5. Practical section
5.2. Performance of the Cost Objective
5.2.1. Regression Analysis
In total, this thesis will include three regression analysis with Labour Productivity acting as the dependent variable (DV) for years 2010, 2012, and 2014. According to Slack, Brandon-Jones, and Johnston (2013), productivity is the most common measure to determine whether the business is successful in keeping its operation costs down while keeping their required levels of performance in the other objectives. It is measured as the ratio of outputs by inputs. Often, companies use partial measures of input instead, so-called single-factor productivity measures, so they are able to make comparisons amongst different operations withing the same industry.
Labour Productivity is measured as sales (output) divided by number of employees (input), to determine the average performance of the employees. The reason why single-factor productivity measures are better to compare the performance of the cost objective among companies, is because single-factor productivity evades the restraint of using all inputs of the operation. Other inputs, like capital, labour costs and materials could vary, but labour productivity judges the ability of the operation of turning inputs into outputs regardless of the costs of the resources employed as inputs. It is important to mention that, in line with Buleje (2014) and as the main premise of the IT Productivity Paradox, it is expected that shortly after the ERP implementation takes place, the company will experience a decrease in productivity.
Subsequently, after two years (Nicolaou, 2004) the sales are expected to increase and the number of workers to decrease. Thus, leading to an increase in the Labour Productivity ratio.
5.2.2. Variables
The dataset used for this study contains a variety of qualitative and quantitative information about a large number of firms within Europe. Among its quantitative variables, it includes information like that found on financial statements, like sales in thousands of euros per year and operating profit. These measurements are used to calculate the performance ratio Labour productivity. Further information as number of employees is also used in the calculations.
For the dependent variable, this study used adoption of ERP with the values 0 and 1.
Indicating that 0 equals no ERP adoption, and 1 means ERP adoption. This regression is also controlled for other independent variables that might have an effect on the dependent variable.
26
The following is a list detailing the control variables and the reasons behind why those were chosen. Generally, control variables are introduced in order to account for other variables that impact the dependent variable, and thereby purify the result attained for the correlation coefficient of the independent variable (Newbold, 2013).
Pavitt classification. The Pavitt Taxonomy, introduced by the author Keith Pavitt (1984), is used in order to account for differences caused by the sector that the company is in.
It especially applicable to the manufacturing industry and was hence used by EFIGE (Pavitt, 1984). In the case of the taxonomy, four different sectors exist, which were separated into four dummy variables, each taking the value 1 if the company is in the respective sector and 0 if not. The sectors are (1) traditional, (2) High Tech, (3) specialised and (4) economies of scale. Bogliacino and Pianta (2011) indicate differences in labour productivity between the sectors, due to which they are accounted for.
Country. It is expected that the dependent variable is also affected by the country in which the company is located, as Mulder and De Groot (2007) found differences in labour productivity between countries. Bogliacino and Pianta (2016) even state that country differences between firms from the same Pavitt sector exist, making it relevant to account for the country effect next to the Pavitt sectors. This variable is categorical, and it is indicated by the country code (e.g. AUT for Austria). Since the variable is qualitative, dummy coding was used (e.g. If the company is located in Austria then 1, if not, then 0).
Firm size. The number of employees indicates the size of the firm in this thesis. The effect is commonly referred to as the firm-size effect (Chan, Chen and Hsieh, 1985). In the case of this thesis, a positive correlation between firm size and labour productivity is expected, as was found by Miller (1978), argued by economies of scale.
Skilled blue collars. Another control that is accounted for is the percentage of skilled blue-collar workers the company has. As the EFIGE dataset focuses on manufacturing companies, the blue-collar workers are more impactful. Müller and Stegmaier (2017) found a positive correlation between skilled employees and labour productivity. Similar was found by Corvers (1997) specifically for the manufacturing sector. Accordingly, the same is expected here and thereby controlled for.
Unskilled blue collars and apprentices. Relating to the above, the percentage of unskilled blue collars and apprentices is included. The effect expected is negative, as Corvers (1997) found a negative correlation in the manufacturing sector. However, it needs to be pointed out that the two worker types are not exhaustive, but most representative for the sample chosen for the research.
Profit or loss before tax in thousands of euros. Another control variable taken into account in the regression analysis is the profit or loss before tax. Tan (2016) finds that higher labour productivity is associated with higher profitability. Hence, the same assumption is made for this thesis’ regression.
27
Added Value in thousands of euros. Yet another variable that is expected to impact the dependent variable is added value. As added value is made up of the sales price of a product and the cost of creation, and the higher labour productivity is associated with lower costs, a positive correlation is expected. (Amanor-Boadu, 2003)
The following table presents all the dependent and independent variables included in the regression function and the values that they could take.
Type of
Variable Name of variable Values
Dependent Labour Productivity (LP) Natural Logarithm of Labour Productivity
Independent ERP adoption (1): ERP adoption
(0): non-adoption of ERP
Control Country
(AUT): Austria (GER): Germany (FRA): France (SPA): Spain (HUN): Hungary (ITA): Italy
Control Pavitt classification
(1): Traditional (2): High Tech (3): Specialized (4): Economies of Scale
Control Firm size Continuous
Control Skilled blue collars Percentage
Control Unskilled blue collars and apprentices Percentage
Control Profit or loss before tax in thousands of
euros Continuous
Control Added Value in thousands of euros Continuous
Table 3. List of dependent and independent variables used for the multiple regression analysis
28
5.2.3. Sample and descriptive statistics
The sample for the analysis was obtained by filtering the information in order to exclude observations that did not have data on the specific variables. Furthermore, an observation which had a negative value for sales was assumed to be a mistake and excluded from the sample. Moreover, after filtering, there were only four observations from companies located in the UK, which is why they were deleted from the sample as the number was much lower than that of the other countries. Additionally, data was also filtered using the firm size variable, to exclude companies with less than 10 employees. Companies with one to nine employees are considered by the Small Business Administration as microbusinesses (Headd, 2015). Finally, following Buleje (2014), outliers were excluded from the sample by deleting any observation falling outside of three standard deviations, otherwise known as the Z-scores.
The number of observations per each variable are summarized in the following table:
Variable name Number of observations (N)
Labour Productivity 2010 351
Labour Productivity 2012 350
Labour Productivity 2014 351
Table 4. Number of observation in samples per each year
It is important to note that the dataset started with 14,759 firms. Following Field (2013), the sample size was checked in order to determine if it exceeded 15 * number of independent variables (17). The results, 255, indicates that the sample should not have less than this number. All of the number of observations indicated above are higher than 255.
Moreover, the following table indicates the number of firms in the initial sample (353 firms before deleting outliers), that belong to each of the Pavitt classifications.
Pavitt classification Number of observations
(1) Traditional 106
(2) High Tech 161
(3) Specialized 63
(4) Economies of Scale 23
Total 353
Table 5. Number of observations per Pavitt classification
29
The firms from the original sample (353 firms) were also classified depending on the country where they are located.
Country Number of observations
Austria 13
Germany 26
France 47
Spain 164
Hungary 38
Italy 65
Total 353
Table 6. Number of observations per country
Finally, the mean, median, standard deviation, minimum and maximum value for each dependent variable for each year are presented in the following table.
Dependent variable Mean Median Standard Deviation
Minimum Value
Maximum Value Labour Productivity 2010 195.24 138.15 191.06 13.47 1,545.37 Labour Productivity 2012 203.88 137.52 212.07 11.99 2,002.28 Labour Productivity 2014 212.84 143.59 235.06 11.56 2,326.04
Table 7. Descriptive Statistics
5.2.4. Assumptions
In order to perform a regression analysis, some assumptions must be met. The importance of the assumptions relies on the ability to draw conclusions from the regression analysis, and interpretation of the results. Since this bachelor’s thesis includes a multiple variable regression analysis, four main assumptions are required. Following Newbold, Carlson, and Thorne (2012), the following tests were performed in order to fulfil the assumptions.
Multicollinearity: This assumption requires that no multicollinearity exists among the independent variables. The concept means that one independent variable significantly predicts another independent variable. To test for this assumption, the variance inflation factor (VIF) is
30
Table 8. Variance inflation factor calculated for the year 2010
For the interpretation of the results, no multicollinearity is assumed when the VIF number does not exceed 10 (The Pennsylvania State University, n.d.). As it can be observed from the table, none of the values is higher than 10, therefore, the assumption is fulfilled. The same tests were performed for the years 2012 and 1014.
Variable VIF 2012 VIF 2014
Firm size 4.834 4.455
Profit/loss before tax 2.854 3.145
Added Value 6.727 7.336
Economies of Scale 1.273 1.270
Skilled blue collars 1.905 1.914 Unskilled blue collars 2.104 2.137
Table 9. Variance inflation factor calculated for the year 2012 and 2014
31
Similarly to results obtained for 2010, referring VIF for 2012 and 2014, no multicollinearity is assumed since the vales are lower than 10. Therefore, the assumption is fulfilled.
Homoscedasticity: This assumption states that there should be a constant variance of the error term. Homoscedasticity is assumed for statistical analysis; accordingly, the aim is to reject the opposite: Heteroscedasticity. The distribution of the residuals should be random without trends or funnel shapes (Newbold et al., 2013). The following scatterplot of the standardized predicted value and the standardized residual can be used to visually assess the existence of Homoscedasticity. It was constructed using the dependent variable Labour Productivity for the year 2010 before its logarithmic transformation.
Figure 2. Scatterplot for regression standardized residual and regression standardize predicted value for the non-log-transformed dependent variable in the year 2010
It can be observed from the graph that the values are forming clusters. Therefore, in this case Homoscedasticity is not assumed. For this reason, the dependent variable was transformed with natural logarithm. The next chart was constructed using the standardized predicted value and the standardized residuals from the regression model estimating the log-transformed dependent variable, for the same year.
32
Figure 3. Scatterplot for regression standardized residual and regression standardize predicted value after the log-transformation for the dependent variable in the year 2010
It this case, it can be observed that the values are more randomly distributed with similar distances throughout the plot. Accordingly, homoscedasticity is assumed.
Similarly, the variables Labour Productivity for the years 2012 and 2014 were also transformed using logarithms. The respective scatterplots are presented in the figures.
Figure 4. Scatterplot for regression standardized residual and regression standardize predicted value after the log-transformation for the dependent variable in the year 2012
Figure 5. Scatterplot for regression standardized residual and regression standardize predicted value after the log-transformation for the dependent variable in the year 2014
Henceforth, homoscedasticity is assumed for all years.
Autocorrelation: According to Newbold et al. (2012), autocorrelation means that observation’s error terms are correlated. The author further explains that the most common test to measure autocorrelation is the Durbin-Watson statistic. The test is interpreted as follow:
33
- Values of 0 to 2: indicates positive autocorrelation.
- 2: means no autocorrelation
- Values 2 to 4: indicates negative autocorrelation. (Newbold et al., 2012).
The subsequent figure summarizes the output of the Durbin-Watson test before and after log-transformation of Labour Productivity for each year.
Year Durbin-Watson (no log-transformation)
Durbin-Watson (after log-transformation)
2010 2.082 2.004
2012 2.113 2.027
2014 2.041 2.000
Table 10. Durbin-Watson statistic before and after log-transformation for every year As it can be observed from the figures above, the non-transformed variable showed only very low autocorrelation. Nevertheless, the log-transformation significantly improved the results of the test, to even lower to zero autocorrelation. For this reason, using the log-transformed variable, it can be stablished that this assumption is met.
Normal distribution of the error terms: Newbold et al. (2012) states that this test can assess using P-P Plot, or a frequency histogram of standardized residual. The figures were constructed with the values of the year 2010, before the dependent variable was log-transformed.
Figures 5 and 6. Histogram and P-P Plot before the log-transformation for the dependent variable in the year 2010
34
For a normal distribution to exist, the histogram shows a bell-shaped figure. As it can be observed from figure x, the distribution is skewed to the right. Moreover, in the P-P Plot, a normal distribution is indicated by the values closely following the straight line. Nevertheless, the values diverge from the line. This can serve as reiteration as to why use the logarithmic transformation. Below, the same plots are presented with the log-transformed Labour Productivity variable for the year 2010.
Figures 7 and 8. Histogram and P-P Plot after the log-transformation for the dependent variable in the year 2010
As a result from the logarithmic transformation, the histogram shows a rather normal distribution and the values in the P-P Plot closely follow the line.
In conclusion, the dependent variable was log-transformed in order to fulfil the assumptions needed for the multiple variable regression analysis. Accordingly, the models using the log-transformed Labour Productivity were chose to perform the regression analysis.
5.2.5. Statistical Tests and Results
A total of three regressions were done for the years 2010, 2012, and 2014. The purpose was to determine the relationship between adoption of ERP and improvement in performance ratio Labour Productivity. These analyses will serve as a support to enhance the evidence that was gathered to answer the research question. By using dummy coding to distinguish between adopters and non-adopters, it was possible to estimate the impact that ERP = 1 (an increase of one unit on the independent variable X), has on the dependent variable. As compared with ERP = 0 for non-adopters. Furthermore, the regression was done for three different years with the purpose of investigating how, during the four years observed, the performance of the companies changes.
Table 8 presents a summary of the results from the regression analysis for the variable Labour Productivity in year 2010.
35
Coefficients β1 Standard Error P-value
Intercept β0 5.22310 0.21899 0.00000
ERP adoption 0.17105 0.07563 0.02435
Firm size -0.00247 0.00041 0.00000
Profit/loss before tax -0.00003 0.00001 0.00002
Added value 0.00005 0.00001 0.00000
GERMANY 0.41700 0.22005 0.05894
FRANCE 0.02524 0.20789 0.90344
SPAIN -0.26946 0.19867 0.17589
HUNGARY -0.83425 0.22622 0.00026
ITALY 0.26544 0.20283 0.19154
High Tech -0.23419 0.08474 0.00603
Specialized -0.31584 0.10267 0.00227
Economies of Scale -0.62953 0.15462 0.00006
Skilled Blue Collars -0.00421 0.00185 0.02385
Unskilled Blue Collars 0.00215 0.00189 0.25578
Table 11. Results for regression. DV: ln (Labour Productivity) for the year 2010.
Since the dependent variable was transformed using the logarithm function, to interpret the results it is necessary to multiply 100 * β1 obtained for each of the independent variables, to get the percentage of change in the dependent variable (The Wharton School of the University of Pennsylvania, 2001). Hence:
In the year 2010, firms that adopted ERP achieved an estimated increase of 17% in their Labour Productivity if the other independent variables are held constant. For this regression model, an adjusted R2 of 0.441 was achieved. This means that the regression constructed is able to estimate approximately 44.1% of the variation of the values from the dependent variable Labour Productivity.
As can be observed from the descriptive statistics presented in table 7, the firms in the sample made on average 195.24 thousands of euros per employee in 2010. Therefore, if the companies which adopted ERP presented an estimate of 17% increase in their labour productivity figure, this can be translated as the following: Compared to non-adopters, during 2010, ERP-adopting firms made an average of 33.4 thousands of euros more per employee.
In a similar manner, the results obtained for the year 2012 are summarized in the table below.
Coefficients β1 Standard Error P-value
Intercept β0 5.36235 0.21777 0.00000
ERP adoption 0.15652 0.07482 0.03719
Firm size -0.00203 0.00041 0.00000
Profit/loss before tax -0.00003 0.00001 0.00006
36
Added value 0.00004 0.00001 0.00000
GERMANY 0.36581 0.21836 0.09482
FRANCE 0.02007 0.20685 0.92276
SPAIN -0.38349 0.19871 0.05447
HUNGARY -1.02815 0.22874 0.00001
ITALY 0.18734 0.20200 0.35438
High Tech -0.23553 0.08355 0.00510
Specialized -0.33537 0.10154 0.00106
Economies of Scale -0.58986 0.15394 0.00015
Skilled Blue Collars -0.00382 0.00184 0.03851
Unskilled Blue Collars 0.00210 0.00187 0.26421
Table 12. Results for regression. DV: ln (Labour Productivity) for the year 2012.
It can be seen that the difference between year 2010 and year 2012, in terms on performance of the labour productivity ratio, is not large. In the year 2012, ERP-adopting firms experienced, on average, and increase of 15.7 percent in their sales per employee ratio, compared to non-adopting companies. In the same manner as with the figures of 2010, the descriptive statistics table can be used to estimate the change in thousands of euros per employee. 15.7 % of 203.88, equals 32 thousand euros more per employee during 2012, for companies who invested in ERP. Comparing with 2010, the values are very similar.
Furthermore, the regression for the year 2012 yielded a result of 0.442 in the adjusted R2. Henceforth, the regression constructed is able to explain 44.2% of the variation of the values of the dependent variable Labour Productivity. This result is almost exactly the same as 2010’s adjusted R2 (0.441). Considering that a very similar sample and the same statistical model (with the same control variables) were used for the analysis, this is expected.
Lastly, the results for the regression analysis for the year 2014 are presented in the following table.
Coefficients β1 Standard Error P-value
Intercept β0 5.46167 0.22826 0.00000
ERP adoption 0.15015 0.07862 0.05700
Firm size -0.00185 0.00042 0.00001
Profit/loss before tax -0.00003 0.00001 0.00562
Added value 0.00004 0.00001 0.00000
GERMANY 0.32730 0.23035 0.15628
FRANCE -0.02538 0.21977 0.90814
SPAIN -0.36751 0.20981 0.08075
HUNGARY -1.09346 0.24093 0.00001
ITALY 0.11134 0.21314 0.60174
High Tech -0.26087 0.08814 0.00330
Specialized -0.33041 0.10792 0.00238
37
Economies of Scale -0.55829 0.16196 0.00064
Skilled Blue Collars -0.00442 0.00194 0.02317
Unskilled Blue Collars 0.00141 0.00199 0.47920
Table 13. Results for regression. DV: ln (Labour Productivity) for the year 2014.
The results for the regression analysis performed for the year 2014 do not differ a lot from the results obtained from the years 2010 and 2012. On average, for the year 2014, the firms that adopted ERP saw an increase of 15.015 percent in their sales per employee, compared to non-adopting firms. In thousands of euros per employees, this represents a figure of approximately 31.9. It can be observed that the three years in the analysis display similar results.
Furthermore, this last regression is able to explain 41.7% of the variation in the values of Labour Productivity.
38
Table 14. Results summary in percentage and in thousands of euros per year
As it can be observed, ERP had a significant impact on labour productivity. These suggest that the implementation of ERP systems can help improve performance of firms in terms of sales per employee. Similar results were found by the study conducted by Hitt et al.
(2002).
Hitt et al. (2002) interprets from the results found in their analysis that the firms are generating more revenues per unit of input. The results in this thesis confirm the findings stated by the authors. Moreover, it is believed that the improvement seen in this study, for ERP-adopting firms in the dependent variable labour productivity, is caused by a better performance of the cost objective of operations management. As previously explained, the five objectives are interrelated, and improving the performance of each one of them, is usually followed by a better cost performance.
On the other hand, while these findings confirm the conclusions found by some
On the other hand, while these findings confirm the conclusions found by some