit model that confirmed a positive statistically significant path leading from perceived ease of use through perceived usefulness and behavioural intention to use to actual system use. This path commences with a significant relation between perceived ease of use and perceived usefulness (explaining 26% of the variance). This finding is consistent with findings reported by Amoako-Gyampah and Salam (2004), Averweg (2008), Chau and Hu (2002), Jamaluddin (2013), Koh et al. (2010) and Venkatesh and Davis (2000).

As expected, perceived usefulness of the information system was strongly associated with the behavioural intention to use it (76% of the variance explained). This finding is consistent with that reported by Davis (1989) (as cited in Koh et al., 2010), Malhotra and Galletta (1999) and Taylor and Todd (1995), but inconsistent with findings of Gumussoy et al. (2007), who reported it to be of medium practical effect. This discrepancy may be attributed to the fact that the Gumussoy et al. study focused on potential ERP users and not existing users as in this study. Potential users will obviously not have any experience with ERP. Therefore, they will not have a well-founded sense of usefulness. Amoako-Gyampah and Salam (2004) reported this relation to be non-significant. The Amoako-Gyampah and Salam study was done during an ERP implementation amidst prospective users with no or little system experience. Their study focused primarily on three external project-related variables, namely (1) project communication related to the ERP system, (2) shared belief in the benefits of the ERP system and (3) training on the ERP system. Obviously these participants had no sense of usefulness and were unable to formulate an intention towards using the system.

The behavioural intention to use the system was strongly related to actual system use (38% of the variance shared). This finding is consistent with that reported by Davis (as cited in Koh et al., 2010) and Malhotra and Galletta (1999), but inconsistent with Shih and Huang (2009), who reported it as being of medium effect. Shih and Huang conducted their research prior to an ERP implementation incorporating additional behavioural constructs such as: (1) top management support, (2) computer self-efficacy and (3) computer anxiety. This finding of Shih and Huang could possibly be attributed to a lack of hands-on experience; thus, potential users were unable to formulate a behavioural intention to use the system.

The significance of identifying this relation path to system use is that organisations can focus on appropriate interventions prior to system upgrades and modifications, in order to secure the optimal use of their ERP system amongst existing users. Therefore, during such ERP renewal projects, change management endeavours could focus on influencing the user's perceptions regarding the ease of use together with usefulness of the adapted system. Subsequently, through emphasising the benefits of these adaptations, their intentions to use the modified system can positively be influenced.

The results of this study confirm the important role of the perceived ease of use and perceived usefulness of technology in the acceptance thereof. Perceived ease of use strongly affected perceived usefulness, whilst perceived usefulness strongly affected attitudes towards system use as well as the behavioural intention to use it. The findings confirmed that perceived ease of use indirectly affected individuals’ attitudes towards use (via perceived usefulness). Moreover, perceived ease of use also indirectly affected the behavioural intention to use the system (via perceived usefulness). Perceived usefulness of the information system seems to play a pivotal role in the model: it affected attitudes towards system use, but also the behavioural intention to use the system.

Attitude towards system use did not play an important role in affecting behavioural intentions to use it or the actual use thereof in this study. Although perceived usefulness of the information system strongly affected attitudes towards using it, the relation of attitudes to behavioural intention to use was insignificant. One reason for this finding could be a possible overlap between the constructs of perceived usefulness and attitudes towards use. However, an evaluation of the measurement models showed that these constructs could be modelled independently. This observation corresponds with the model of Gumussoy et al. (2007), but differs from that of Amoako-Gyampah and Salam (2004), who found the strength of the relation between attitude towards using and behavioural intention to use highly significant (r = 0.75). This inconsistency could be attributed to differences in user cultures in South Africa and the United States of America or the period of user exposure to ERP. The South African study was conducted amongst existing ERP users whereas Amoako-Gyampah and Salam focused on potential ERP users prior to an ERP implementation.

This study had several limitations. Firstly, the sample size has a limitation, specifically the distribution of cultural groups, age generations and gender. This limitation could be ratified by a stratified random sampling approach of larger samples and by securing adequate representation of participants in each of these categories. Another limitation was that the measurement of this technology acceptance model variables was based upon self-reports. According to Halbesleben and Buckley (2004), the exclusive use of self-report measures increases the likelihood that at least part of the shared variance between measures can be attributed to method variance. Thirdly, the study population was very homogeneous, since from a sample of 241 respondents, 64.7% were male and 75.5% were white. South Africa's multicultural society demands studying the constructs of technology acceptance from different cultural group perspectives, by ensuring construct equivalence in the absence of item bias for these groups. Stratified random sampling might amend this inadequacy.