279 279 279
THE RELEVANCE OF EMPLOYEE-RELATED 
RATIOS FOR EARLY DETECTION OF 
CORPORATE CRISES
MARIO SITUM
1
   Received: 7 February 2014
Accepted: 27 January 2015
ABSTRACT: The purpose of this study was to analyse whether employee-related ratios de-
rived from accounts have incremental predictive power for the early detection of corporate 
crises and bankruptcies. Based on the literature reviewed, it can be seen that not much 
attention has been drawn to this task, indicating that further research is justified. For 
empirical research purposes, a database of Austrian companies was used for the time pe-
riod 2003 to 2005 in order to develop multivariate linear discriminant functions for the 
classification of companies into the two states; bankrupt and non-bankrupt, and to de-
tect the contribution of employee-related ratios in explaining why firms fail. Several ratios 
from prior research were used as potential predictors. In addition, other separate ratios 
were analysed, including employee-related figures. The results of the study show that while 
employee-related ratios cannot contribute to an improvement in the classification perfor-
mance of prediction models, signs of these ratios within the discriminant functions did 
show the expected directions. Efficient usage of employees seems to play an important role 
in decreasing the probability of insolvency. Additionally, two employee-related ratios were 
found which can be used as proxies for the size of the firm. This had not been identified in 
prior studies for this factor.
Keywords: bankruptcy prediction, crisis indicators, discriminant analysis, ratio analysis
JEL Classification: C12, C38, G33
1. INTRODUCTION
Fast-changing environmental conditions increase the challenges faced by enterprises to 
remain successful in today’s markets. The insolvency statistics for Europe show that after 
the start of financial crisis in 2007/2008, insolvency rates increased and the situation con-
tinued to deteriorate thereafter. Nevertheless, the problem of business failures and the po-
tential for insolvency is still an interesting topic within management science, as the dam-
ages on a macroeconomic level are significant. Therefore, it is both necessary and useful 
to direct research towards the early prediction of corporate crises and financial distress. 
It is generally accepted that prediction models should recognize potential economic and 
1 University of Applied Sciences, Fachhochschule Kufstein Tirol Bildungs GmbH, Kufstein, Austria; e-mail: mario.
situm@fh-kufstein.ac.at
      ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015  |  279-314

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 280
financial difficulties as early as possible. The best timing for the detection of crisis is at the 
strategic crisis stage, which is in practice very difficult to detect, due to the weak signals 
apparent during this stage (Pretorius, 2008, p. 416; Exler & Situm, 2013, p. 162). This type 
of crisis is not fully visible in financial statement figures and other non-financial ratios 
are therefore needed, such as market-based indicators and macroeconomic variables, to 
make them visible and to achieve stronger and more reliable early warning signals. If this 
is possible, sufficient time will then remain to induce turnaround activities, which are less 
costly and far more efficient than in the later stages of crisis.
Even if it is currently recognized that a properly functioning forecasting tool should in-
clude a combination of the aforementioned variables (Grunert, Norden & Weber, 2006; 
Altman, Sabato & Wilson, 2010; Madrid-Guijarro, Garcia-Perez-de-Lema & van Auken, 
2011; Iazzolino, Migliano & Gregorace, 2013), the relevance of figures from financial state-
ments as discriminating variables between failed and non-failed firms remains prominent. 
This paper focuses on this aspect and devotes special attention to accounting ratios related 
to employee data. An extensive review of 238 papers dealing with the search for potential 
variables for the differentiation between the two types of firms revealed that such ratios 
have not received much attention in this field of study.
Therefore, within this research, several ratios related to employees from prior research 
and several new ratios not found to be analysed in the past were tested to assess their 
ability to act as prediction variables. An analysis was made based on a database including 
accounting figures of Austrian firms for the years 2003 and 2004. The aim was to explore 
how solvent and insolvent firms differ in such ratios and whether the performance of 
insolvency prediction models based on multivariate linear discriminant analysis could 
be improved, when these ratios are included with traditional accounting ratios already 
known as prediction variables.
It was found that several employee-related variables showed discriminatory power be-
tween the two types of firms. Nevertheless, this ability could not be exploited to attain im-
proved classification results. Despite this, signs of the variables indicated a direction which 
was consistent with expectations and can be interpreted economically. Summarized, it can 
be said that employee-related variables carry certain information which is relevant for 
discrimination between failed and non-failed firms, but their predictive power is some-
how limited and therefore not sufficient in order to be included into a well-functioning 
forecasting model for bankruptcies.
This paper is organized as follows: First, a literature review is given about the origins of us-
ing accounting ratios for the prediction of insolvencies. This is followed by some insights 
into more recent research, where results concerning the usage of non-financial indica-
tors within bankruptcy prediction are summarized. Additionally, some relevant papers 
are highlighted with reference to their main conclusions, where employee-related ratios 
were investigated concerning predictability. Second, three research hypotheses and three 
research questions are then presented. Third, a description about the database and the 
selection of samples for empirical tests is given. Fourth, a presentation about the ratios 

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 281
applied within this work is provided, where some of them are based on literature review 
and some were not found to be used in previous studies. Fifth, the statistical analyses and 
results are presented, which are necessary to extract the most relevant variables for the 
construction of an insolvency prediction model. Sixth, the equations obtained for mul-
tivariate linear discriminant functions for the different years and combinations of ratios 
including their performance quality are presented. Seventh, the main conclusions of the 
result are summarized. At last, implications and limitations of the study are described, and 
recommendations for future research are given. Within this section the research hypoth-
eses are tested and the research questions are answered.
2. LITERATURE REVIEW
Numerous research papers can be found in the field of business failure prediction which 
extracted a number of differing variables which are suitable for the prediction of busi-
ness failures and bankruptcies. The most prominent variables are accounting ratios, which 
were derived from the financial statements of companies. Early research stated that ac-
counting ratios show the potential to differentiate between bankrupt and non-bankrupt 
firms (Beaver, 1966; Altman, 1968). Several other papers confirmed these findings based 
on empirical results (Altman, Haldeman & Narayanan, 1977; Dambolena & Khoury, 1980; 
Zmijewski, 1984; Casey & Bartczak, 1985; Chalos, 1985; Gombola, Haskins, Ketz & Wil-
liams, 1987; Gilbert, Menon & Schwartz, 1990; Platt, Platt & Pedersen, 1994; McKee, 1995; 
Foster, Ward & Woodroof, 1998; Doumpos & Zopounidis, 1998).
Further research concluded that prediction models using only accounting ratios are in-
ferior to models which combine accounting ratios with other financial and non-financial 
ratios. Financial ratios could comprise data not available from financial statements and 
replicate market data of e.g. stock prices, stock volatilities etc. Non-financial ratios can 
include factors such as management quality or efficiency, which are not directly observ-
able, but can be estimated by appropriate quantitative measures. It is currently generally 
accepted that insolvency prediction models should include a mix of accounting, financial 
and non-financial variables as the accuracy of prediction can be increased in contrast to 
models including only accounting ratios (Thornhill & Amit, 2003; Grunert, Norden & 
Weber, 2006; Altman, Sabato & Wilson, 2010; Madrid-Guijarro, Garcia-Perez-de-Lema & 
van Auken, 2011; Iazzolino, Migliano & Gregorace, 2013).
The study by Thornhill & Amit (2003) found that deficiencies in general and financial man-
agement can be used as variables to explain why younger firms are more likely to fail. For 
older firms, failure is more dependent on external forces. A similar conclusion was made 
by Grunert, Norden & Weber (2006), where the factor of management quality displayed 
statistical significance to such an extent that this non-financial variable contributed to an im-
proved classification. Furthermore, this research found in general that models incorporating 
financial and non-financial factors lead to significantly more accurate default probabilities  
than the single use of either financial or non-financial factors. This result was confirmed by 
the study of Altman, Sabato & Wilson (2010), where the inclusion of non-accounting data to 

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 282
the basic z-score model significantly improved classification performance. This study con-
cluded that needing a longer time to file accounts after the year end is associated with a high-
er probability of difficulties. It was also stated that firms which have an audit qualification are 
more prone to failure, based on the indication that the long-term viability is in some doubt.
The study of Madrid-Guijarro, Garcia-Perez-de-Lema & van Auken (2011) analysed fac-
tors affecting the external and internal environment of the firm and their impact on finan-
cial distress. They showed that high competition among existing firms in the industry and 
high bargaining power of customers increase the probability of distress. A higher tech-
nological level was negatively associated with bankruptcy. The overall conclusion of the 
study was that some strategic variables have a close association with financial distress. Iaz-
zolino, Migliano & Gregorace (2013) investigated the contribution of intellectual capital 
(human capital, structural capital, relational capital) for the purposes of aiding prediction. 
Intellectual capital showed a contribution for credit risk decisions and was useful for the 
classification of defaulted and non-defaulted firms. The general conclusion followed that 
of the aforementioned studies, namely that a scoring model should include financial and 
non-financial information in order to improve prediction accuracy and model quality.
The inclusion of employee-related ratios – these are ratios where certain variables con-
cerning employees and their associated costs (e.g. number of employees, ln(number of 
employees), sales/number of employees, EBIT/employee costs etc.) are considered – was 
not analysed extensively within research. The review of 238 papers related to the task of 
crises and insolvency prediction revealed that only in a few studies were such ratios in-
cluded in the starting base (i.e. a catalogue of potential explanatory variables), but hardly 
any of these were detected as having discriminatory power to divide between failed and 
non-failed firms. In situations of financial distress or crisis, entrepreneurs try to improve 
the company´s results through various measures, and cost cutting seems to be one of the 
most effective measures undertaken in this regard. Firms which recovered from crisis had 
improved their operating performance through cost rationalization, lay-offs, closures and 
the integration of business units (Sudarsanam & Lai, 2001; Pretorius, 2008). Unsuccessful 
enterprises are mostly unable to exploit these opportunities due to different circumstanc-
es. This includes the inability to efficiently use employee resources.
The professional and economic use of staff seems important for corporate success and it 
also depends on management qualities as to how well these aspects are fulfilled. It is pos-
sible for companies to increase EBIT via tight control of labor costs (Kim & Gu, 2006). 
Therefore, ratios associated and related to employee-figures could be seen as a measure-
ment of management efficiency. It is worth attempting to analyse this aspect by using 
employee-related ratios from prior research, but also with some new ratios which have not 
been used as potential explanatory variables in previous studies. Before this, some results 
from papers are presented where employee-related ratios were considered, where differing 
results were obtained concerning their suitability as predictors for insolvencies and crises.
Bruse (1978) conducted one of the first studies which considered employees for the pre-
diction of the potential growth of a company. He explicitly analysed growing and non-

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 283
growing firms in Germany and developed a model that was able to distinguish between 
these two types of firms. The ratios sales/number of employees and staff costs/sales can be 
found within his starting catalogue. Only the second ratio showed the ability to forecast 
corporate growth alongside ratios of liquidity and turnover. Within the work of Gebhardt 
(1980), three employee-related ratios were defined as starting variables. These were staff 
costs/sum of costs, staff costs/revenues and value added/staff costs. None of these varia-
bles displayed statistical significance within univariate and multivariate analyses and were 
therefore not considered to act as predictors to distinguish between failed and non-failed 
firms. 
Wilson, Chong & Peel (1995) analysed the ability of the ratio of directors remuneration/
employee remuneration to act as a discriminatory variable within a logistic model for 
the distinction between failed and distressed acquired firms. The resulting negative sign 
signifies that the more directors earn relative to staff, the more likely it is that a firm can 
be assigned as distressed acquired. No specific explanation was given within their work 
for this occurrence. Within the studies of Lennox (1999a and 1999b), the number of em-
ployees appeared as a relevant variable for discrimination between failed and non-failed 
firms. This ratio can be seen as a proxy for the size of the firm. Small firms and indirectly, 
firms with a low number of employees, are more likely to fail. This aspect confirms results 
from studies before and after Lennox, where the size of the firm played a crucial role for 
discrimination between bankrupt and non-bankrupt firms, even if size was sometimes 
measured by using different variables (Altman, Haldeman & Narayanan, 1977; Ohlson, 
1980; Theodossiou, Kahya, Saidi & Phillipatos, 1996; Dawley, Hoffman & Brockman, 
2003; Bhattarcharjee, Higson, Holly & Kattuman, 2009; Chancharat, Tian, Davy, McCrae 
& Lodh, 2010; Pervan & Visic, 2012; Situm, 2014). 
Within the work of Whitaker (1999), a more complex ratio was constructed for the pre-
diction of a company’s recovery from financial distress. The ratio was defined as number 
of employees/total assets (following year)/number of employees/total assets (pre-distress 
year). A decrease in the number of employees can help firms to recover and can therefore 
provide valuable signals concerning the economic health of a firm. Gudmundsson (2002) 
investigated the potential role of specific variables for the prediction of bankruptcy in 
the airline industry. Three non-financial ratios were included in the analyses: Number 
of pilots per aircraft, number of employees per aircraft and number of hours flown per 
pilot. Only the second variable showed statistical significance with a positive sign. This 
meant that non-distressed airlines exhibited a lower value compared to distressed ones. 
The fewer employees used per aircraft in action, the lower the probability of failure.
Neves & Vieira (2004) found that the ratio percentage of value added for employees and 
value added per employee were explanatory variables for the discrimination between 
bankrupt and non-bankrupt companies. The second variable was one of the most signifi-
cant signals for financial distress. Distressed firms showed much lower values for value 
added compared to non-distressed firms. The ability of this variable to act as a predictor 
was also found within the studies of Nam, Kim, Park & Lee (2008) and Lin, Wang, Wu 
& Chuang (2009). Yim & Mitchell (2007) analysed the ratio of sales/employees within 

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 284
their study to distinguish between failed and non-failed firms in the financial industry. It 
showed no significance and therefore did not appear as a predictor within their forecast-
ing model. Nam, Kim, Park & Lee (2008) used the growth rate of added value/employee as 
a potential variable and recognized that it does not have any discriminatory power. 
Wetter and Wennberg (2009) analysed the effect of human and social capital on a firm’s 
performance and the ability of related measures to assist in the prediction of bankruptcies. 
Their conclusion was that these factors have the discriminatory power to divide between 
successful and unsuccessful firms. Bartual, Garcia, Gimenez & Romero-Civera (2012) 
began their analyses with two employee-related variables: sales/personnel expenses and 
sales/(financial expenses + personnel expenses). Only the second variable was statistically 
significant and therefore suitable to discriminate between failed and non-failed firms. 
Firms exhibiting a higher value of this ratio are more stable and therefore less vulnerable 
to problems. Resistance against crises and bankruptcies can be optimized by an increase 
in sales and the reduction of personnel expenses.
In summary, it can be concluded that the focus in research on employee-related ratios 
within business failure and insolvency prediction is relatively low when compared to the 
numerous studies conducted in this field. Due to this lack of analysis concerning these 
ratios as potential predictors, it is interesting and useful to conduct a separate study where 
some of the existing, but also some new, as-yet unanalysed ratios is investigated, which 
were not considered within prior research. From reviewing the literature, it can be ex-
pected that some variables will show discrimination ability whereas others will not. This 
will be the main task of this paper, but also attention will be given to the contribution of 
each variable concerning the identification of the corporate economic situation.
3. RESEARCH HYPOTHSES AND QUESTIONS
Based on the findings from previous literature, it seems that some financial statement fig-
ures and other employee-related ratios have a certain explanatory power for the event of 
bankruptcy. Firms in financial distress need to implement turnaround activities in order 
to recover. Employees are a cost factor affecting financial statement figures and it is gener-
ally possible from a practical viewpoint to improve different ratios through a reduction of 
the costs associated with employees. Generally, it is expected that firms with an ineffective 
use of employees and high staff costs are more likely to become bankrupt. 
H1: The higher the proportion between staff costs to sales, the higher the probability of 
insolvency.
H2: When employee-related ratios are added to prediction models with “traditional” ac-
counting ratios, then the prediction performance of such models can be improved.
H3: The number of employees and associated ratios with the number of employees are 
potential proxies for the size of the firm.

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 285
The third hypothesis is of relevance due to prior studies, where the variable number of 
employees and ln(number of employees) were found to be proxies for the size of the firm. 
In previous research, attempts to find other proxies for the size of the firm associated with 
the number of employees were not made, with the result that this approach is something 
new in comparison to prior research. Additionally, several research questions shall be 
answered with the empirical data of this study. First, how can employee-related ratios 
contribute to early detection of corporate crises and bankruptcies? Second, which of the 
employee-related ratios are potential predictors for the construction of a business failure 
prediction model? Last, can the inclusion of such factors be helpful to improve the predic-
tion accuracy of an insolvency prediction model?
4. DATABASE
The database for this study consists of Austrian companies divided into the categories bank-
rupt and non-bankrupt. The observation period ranges from the years 2003 to 2004. The 
selected firms were not matched pairwise, as in many other previous studies due to several 
problems with this selection technique. An attempt was made to obtain a sample which is 
representative of the whole population. Thomas, Edelman and Crook (2002) propose such 
an approach. Nevertheless, this procedure also provides problems in terms of statistical es-
timation. If too few bankrupt firms are present in the sample, then their proportion is un-
derestimated and developed models are much better at detecting non-bankrupt firms. First, 
a random initial sample was selected for the observation period. Here, 17 bankrupt firms 
were found from a database for the period 2003 and 2004. Then a random sample of non-
bankrupt firms was chosen for the same period for 170 companies. Therefore, the propor -
tion between non-bankrupt and bankrupt firms is 10:1. Similar proportions had also been 
used within different prior studies (Baetge, Beuter & Feidicker, 1992; Begley, Ming & Watts, 
1996; Lennox, 1999a; Lennox, 1999b; Shah & Murtaza, 2000; Paradi, Asmild & Simak, 2004; 
Hol, 2007; Iazzolino, Migliano & Gregorace, 2013; Chaudhuri, 2013)
Second, a random test sample was obtained in order to assess the performance of the de-
veloped models and their ability to be used for practical purposes. Here, 10 bankrupt and 
100 non-bankrupt firms were chosen randomly. The composition of the firms within the 
different samples is presented in table 1. 
Table 1: Composition of firms in initial and validation samples
2003
[two years prior to 
bankruptcy]
2004
[one year prior to 
bankruptcy]
Bankrupt Non-bankrupt Bankrupt Non-bankrupt
initial sample 17 170 17 170
test sample 10 100 10 100

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 286
5. METHODOLOGY AND RESEARCH DESIGN
In order to test the research hypotheses and research questions, different statistical tests 
and applications were applied within this study. First, descriptive statistics for the bank-
rupt and non-bankrupt companies were computed. Second, a test for normality based on 
Kolmogorov-Smirnov was applied, in order to determine whether the selected ratios were 
normally distributed. Normally distributed data are an important theoretical pre-condi-
tion for the application of multivariate linear discriminant analysis. Third, the differences 
in means, medians and variances for the two groups were analysed, in order to detect 
whether there are differences between the two groups in the variables. This analysis shall 
determine which of the variables are the most effective for discriminating between bank-
rupt and non-bankrupt companies. Fourth, a correlation analysis was computed to recog-
nize how the variables are correlated with each other. This application was complemented 
by a factor analysis, where the loadings of the variables to certain factors were determined. 
Last, multivariate linear discriminant functions were computed which are suitable to divide 
a posteriori between failed and non-failed companies two years and one year prior to the 
event of bankruptcy. In order to test the incremental informational content of employee-
related ratios, three types of functions were computed for this purpose. These are functions 
containing only traditional ratios, only employee-related ratios and a combination of both. 
The validity of the models was then tested with the companies from the test group. The 
quality and accuracy of the models was evaluated using appropriate performance measures.
6. SELECTION OF VARIABLES
The variables for the purpose of this study were selected based on their appearance in 
previous literature. As already stated in this paper, variables related to employees have not 
been extensively analysed in prior studies. Some traditional ratios and some employee-
related were therefore selected. Following accounting variables appeared relatively often 
in previous studies:
• Total Equity/Total Assets
 (Laitinen & Laitinen, 2000; Grunert, Norden & Weber, 2005; Pompe & Bilderbeek, 
2005; Shin, Lee & Kim, 2005; Min & Lee, 2005, Muller, Steyn-Bruwer & Hamman, 
2009; Bartual, Garcia, Gimenez & Romero-Civera, 2012)
• Total Debt/Total Assets
 (Ohlson, 1980; Zmijewski, 1984; Frydman, Altman & Kao, 1985; Pacey & Pham, 1990; 
Bryant, 1997; Doumpos & Zopounidis, 1998; Andandarajan, Lee & Anandarajan, 
2001; Brabazon & Keenan, 2004; Neves & Vieira, 2006; Pervan & Kuvek, 2013; Chaud-
huri, 2013).
• EBIT/Total Assets
 (Altman, 1968; Gilbert, Menon & Schwartz, 1990; Coats & Fant, 1993; Altman & Saun-
ders, 1998; Grunert, Norden & Weber, 2005; Chen, Marshall, Zhang & Ganesh, 2006; 
Li & Sun, 2011; Bartual, Garcia, Gimenez & Romero-Civera, 2012; Iazzolino, Migliano 
& Gregorace, 2013)

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 287
• Ln(Total Assets)
 (Ohlson, 1980; Frydman, Altman & Kao, 1985; Barniv & Raveh, 1989; Whitaker, 1999; 
Chi & Tang, 2006; Pervan & Visic, 2012, Situm, 2014)
• Ln(Sales)
 (Chancharat, Tian, Davy, McCrae & Lodh, 2010; Situm, 2014)
Additionally, the following ratios were included within this study which were derived part-
ly from previous literature. Also displayed are new ratios which have not been analysed 
in this form within prior research. Several of them contain figures related to employees.
Table 2: Additional ratios for analysis
Ratios on the right side are defined as “new” , because these ratios were not found to be 
considered as potential prediction variables based on an extensive literature review of 
238 papers related to the topic of crisis- and insolvency prediction 
Ratios found in previous research “new” ratios not found to be used in 
previous research
Sales/Total Assets  
[Altman, 1968; Brabazon & Keenan, 2004; 
Dietrich, Arcelus & Srinivasan, 2005; 
Bartual, Garcia, Gimenez & Romero-
Civera, 2012; Tsai, 2013]
Ln(Sales/Total Assets)
Ln(Number of Employees) [Situm, 2014] Ln(Sales/Number of Employees)
Staff Costs/Sales  
[Bruse, 1978; Gebhardt, 1980]
Ln(Staff Costs/Number of Employees)
EBIT/Sales  
[Marchesini, Perdue & Bryan, 2005]
EBITDA/Staff Costs
Sales/Staff Costs  
[inverse relation to the ratio staff costs/sales 
based on Bruse, 1978; Gebhardt, 1980]
EBIT/Staff Costs
7. STATISTICAL ANALYSES
7.1 Descriptive statistics
Table 3 provides the means, medians and standard deviations for the chosen variables for 
two years and one year prior to bankruptcy. The mean of total equity/total assets deterio-
rated for bankrupt firms from 2003 to 2004, which indicates that bankrupt firms incur 
additional losses as insolvency approaches. Firms in financial trouble are financing their 
operating business with liabilities, with the result that they are exhibiting much higher lev-
erage ratios in mean and median than solvent firms. Ln(sales), ln(number of employees) 
and ln(total assets) are all measures associated with the size of the firm. All three variables 
showed higher means for the solvent firms in comparison to the bankrupt firms for the 
two observation periods. This indicates that bankrupt firms are in mean, but also in me-
dian, smaller than non-bankrupt ones. Staff costs/sales are much lower for solvent firms 

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 288
in mean, which indicates that employee-resources are used more efficiently by financially 
healthy companies. Higher efficiency of non-bankrupt firms can also be argued by the ra-
tios EBITDA/staff costs, EBIT/staff costs and EBIT/Sales. These variables showed in mean 
and in median higher values for non-bankrupt than for bankrupt firms.
Table 3: Descriptive statistics
2003
[two years prior to bankruptcy]
2004
[one year prior to bankruptcy]
Ratio Class Mean Median Stand.-Dev. Mean Median Stand.-Dev.
Total Equity/
Total Assets
0 -0.062 0.056 0.549 -0.851 -0.330 1.468
1 0.153 0.177 0.510 0.223 0.193 0.295
Total Debt/
Total Assets
0 1.062 0.944 0.549 1.851 1.330 1.468
1 0.848 0.823 0.510 0.777 0.807 0.295
Sales/Total 
Assets
0 2.492 1.694 2.492 1.832 1.238 2.068
1 1.800 1.386 1.527 1.693 1.301 1.485
ln(Sales/Total 
Assets)
0 0.225 0.527 1.777 -0.007 0.214 1.271
1 0.143 0.326 1.171 0.088 0.263 1.102
ln(Sales)
0 11.221 11.235 1.412 11.390 11.549 1.100
1 12.071 11.988 1.243 12.237 12.031 1.205
ln(Number of 
Employees)
0 3.418 3.689 1.175 2.687 2.639 1.309
1 3.834 4.025 1.715 3.749 3.912 1.537
ln(Sales/
Number of 
Employees)
0 11.221 11.235 1.412 11.390 11.549 1.100
1 12.071 11.988 1.243 12.237 12.031 1.205
ln(Staff 
Costs/
Number of 
Employees)
0 10.346 10.409 0.542 10.481 10.627 0.591
1 10.676 10.681 0.726 10.690 10.686 0.615
Staff Costs/
Sales
0 3.736 0.325 13.810 0.840 0.367 1.500
1 0.430 0.274 0.843 0.347 0.262 0.474
Sales/Staff 
Costs
0 4.188 3.076 4.444 4.198 2.728 6.051
1 7.130 3.653 11.442 11.982 3.817 34.833
EBITDA/
Staff Costs
0 -0.152 0.107 1.288 -0.503 -0.230 1.514
1 0.509 0.221 1.571 1.184 0.324 4.072
EBIT/Staff 
Costs
0 -0.251 0.036 1.284 -0.695 -0.312 1.463
1 0.021 0.102 2.070 0.770 0.162 3.354
EBIT/Sales
0 -17.311 0.017 71.301 -1.507 -0.058 4.217
1 -1.254 0.027 15.052 0.043 0.033 0.291

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 289
EBIT/Total 
Assets
0 -0.077 0.024 0.292 -0.531 -0.040 0.994
1 -0.048 0.034 0.909 0.056 0.045 0.163
ln(Total 
Assets)
0 14.415 14.831 1.126 14.084 14.735 1.783
1 15.762 15.743 1.807 15.898 15.735 1.742
7.2 Test for normality of data
The test of normality based on Kolmogorov-Smirnov at the 5 percent level is reported in 
Table 4. Normality of data cannot be assumed for the majority of the variables. The only 
variable which showed normality for both groups and for both time periods is ln(number of 
employees). Similarly, ln(total assets) showed normal distribution for both groups of firms, 
but only two periods prior to the event of insolvency. As within this study multivariate lin-
ear discriminant analysis has been applied, the occurrence of non-normal data could be 
a problem for model building, due to the risk that classification accuracy can be affected 
(Hopwood, McKeown & Mutchler, 1988; Klecka, 1989; Subhash, 1996; Keasey & Watson, 
1991). This is not an extreme problem however, when departures from normality are only at 
a low level (Hopwood, McKewon & Mutchler, 1988; Silva, Stam & Neter, 2002; Feldesman, 
2002). For some constellations of probability distribution, a departure can also be beneficial 
for improved discrimination between both groups, so that better classification accuracies 
can be achieved in comparison to logistic regression (Pohar, Blas & Turk 2004).
Table 4: Kolmogorov-Smirnov test for normality of data
Values in bold denote normally distributed data at 5 percent significance level
2003
[two years prior 
to bankruptcy]
2004
[two years prior  
to bankruptcy]
Ratio Class Statistic Sign. Skewness Statistic Sign. Skewness
Total Equity/
Total Assets
0 .361 .000 -3.344 .263 .003 -1.724
1 .253 .000 -4.526 .123 .000 -0.672
Total Debt/
Total Assets
0 .361 .000 3.344 .263 .003 1.724
1 .253 .000 4.527 .123 .000 0.672
Sales/Total 
Assets
0 .201 .065 1.963 .252 .005 2.010
1 .120 .000 1.834 .142 .000 1.884
ln(Sales/Total 
Assets)
0 .209 .046 -2.486 .142 .200 -0.571
1 .113 .000 -1.654 .116 .000 -1.153
ln(Sales) 0 .201 .067 -1.587 .182 .138 0.275
1 .072 .030 0.579 .093 .001 0.755
ln(Number of 
Employees)
0 .135 .200 -0.969 .119 .200 -0.294
1 .060 .200 0.199 .056 .200 0.075
2003
[two years prior to bankruptcy]
2004
[one year prior to bankruptcy]
Ratio Class Mean Median Stand.-Dev. Mean Median Stand.-Dev.

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 290
ln(Sales/
Number of 
Employees)
0 .201 .067 -1.587 .182 .138 0.275
1 .072 .030 0.579 .093 .001 0.755
ln(Staff Costs/
Number of 
Employees)
0 .137 .200 0.105 .131 .200 -1.156
1 .128 .000 1.373 .101 .000 -0.652
Staff Costs/Sales 0 .519 .000 4.121 .420 .000 3.077
1 .309 .000 8.385 .234 .000 6.871
Sales/Staff Costs 0 .277 .001 2.441 .370 .000 3.752
1 .270 .000 5.465 .368 .000 6.974
EBITDA/Staff 
Costs
0 .365 .000 -3.745 .212 .040 -1.635
1 .234 .000 4.662 .368 .000 6.120
EBIT/Staff 
Costs
0 .394 .000 -3.834 .249 .006 -1.869
1 .313 .000 -5.735 .364 .000 6.536
EBIT/Sales 0 .536 .000 -4.123 .434 .000 -3.350
1 .478 .000 -12.954 .280 .000 -4.890
EBIT/Total 
Assets
0 .268 .002 -2.275 .273 .002 -1.929
1 .370 .000 -11.711 .174 .000 -0.320
ln(Total Assets) 0 .173 .185 -0.271 .172 .196 -0.690
1 .067 .060 0.155 .076 .017 0.657
The problem of non-normal data appeared in several studies (Hauschildt, Rößler & 
Gemünden, 1984; Pacey & Pham, 1990; Barniv & McDonald, 1992; Baetge, Beuter & Fei-
dicker, 1992; Lennox, 1999a; Chi & T ang, 2006; Yim & Mitchell, 2007; Samad, Y usof & Shaha-
rudin, 2009), where this aspect was handled differently. Additionally, the approach of logistic 
regression should be more sound, as it does not demand normally distributed data, but sev-
eral studies showed that this method is not generally able to deliver superior classification re-
sults when compared to multivariate linear discriminant analysis (Gentry, Newbold & Whit-
ford, 1985; Gombola, Haskins, Ketz & Williams, 1987; Barniv & Raveh, 1989; Pacey & Pham, 
1990; Barniv & McDonald, 1992; Neophytou & Mar Molinero, 2004; Kim & Gu, 2006).
The aim of this study is not to develop a forecasting model. This study aims to test the po-
tential prediction power of employee-related ratios in order to differentiate between bank-
rupt and non-bankrupt firms. Despite the non-normality of data being a given, multivari-
ate linear discriminant analysis can nevertheless be used for such an attempt (Feldesman, 
2002; Neophytou & Mar Molinero, 2005; Kim & Gu, 2006). Therefore, further progress was 
conducted without outlier deletion techniques or attempts concerning the normalization of 
data. Nevertheless, it must be kept in mind that this theoretical pre-condition is generally 
violated and that this may be attributable to weaker model quality and classification results. 
This aspect is also discussed within the section covering the limitations of the study.
2003
[two years prior 
to bankruptcy]
2004
[two years prior  
to bankruptcy]
Ratio Class Statistic Sign. Skewness Statistic Sign. Skewness

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 291
7.3 Tests for differences in means and variances
A test for differences in means and in variances at 5 percent level can be applied to detect 
the variables with the highest potential as discriminators between the two groups of com-
panies. Mainly due to non-normally distributed data additionally, a U-test was considered 
(Ho, 2006, p. 357 and 368). In this case it is the more suitable method for decision and 
evaluation, and the results from the two aforementioned methods are displayed for infor-
mational purposes. The results are presented in the Tables 5 and 6. Many more variables 
can be found for the period one year prior to bankruptcy, which indicates that the signal-
ling power increases as the event of bankruptcy approaches. This is in congruence with 
the generally accepted view that early detection is much more difficult (i.e. the signals are 
much weaker or less forthcoming) when the distance in time to the event of bankruptcy 
increases (Altman, 1968; Blum, 1974; Altman, Haldeman & Narayanan, 1977; Dambolena 
& Khoury, 1980; Mensah, 1984; Barniv & McDonald, 1992; Lennox 1999a; Laitinen & 
Laitinen, 2000; Chi & Tang, 2006; Korol & Korodi, 2011). According to the results, certain 
variables remain, which could act as potential predictors for the models.
Table 5: Parametric and non-parametric test for differences two years prior to bankruptcy
Values in bold denote statistically significant differences at the 5 percent level
 
2003
[two years prior to bankruptcy]
  t-test Levene-test U-Test
 Variables T Sign. F Sign. U Sign.
Equity/Total Assets -1.646 0.102 0.118 0.731 914.000 0.013
Total Debt/Total Assets 1.644 0.102 0.119 0.730 914.000 0.013
Sales/Total Assets 1.664 0.098 5.483 0.020 1252.000 0.364
ln(Sales/Total Assets) 0.261 0.794 1.377 0.242 1252.000 0.364
ln(Sales) -2.654 0.009 0.031 0.860 955.000 0.021
ln(Number of 
Employees)
-0.974 0.331 2.806 0.096 1237.500 0.329
ln(Sales/Number of 
Employees)
-2.654 0.009 0.031 0.860 955.000 0.021
ln(Staff Costs/
Number of 
Employees)
-1.817 0.071 0.136 0.713 916.000 0.013
Staff Costs/Sales 3.139 0.002 41.265 0.000 1176.000 0.206
Sales/Staff Costs -1.050 0.295 1.877 0.172 1176.000 0.206
EBITDA/Staff Costs -1.677 0.095 0.161 0.689 1014.000 0.043
EBIT/Staff Costs -0.531 0.596 0.099 0.754 1132.000 0.141
EBIT/Sales -2.482 0.014 25.694 0.000 1086.000 0.092
EBIT/Total Assets -0.130 0.897 0.027 0.870 1200.000 0.250
ln(Total Assets) -3.010 0.003 1.869 0.173 720.000 0.001

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 292
All of the significant variables in 2003 based on U-test (the only exception is ln(staff costs/
number of employees) are also statistically significant at the 5 percent level in 2004. This 
indicates that these ratios are able to provide much earlier warning signals concerning 
the economic situation of the firm. In 2004, four additional ratios showed discriminatory 
power to make a distinction between the two types of firms (ln(number of employees), 
EBIT/staff costs, EBIT/sales and EBIT/total assets). All other ratios are insignificant and 
can therefore be excluded from further analysis. A more profound insight can be achieved 
using correlation and factor analysis.
Table 6: Parametric and non-parametric test for differences one year prior to bankruptcy
Values in bold denote statistically significant differences at the 5 percent level
 
2004
[one year prior to bankruptcy]
  t-test Levene-test U-Test
Variables T Sign. F Sign. U Sign.
Equity/Total Assets -8.194 0.000 119.091 0.000 574.000 0.000
Total Debt/Total 
Assets
8.194 0.000 119.091 0.000 574.000 0.000
Sales/Total Assets 0.353 0.724 1.818 0.179 1377.000 0.749
ln(Sales/Total 
Assets)
-0.334 0.739 0.827 0.364 1377.000 0.749
ln(Sales) -2.784 0.006 0.639 0.425 853.000 0.005
ln(Number of 
Employees)
-2.748 0.007 0.430 0.513 870.500 0.007
ln(Sales/Number of 
Employees)
-2.784 0.006 0.639 0.425 853.000 0.005
ln(Staff Costs/
Number of 
Employees)
-1.340 0.182 0.067 0.796 1202.000 0.253
Staff Costs/Sales 3.065 0.003 23.353 0.000 1051.000 0.064
Sales/Staff Costs -0.918 0.360 1.858 0.175 1051.000 0.064
EBITDA/Staff 
Costs
-1.693 0.092 0.375 0.541 846.000 0.005
EBIT/Staff Costs -1.781 0.077 0.097 0.756 770.000 0.002
EBIT/Sales -4.795 0.000 77.533 0.000 846.000 0.005
EBIT/Total Assets -6.968 0.000 110.293 0.000 734.000 0.001
ln(Total Assets) -4.085 0.000 0.667 0.415 699.000 0.000

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 293
7.4 Correlation analysis and factor analysis
The complete results of correlation analysis based on Pearson for the two years prior to 
the event of bankruptcy can be found in the appendix of this work. Within Tables 7 and 8, 
the correlations for the most relevant variables based on U-test are reported. The general 
results show some highly positive and significant correlations between variables, which 
imposes multicollinearity. This occurrence can affect the discrimination power of models 
when such variables are included within prediction models (Hosmer & Lemeshow, 2000; 
Thomas, Edelman & Crook, 2002; Silva, Stam & Neter, 2002; Wooldridge, 2006; Asteriou 
& Hall, 2007). Multicollinearity can therefore be assumed for the following constellations:
• Ln(sales) and ln(sales/number of employees) for both years
• Total equity/total assets and EBIT/total assets for 2004
• EBITDA/staff costs and EBIT/staff costs for 2004
This implies that not all of these variables should be used for model building, because 
information redundancy is taken as a given. Besides this, there are several negative cor-
relations which are potentially interesting for model building.
Table 7: Correlation analysis for two years prior to bankruptcy 
Values in bold denote statistically significant correlations at the 1 percent level
 
Equity/Total Assets
Total Debt/Total 
Assets
ln(Sales)
ln(Sales/Number of 
Employees
ln(Staff Costs/
Number of 
Employees)
EBITDA/Staff 
Costs
ln(Total Assets)
Equity/Total Assets 1 -1.000 .273 .273 .087 .343 .316
Total Debt/Total 
Assets
1 -.273 -.273 -.087 -.343 -.316
ln(Sales) 1 1.000 .547 .370 .334
ln(Sales/Number of 
Employees)
1 .547 .370 .334
ln(Staff Costs/
Number of 
Employees)
1 .066 .239
EBITDA/Staff Costs 1 .331
ln(Total Assets) 1

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 294
Table 8: Correlation analysis for one year prior to bankruptcy 
Values in bold denote statistically significant correlations at the 1 percent level
 
Equity/Total Assets
Total Debt/Total 
Assets
ln(Sales)
ln(Number of 
Employees)
ln(Sales/Number of 
Employees
EBITDA/Staff Costs
EBIT/Staff Costs
EBIT/Sales
EBIT/Total Assets
ln(Total Assets)
Equity/Total 
Assets
1 -1.000 .194 .237 .194 .080 .103 .488 .775 .360
Total Debt/Total 
Assets
1 -.194 -.237 -.194 -.080 -.103 -.488 -.775 -.360
ln(Sales) 1 -.253 1.000 .285 .251 .238 .198 .321
ln(Number of 
Employees)
1 -.253 -.273 -.242 .076 .125 .633
ln(Sales/Number 
of Employees)
1 .285 .251 .238 .198 .321
EBITDA/Staff 
Costs
1 .952 .172 .231 .085
EBIT/Staff Costs 1 .187 .276 .048
EBIT/Sales 1 .453 .097
EBIT/Total Assets 1 .195
ln(Total Assets) 1
Further insight into the ability of the ratios is provided with factor analysis based on Va-
rimax-rotation. For the period two years prior to bankruptcy, two factors were extracted. 
These factors are able to explain over 70.7 percent of the variances between the related 
variables. Ln(sales) shows a high loading on the first factor, which is the same for the ratio 
ln(sales/number of employees). As the first ratio is associated with the size of the firm, 
this factor could be assigned as a “size factor” . Therefore the second ratio seems to be a 
good proxy for this aspect too. This is also valid for the variable ln(staff costs/number of 
employees). 
The second factor is driven mainly by the variable equity/total assets ratio, which indicates 
that this factor can be assigned as the “capital structure”. Here, total debt/total assets is 
loaded with the same value as the equity-ratio, but with a negative sign, which was expect-
ed based on correlation analysis. The relation between EBITDA/staff costs, which could be 
assigned as a measure of efficiency, is also loaded on this factor, so that here the aspects of 
profitability and efficiency are additionally included. It is interesting to note that ln(total 
assets) is mainly loaded on the second factor, even if it is a measure of size.

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 295
Table 9: Factor analysis for two years prior to bankruptcy
 
Factors
1 2
Cumulated explained variances in % 35.9935564 70.7114292
Equity/Total Assets ,964
Total Debt/Total Assets -,964
ln(Sales) ,924 ,238
ln(Sales/Number of Employees) ,924 ,238
ln(Staff Costs/Number of Employees) ,757 
EBITDA/Staff Costs ,309 ,512
ln(Total Assets) ,374 ,441
The results of factor analysis for one year prior to bankruptcy exhibited four factors. The first 
factor is related to “capital structure and profitability” , where the equity-ratio showed the 
highest loading. The second and fourth factor can both be associated with the “size of the 
firm” . It is interesting to note that ln(total assets) gained a higher value one year prior to the 
event of bankruptcy and showed a much higher loading. Here it is dominant and responsible 
for the formation of the fourth factor. The third factor shows “efficiency” in the usage of staff 
within the firm, where it is measured by different types of earnings. The general conclusion 
of factor analysis is in congruence with the results from correlation analysis in that certain 
variables will not appear within the models due to multicollinearity, even if they showed 
discriminatory power based on the U-test. To answer which variables these will be, several 
multivariate linear discriminant analyses based on a step-wise method were applied.
Table 10: Factor analysis for one year prior to bankruptcy
 
Factors
1 2 3 4
Cumulated explained variances in % 30.3933421 51.9858626 72.080472 88.8199957
Equity/Total Assets .938 .210
Total Debt/Total Assets -.938 -.210
ln(Sales) .134 .978 .132
ln(Number of Employees) .129 -.282 -.196 .872
ln(Sales/Number of Employees) .134 .978 .132
EBITDA/Staff Costs .152 .971
EBIT/Staff Costs .119 .105 .972
EBIT/Sales .659 .156
EBIT/Total Assets .865 .184
ln(Total Assets) .150 .309 .901

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 296
8. MODEL BUILDING
Resulting from the preliminary analyses, it is now possible to construct prediction models 
based on multivariate linear discriminant analysis for the two periods prior to bankruptcy. 
This method is used here to determine the prediction ability of employee-related ratios 
as potential indicators for the distinction between bankrupt and non-bankrupt firms. It 
cannot be expected that the models obtained will provide good classification results nor 
that they will show a good model quality, due to the violations of certain theoretical as-
sumptions for the correct application of linear discriminant analysis. Within this research, 
the technique of Mahlanobis distance is used for the construction of the classification 
functions, where a stepwise method is applied. First, only “traditional” ratios will be used 
for model building, which do not have any association with the number of employees or 
to staff costs. Second, models fed with all variables will be computed so that the incremen-
tal informational content of employee-related ratios for the task of prediction can be as-
sessed. Third, only models containing employee-related variables were calculated in order 
to determine the predictive ability of these ratios. Last, the models were then tested on the 
firms of the validation group, in order to evaluate their potential for practical purposes 
and to assess model stability.
The relevant statistical ratios concerning pre-analysis for discriminant models are shown 
in Table 11. Based on Box-test, the significances are less than 0.05 for the first two model 
types, meaning that equality of variance-co-variance matrices cannot be assumed (Burns 
& Burns, 2008, p. 598). This is a violation of another theoretical pre-condition for the ap-
plication of linear discriminant analysis, which can affect the prediction accuracy of clas-
sification functions (Feldesman, 2002). The only exception is apparent in the model using 
only employee-related ratios for the period 2003 and 2004. Here, this pre-condition was 
fulfilled so that a better model fit could be expected. Based on the Wilks-Lambda model, 
quality of discriminant functions can be assessed. The significances are all below 0.05, so 
that the discriminant functions are significantly discriminating between the two groups 
of firms. A classification with the functions is more reliable than a random assignment of 
the firms into the two groups.
Even if this aspect is taken as a given, based on the value of Wilks-Lambda there remains a 
large proportion of unexplained variances between the two groups of firms. For the model 
containing only “traditional” ratios/all ratios in 2004, the unexplained variance was about 
73.4 percent, which is much lower than for the other models. This result indicates that 
the chosen ratios are not in the position to fully explain or describe how bankrupt and 
non-bankrupt firms differ in this regard. There is the need to consider additional ratios 
and variables, which need not be accounting ratios at all, so that model quality and per-
formance can be improved. This conclusion is in congruence with the generally accepted 
views in research (Grunert, Norden & Weber, 2006; Muller, Steyn-Bruwer & Hamman, 
2009; Altman, Sabato & Wilson, 2010; Madrid-Guijarro, Garcia-Perez-de-Lema & van 
Auken, 2011; Iazzolino, Migliano & Gregorace, 2013).

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 297
Table 11: Pre-analysis for discriminant models
Models with 
“traditional” ratios
Models with all ratios
Models with “employee-
related” ratios
2003 2004 2003 2004 2003 2004
Box-M 5.079 155.604 5.079 155.604 .511 2.335
Sign. .026 .000 .026 .000 .479 .527
Wilks 
Lambda
.953 .734 .953 .734 .963 .894
Sign. .003 .000 .003 .000 .009 .000
The related discriminant functions are shown in the following equations. These are the 
models with the best model fit. The abbreviations are defined as T = traditional model, A 
= all ratios model and E = employee-related model and the numbers in parenthesis denote 
the respective year:
As can be seen above, functions two and four, but also one and three are the same. With 
discriminant analysis, it was not possible to include an additional employee-related ratio 
to total equity/total assets (in 2004) and to ln(total assets) (in 2003) in order to improve 
the prediction accuracy of the model. The classification results for initial and validation 
sample are presented in the Tables 12 and 13. Generally, the functions did not provide 
satisfactory results, which is attributable to the factors of non-normally distributed data, 
inequality of covariance matrices (for the first two model types) and the composition of 
the samples itself, which will be discussed in a later section. Here, it must once again 
be pointed out that these aspects were ignored, as the purpose of this study was not to 
construct a prediction model. The aim is rather to investigate the predictive power of 
employee-related ratios for the distinction between bankrupt and non-bankrupt firms. 
 
 
The related discriminant functions are shown in the following equations. These are the models 
with the best model fit. The abbreviations are defined as T = traditional model, A = all ratios 
model and E = employee-related model and the numbers in parenthesis denote the respective 
year: 
 
𝑍𝑍 𝑇𝑇 ( 2003 )
= −6.5 6 8 + 0.4 3 5 × ln ⁡ ( 𝑇𝑇𝑇𝑇 𝑇𝑇 𝑇𝑇𝑇𝑇 ⁡ 𝐴𝐴 𝐴𝐴𝐴𝐴𝐴𝐴 𝑇𝑇 𝐴𝐴 )
𝑍𝑍 𝑇𝑇 ( 2004 )
= 1.2 7 1 + 4.0 4 4 ×
𝑇𝑇 𝑇𝑇 𝑇𝑇 𝑇𝑇𝑇𝑇 ⁡ 𝐸𝐸 𝐸𝐸 𝐸𝐸 𝐸𝐸 𝑇𝑇 𝐸𝐸
𝑇𝑇 𝑇𝑇 𝑇𝑇 𝑇𝑇𝑇𝑇 ⁡ 𝐴𝐴 𝐴𝐴𝐴𝐴𝐴𝐴 𝑇𝑇 𝐴𝐴 𝑍𝑍 𝐴𝐴 ( 2003 )
= 𝑍𝑍 𝑇𝑇 ( 2003 )
𝑍𝑍 𝐴𝐴 ( 2004 )
= 𝑍𝑍 𝑇𝑇 ( 2004 )
𝑍𝑍 𝐸𝐸 ( 2003 )
= −6.2 4 9 + 0.5 3 7 × 𝑇𝑇 𝑙𝑙 (
𝑆𝑆 𝑇𝑇 𝑇𝑇𝐴𝐴𝐴𝐴
𝑁𝑁 𝐸𝐸 𝑁𝑁 𝑁𝑁 𝐴𝐴 𝑁𝑁 ⁡ 𝑇𝑇 𝑜𝑜 ⁡𝐸𝐸 𝑁𝑁 𝐸𝐸 𝑇𝑇𝑇𝑇 𝐸𝐸 𝐴𝐴𝐴𝐴 𝐴𝐴
)
𝑍𝑍 𝐸𝐸 ( 2004 )
= −1 2 . 1 6 5 + 0. 8 4 9 × 𝑇𝑇𝑙𝑙 (
𝑆𝑆 𝑇𝑇 𝑇𝑇𝐴𝐴𝐴𝐴
𝑁𝑁 𝐸𝐸 𝑁𝑁𝑁𝑁 𝐴𝐴𝑁𝑁 ⁡ 𝑇𝑇𝑜𝑜 ⁡𝐸𝐸 𝑁𝑁 𝐸𝐸 𝑇𝑇𝑇𝑇 𝐸𝐸 𝐴𝐴𝐴𝐴 𝐴𝐴
) + 0 .6 6 4 × 𝑇𝑇𝑙𝑙 ( 𝑁𝑁𝑁𝑁 𝑁𝑁 𝑁𝑁 𝐴𝐴𝑁𝑁 ⁡ 𝑇𝑇𝑜𝑜 ⁡ 𝐸𝐸 𝑁𝑁 𝐸𝐸 𝑇𝑇 𝑇𝑇𝐸𝐸𝐴𝐴 𝐴𝐴 𝐴𝐴 )
 
As can be seen above, functions two and four, but also one and three are the same. With 
discriminant analysis, it was not possible to include an additional employee-related ratio to 
total equity/total assets (in 2004) and to ln(total assets) (in 2003) in order to improve the 
prediction accuracy of the model. The classification results for initial and validation sample 
are presented in the Tables 12 and 13. Generally, the functions did not provide satisfactory 
results, which is attributable to the factors of non-normally distributed data, inequality of 
covariance matrices (for the first two model types) and the composition of the samples itself, 
which will be discussed in a later section. Here, it must once again be pointed out that these 
aspects were ignored, as the purpose of this study is not to construct a prediction model. The 
aim is rather to investigate the predictive power of employee-related ratios for the distinction 
between bankrupt and non-bankrupt firms.  

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 298
Table 12: Classification results of discriminant models for initial sample 
Models with “traditional” ratios and models with all ratios contain the same combina-
tion of prediction variables for the respective year, so that there is no difference in the 
classification results between these two types of discriminant functions
     
2003 
[two years prior to 
bankruptcy]
2004 
[one year prior to 
bankruptcy]
 
bankrupt
non-bankrupt
Assignment Assignment
bankrupt
non-
bankrupt
Models with 
“traditional” 
ratios
Number of 
firms
bankrupt 11 6 9 8
non-
bankrupt 54 116 8 162
% bankrupt 64.71% 35.29% 52.94% 47.06%
non-
bankrupt 31.76% 68.24% 4.71% 95.29%
Models with 
all ratios
Number of 
firms
bankrupt 11 6 9 8
non-
bankrupt 54 116 8 162
% bankrupt 64.71% 35.29% 52.94% 47.06%
non-
bankrupt 31.76% 68.24% 4.71% 95.29%
Models with 
“employee-
related” 
ratios
Number of 
firms
bankrupt 11 6 11 6
non-
bankrupt 66 104 50 120
% bankrupt 64.71% 35.29% 64.71% 35.29%
non-
bankrupt 38.82% 61.18% 29.41% 70.59%
The most relevant measurement for practical purposes is a type I error, which occurs 
when a bankrupt firm is assigned by the model as being non-bankrupt. Therefore, it is of 
interest to minimize this type of error. Models with high type I error are normally not suit-
able for practical purposes. In the case of the models developed, it can be seen that type 
I errors are relatively high, which is implied by the fact that the cut-off value between the 
two types of firms was set at zero. A firm having a lower value than zero was assigned as 
being bankrupt and otherwise it was assigned as non-bankrupt. Literature provides dif-
ferent methods of adjusting this cut-off point in order to minimize type I error, which will 
not be examined further within this work as it is not the purpose of this study. 
Nevertheless, it is of interest to analyse the signs of the ratios within the equations, as they 
can provide some information about the probability of bankruptcy. Within equation one, 

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 299
the signs are consistent with expectations. The size of the firm was found in many studies 
to be a good discriminator between bankrupt and non-bankrupt firms (Ohlson, 1980; 
Theodossiou, Kahya, Saidi & Philippatos, 1996; McKee, 2007; Fitzpatrick & Ogden, 2011; 
Situm, 2014), and it is unsurprising that this variable also appeared as a discriminator 
within this study. Firms which are great in size are much less likely to go into bankruptcy.
The sign of total equity/total assets for equations two and four is also consistent with find-
ings from prior literature (Laitinen & Laitinen, 2000; Grunert, Norden & Weber, 2005; 
Muller, Steyn-Bruwer & Hamman, 2009). Companies with a higher portion of equity are 
less likely to fail. The discriminatory power of this ratio in 2004 is not surprising, based on 
the preliminary U-test and its significance. 
Table 13: Classification results of discriminant models for validation sample 
Models with “traditional” ratios and models with all ratios contain the same combina-
tion of prediction variables for the respective year, so that there is no difference in the 
classification results between these two types of discriminant functions
     
2003 
[two years prior to 
bankruptcy]
2004 
[one year prior to 
bankruptcy]
 
bankrupt
non-bankrupt
Assignment Assignment
bankrupt
non-
bankrupt
bankrupt
non-
bankrupt
Models with 
“traditional” 
ratios
Number 
of firms
bankrupt 4 6 5 5
non-bankrupt 33 67 6 94
% bankrupt 40.00% 60.00% 50.00% 50.00%
non-bankrupt 33.00% 67.00% 6.00% 94.00%
Models with 
all ratios
Number 
of firms
bankrupt 4 6 5 5
non-bankrupt 33 67 6 94
% bankrupt 40.00% 60.00% 50.00% 50.00%
non-bankrupt 33.00% 67.00% 6.00% 94.00%
Models with 
"employee-
related" 
ratios
Number 
of firms
bankrupt 7 3 6 4
non-bankrupt 40 60 29 71
% bankrupt 70.00% 30.00% 60.00% 40.00%
non-bankrupt 40.00% 60.00% 29.00% 71.00%
The fifth equation, ln(sales/number of employees), appeared with a positive sign. This 
ratio indirectly replicates the size of the firm based on factor analysis, so that the direction 
of the sign follows the expectation. A higher value for this ratio improves the score of the 
discriminant functions and tends towards the assignment of a company as non-bankrupt. 

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 300
Therefore, the greater the size of the firm, the less likely it will become bankrupt. This 
result is in congruence with the findings concerning the model comprising the traditional 
ratios. In equation six, ln(sales/number of employees) showed a positive sign. It can be de-
fined as a measure of efficiency. The more sales with fewer employees that can be achieved, 
the more efficiently a firm can be said to be working. This efficiency contributed positively 
to discriminant value. Firms with higher efficiency are less likely to fail. Ln(number of em-
ployees) is a proxy for the size of the firm and therefore has a similar function to ln(total 
assets). Firms with a higher number of employees are greater in size and therefore have a 
lower probability of failure. This aspect is in congruence with results from previous litera-
ture (Situm, 2014). Therefore, the positive sign of this ratio satisfies expectations.
A significantly better comparison of the models can be reached when certain performance 
measures are computed (calculations were based on Metz, 1978; Fawcett, 2006; Anderson, 
2007; Ooghe & Spaenjers, 2010). Based on AUC, it can be seen that the models one year 
prior to bankruptcy have a higher performance when compared to the models two years 
prior to bankruptcy. It is interesting to note that the results for validation sample in 2004 
are quite good, and for the model with “traditional” ratios (and also for the model with all 
variables), they are better than for the initial sample. Nevertheless, type I error remains 
relatively high, but could be adjusted appropriately to achieve better results. The model 
containing only employee-related ratios in 2004 was slightly weaker in performance, but 
showed a better ability to correctly assign bankrupt firms due to lower type I error. The 
overall accuracy of this model in 2004 was relatively weak, which means that the model 
produced high type II errors. It must be emphasized again that the model with all vari-
ables is the same as the model with traditional ratios. Stepwise discriminant analysis was 
not able to include any of the employee-related variables in order to improve prediction 
accuracy. This occurrence leaves doubt about the incremental informational content of 
employee-related ratios for bankruptcy prediction. 
Table 14: Performance measures for the models
Values in bold denote statistically insignificant values
2004 
[one year prior to bankruptcy]
 
Models with 
“traditional “ ratios
Models with all ratios
Models with 
“employee-related” 
ratios
 Initial Validation Initial Validation Initial Validation
AUC 80,14% 83,60% 80,14% 83,60% 79,34% 76,60%
Accuracy 91,44% 90,00% 91,44% 90,00% 70,05% 70,00%
Type I Error 47,06% 50,00% 47,06% 50,00% 35,29% 40,00%
Type II Error 4,71% 6,00% 4,71% 6,00% 29,41% 29,00%

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 301
 
2003 
[two years prior to bankruptcy]
 
Models with 
“traditional “ ratios
Models with all ratios
Models with 
“employee-related” 
ratios
 Initial Validation Initial Validation Initial Validation
AUC 50.59% 43.90% 50.59% 43.90% 54.19% 58.10%
Accuracy 67.91% 64.55% 67.91% 64.55% 61.50% 60.91%
Type I Error 35.29% 60.00% 35.29% 60.00% 35.29% 30.00%
Type II Error 31.76% 33.00% 31.76% 33.00% 38.82% 40.00%
For the period two years prior to bankruptcy, the models delivered a weaker performance 
and in contrast to the models in 2004, the results for validation sample are weak. The 
AUC measures were all relatively low and not statistically significant. Therefore, none of 
the models in 2003 are suitable for practical purposes in the apparent combination. They 
mainly exhibited Gini coefficients below 0.5, and based on this measure their usefulness 
can be dismissed (Anderson, 2007, p. 205). It is assumed that it would also be possible here 
to adjust cut-off values to achieve an improved performance. It is noticeable that accura-
cies in 2003 are much lower than for 2004, which indicates that the informational content 
of the ratios is not strong enough for early detection of corporate crises. This is true for 
all analysed ratios in the course of this study. Such a statement also confirms the results 
from previous literature that the signalling power of predictors is weaker as the distance 
to the event of bankruptcy increases (Altman, 1968; Dambolena & Khoury, 1980; Lennox, 
1999a; Laitinen & Laitinen, 2000; Chi & Tang, 2006; Korol & Korodi, 2011).
9. CONCLUSIONS
The results of this study provide some interesting conclusions concerning the suitability of 
employee-related ratios for the task of prediction. The models developed within this study 
did not show a satisfactory performance, which is attributable to different factors which 
will be highlighted within the next chapter. The signalling power for the period two years 
prior to bankruptcy was limited and the performance of the models was clearly inferior 
to the models developed for the period one year prior to bankruptcy. Such an observa-
tion confirms the results from numerous prior research. It is remarkable that the model 
based on traditional ratios in 2004 provided a better performance in terms of classifying 
the firms from validation sample than it did from initial sample. This provides evidence 
that the model does have a certain discriminatory power and stability which could be ex-
ploited if cut-off values are appropriately adjusted in order to minimize type I error. Even 
if this is possible, it does not seem useful to make this attempt, as the function contains 
only one variable, in that it is a univariate approach. Several problems are associated with 
such a model, which deny a proper practical application.

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 302
The results show that it was not possible to increase the prediction performance of the 
models when employee-related ratios were incorporated within the models. It must be 
mentioned here that it was not possible to build a model containing traditional and em-
ployee-related ratios due to statistical restrictions. Models built solely on employee-related 
ratios are inferior with respect to their predictive power when compared to models using 
traditional ratios and a combination between classical and employee-related ratios. Gen-
erally, their ability to act as predictors seems questionable based on these results, although 
several ratios showed discriminatory power grounded on the tests for differences.
The signs of the independent variables within the discriminant functions are all in con-
gruence with expectations. A closer look at the weightings and interpretation of the dis-
criminators helps to draw some useful conclusions. The size of the firms seems to play an 
important role in the distinction between bankrupt and non-bankrupt firms, because this 
factor was implicitly inherent within the different functions replicated by different vari-
ables. Based on the results from factor analysis, the following proxies were found which 
were all related to the size of the firm. The final two proxies had not been found and at-
tributed to this factor in previous studies until now.
• ln(total assets)
• ln(number of employees)
• ln(staff costs/number of employees)
• ln(sales/number of employees)
From a theoretical viewpoint, it is assumed that firms with a greater size have a great-
er chance of survival due to different factors such as less business risk per dollar of as-
sets invested, easier access to borrowing markets or more tax offsets per dollar of assets 
(Castanias, 1983, p. 1628 – 1629; Theodossiou, Kahya, Saidi & Philippatos, 1996, p. 704). 
These theoretical statements were confirmed within different studies, resulting in firms 
with a greater size having a lower probability of bankruptcy (Ohlson, 1980; Chatterjee 
& Srinivasan, 1992; Theodossiou, Kahya, Saidi & Philippatos, 1996; Dawley, Hoffman & 
Brockman, 2003; Chava & Jarrow, 2004; Chi & Tang, 2006; Hol, 2007; Pervan & Visic, 
2012; Situm, 2014). The same result was found within this study.
The factor of efficiency appeared within this study, but it was not as important a factor as 
expected. Ln(sales/number of employees) appeared as an independent variable within the 
functions containing only employee-related variables for the periods one and two years 
prior to bankruptcy. The respective weightings are relatively low, with the result that the 
contribution of the values to the score of the function is marginal. An argument can be 
made that the data set was not concentrated on a specific industry. Certain industries need 
comparatively much more employees than others to generate sales and results. Such an 
aspect cannot be directly associated with inefficiency to the extent that this can explain 
the small weightings of the related ratios. Despite this, the signs suggest that firms with a 
higher efficiency are less likely to fail. This result confirms the findings of previous studies 
(Chen, Marshall, Zhang & Ganesh, 2006; Y eh, Chi & Hsu, 2010). It is also a sign that firms 
are stable at the operational level, meaning that the pre-conditions for stabilization and 
profit generation are taken as a given (Sudarsanam & Lai, 2001).

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 303
The capital structure of the firm was replicated with the ratio of total equity/total assets, 
which was also a predictor in other studies (Laitinen & Laitinen, 2000; Grunert, Norden 
& Weber, 2005; Pompe & Bilderbeek, 2005; Shin, Lee & Kim, 2005; Min & Lee, 2005, Ba-
hiraie, Akma bt Ibrahim & Azhar, 2009; Muller, Steyn-Bruwer & Hamman, 2009; Bartual, 
Garcia, Gimenez & Romero-Civera, 2012). The interesting finding from this study is that 
it appeared alone within the discrimination functions without any other supporting ratios 
for one year prior to bankruptcy. This is not surprising as this was also visible from the 
results of means and medians from descriptive statistics. This result supports the gener-
ally accepted view that companies with a high (positive) equity ratios are less likely to fail. 
Nevertheless, it is interesting that the ratio did not provide an early warning signal two 
years prior to bankruptcy. This aspect could be attributable to the database however, as the 
differences in means and medians for this ratio in 2003 were relatively low.
The overall conclusion of this work is that employee-related ratios cannot provide ad-
ditional information which could be exploited for the task of prediction. Therefore, their 
inclusion within bankruptcy prediction models is neither recommendable nor beneficial.
10.  IMPLICATIONS, LIMITATIONS AND RECOMMENDATIONS FOR 
FUTURE RESEARCH
10.1 Research hypotheses
The first hypothesis assumed that the higher the ratio of staff costs/sales, the higher the 
probability of insolvency. This statement can be confirmed by the findings of this study. 
The ratio was much higher in mean for bankrupt firms when compared to non-bankrupt 
firms. Nevertheless, the medians were not dispersed so much that the differences were not 
statistically significant based on the results from U-test. Therefore, it was also not consid-
ered within further model building, as it cannot provide incremental explanatory power 
to divide between the two types of firm. Consequently, hypothesis one must be rejected. 
The second hypothesis assumed that a combination of traditional and employee-related 
ratios could improve the classification performance of prediction models. This hypoth-
esis must be rejected based on the performance measures of the derived functions and 
the classification results. Employee-related ratios did not show incremental information 
content which could be exploited for improved prediction accuracy, even if some of them 
showed discriminatory power.
The third hypothesis assumed that the number of employees and related ratios could act 
as proxies for the size of the firm. This hypothesis can be accepted to a significant degree. 
Based on factor analyses, several ratios concerning employee-related figures were loaded 
on size factor. The ratios were ln(staff costs/number of employees) for two years, ln(sales/
number of employees) for both years and ln(number of employees) for one year prior to 
bankruptcy. Therefore, these ratios showed the potential to act as proxies for the size of the 
firm. The first two variables had not been found in previous studies as proxies for the size 
of the firm, which is therefore a new finding from this study.

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 304
10.2 Research questions
The contribution of employee-related ratios for the task of prediction is limited due to the 
results of this study and the chosen variables did not show signalling power, which could 
have been exploited for improved classification accuracy of the models. Employee-related 
ratios are inferior concerning their ability to act as crisis indicators. Therefore, they did 
not contribute to the early detection of bankruptcies. Only some of the employee-related 
ratios were able to act as discriminators between the two types of firms. The respective 
variables mainly characterized the factors of size and efficiency. The size of the firm was 
found to be a relevant predictor within numerous previous studies and the theoretical 
considerations have already been highlighted within this work. Efficiency can be a rel-
evant factor for the distinction between bankrupt and non-bankrupt firms, but the dis-
criminatory power is somewhat limited. This may be attributable to the fact that the study 
included different industries with varying dependencies on the number of employees re-
quired for business success. Nevertheless, the inclusion of such factors was not able to 
improve the prediction quality of the models.
10.3 Limitations and recommendations for future research
The results of this study are based on a database of Austrian companies for the period 2003 
to 2004. As a consequence, certain limitations concerning the generalizability of the re-
sults must be stated. Therefore, additional research for different countries and observation 
periods would be necessary to develop a more profound statement about the potential of 
employee-related ratios to assist in bankruptcy prediction.
The classification performance of the developed models was not satisfying. Here, several 
problems appeared which potentially affected classification accuracy. First, the theoretical 
pre-conditions for application of multivariate linear discriminant analysis were mostly 
violated by the database and the selected ratios. Several distributions showed a strong 
departure from normality, meaning that the proper application of multivariate linear dis-
criminant analysis was theoretically not taken as a given. Additionally, the equalities of 
variance-co-variance matrices based on Box-test were not given, with the result that an-
other important requirement was violated (this point was not true for the models includ-
ing only employee-related ratios). These two aspects together can explain a significant 
proportion of the misclassifications assigned by the discriminant functions and the poor 
model quality. These points are additionally the reason why such a high proportion of 
unexplained variances remained, based on Wilks-Lambda.
Despite this, it was possible to realize the effect of employee-related ratios on the probabil-
ity of insolvency based on the signs within the functions. All signs showed the expected 
direction. Therefore, multivariate linear discriminant analysis can be applied as an ana-
lytical tool to answer specific research questions, even if the theoretical pre-conditions 
for proper application of this method are violated. It remains a valuable instrument for 
researchers in order to analyse further ratios which could potentially be of interest in 

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 305
explaining the differences between bankrupt and non-bankrupt firms. Nevertheless, due 
to the restriction problems, it does not appear to be the correct instrument for the future 
development of a perfectly functioning forecasting tool and the foundation of a theory for 
the explanation of crises and insolvencies.
The proportion between bankrupt and non-bankrupt firms was unequal, with the result 
that different prior probabilities were incorporated within the models. Even if the event of 
insolvency was overrepresented in the initial sample when compared to the real insolven-
cy rate of the whole population, the models understated the event of bankruptcy as they 
showed a much higher ability to detect non-bankrupt firms. This is a general problem of 
sampling for the purpose of business failure prediction, which can only be overcome when 
a much larger sample size is used (Thomas, Edelman & Crook, 2002, p. 122; Anderson, 
2007, p. 350), which was not possible within this study due to restricted data size.
Finally, the models were not adjusted concerning their cut-off values. The threshold for 
classification was set at zero, but it is possible for optimization of model performance to 
adjust this value with appropriate techniques. This means a reduction of type I error with 
an accompanied increase in type II error. Based on the results for performance measures, 
such an adjustment only makes sense for the model in 2004, which includes the tradi-
tional ratios, but makes no sense due to its univariate character. None of the other models 
developed are useful for practical purposes and are not suitable as crisis indicators. 

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 306
APPENDIX
Table 15: Correlation analysis for the year 2003 (two years prior to the event of bankruptcy)
*)  statistically significant correlations at the 5 percent level 
**)  statistically significant correlations at the 1 percent level
 
EquityTotalAssets
TotalDebt/TotalAssets
Sales/TotalAssets
ln(SalesTotal/Assets)
ln(Sales)
ln(Number of Employees)
ln(Sales/Number of Employees
ln(Staff Costs/Number of 
Employees)
Staff Costs/Sales
Sales/Staff Costs
EBITDA/Staff Costs
EBIT/Staff Costs
EBIT/Sales
EBIT/TotalAssets
ln(TotalAssets)
Total 
EquityTotalAssets
1 -1.000** .026 .113 .273** .214** .273** .087 -.339** .064 .343** .342** .340** .160* .316**
Total Debt/
TotalAssets
1 -.026 -.113 -.273** -.214** -.273** -.087 .339** -.064 -.343** -.343** -.340** -.160* -.316**
Sales/TotalAssets 1 .747** .225** .066 .225** -.122 -.119 .249** -.103 .080 .119 .060 -.291**
ln(SalesTotal/
Assets)
1 .287** .242** .287** -.105 -.419** .185* -.154* .327** .495** .071 -.256**
ln(Sales) 1 -.194** 1.000** .547** -.375** .604** .370** .324** .362** .239** .334**

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 307
ln(Number of 
Employees)
1 -.194** -.237** -.120 -.201** -.041 .124 .134 .165* .629**
ln(Sales/Number 
of Employees)
1 .547** -.375** .604** .370** .324** .362** .239** .334**
ln(Staff Costs/
Number of 
Employees)
1 .037 .024 .066 .015 -.036 .010 .239**
Staff Costs/Sales 1 -.086 -.304** -.320** -.906** -.163* -.091
Sales/Staff Costs 1 .122 .047 .066 .063 .115
EBITDA/Staff 
Costs
1 .586** .302** .223** .331**
EBIT/Staff Costs 1 .588** .217** .122
EBIT/Sales 1 .120 .043
EBIT/TotalAssets 1 .275**
ln(TotalAssets) 1
 
EquityTotalAssets
TotalDebt/TotalAssets
Sales/TotalAssets
ln(SalesTotal/Assets)
ln(Sales)
ln(Number of Employees)
ln(Sales/Number of Employees
ln(Staff Costs/Number of 
Employees)
Staff Costs/Sales
Sales/Staff Costs
EBITDA/Staff Costs
EBIT/Staff Costs
EBIT/Sales
EBIT/TotalAssets
ln(TotalAssets)

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 308
Table 16: Correlation analysis for the year 2004 (one year prior to the event of bankruptcy)
*)  statistically significant correlations at the 5 percent level 
**)  statistically significant correlations at the 1 percent level
 
EquityTotalAssets
TotalDebt/TotalAssets
Sales/TotalAssets
ln(SalesTotal/Assets)
ln(Sales)
ln(Number of Employees)
ln(Sales/Number of Employees
ln(Staff Costs/Number of 
Employees)
Staff Costs/Sales
Sales/Staff Costs
EBITDA/Staff Costs
EBIT/Staff Costs
EBIT/Sales
EBIT/TotalAssets
ln(TotalAssets)
Total 
EquityTotalAssets
1 -1,000** -,089 -,046 ,194** ,237** ,194** ,179* -,266** -,014 ,080 ,103 ,488** ,775** ,360**
Total Debt/
TotalAssets
1 ,089 ,046 -,194** -,237** -,194** -,179* ,266** ,014 -,080 -,103 -,488** -,775** -,360**
Sales/TotalAssets 1 ,802** ,334** -,104 ,334** ,050 -,181* ,307** -,130 -,083 ,082 ,040 -,356**
ln(SalesTotal/Assets) 1 ,218** ,078 ,218** -,074 -,347** ,180* -,206** -,138 ,207** ,072 -,402**
ln(Sales) 1 -,253** 1,000** ,402** -,505** ,482** ,285** ,251** ,238** ,198** ,321**
ln(Number of 
Employees)
1 -,253** ,077 ,035 -,253** -,273** -,242** ,076 ,125 ,633**

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 309
ln(Sales/Number of 
Employees)
1 ,402** -,505** ,482** ,285** ,251** ,238** ,198** ,321**
ln(Staff Costs/
Number of 
Employees)
1 ,126 -,251** -,166* -,192** -,098 ,004 ,380**
Staff Costs/Sales 1 -,160* -,173** -,166* -,783** -,310** -,096
Sales/Staff Costs 1 ,377** ,409** ,039 ,053 -,003
EBITDA/Staff Costs 1 ,952** ,172* ,231** ,085
EBIT/Staff Costs 1 ,187* ,276** ,048
EBIT/Sales 1 ,453** ,097
EBIT/TotalAssets 1 ,195**
ln(TotalAssets) 1
 
EquityTotalAssets
TotalDebt/TotalAssets
Sales/TotalAssets
ln(SalesTotal/Assets)
ln(Sales)
ln(Number of Employees)
ln(Sales/Number of Employees
ln(Staff Costs/Number of 
Employees)
Staff Costs/Sales
Sales/Staff Costs
EBITDA/Staff Costs
EBIT/Staff Costs
EBIT/Sales
EBIT/TotalAssets
ln(TotalAssets)

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 310
REFERENCES
Altman, E. I., Sabato, G. & Wilson, N. (2010). The value of non-financial information in small and medium-sized 
enterprise risk management. The Journal of Credit Risk, 6(2), 1–33.
Altman, E. I. & Saunders, A. (1998). Credit risk measurement: Development over the last 20 years. Journal of 
Banking & Finance, 21, 1721–1742.
Altman, E. I., Haldeman, R. G. & Narayanan, P . (1977). ZETA
TM 
analysis: A new model to identify bankruptcy 
risk of corporations. Journal of Banking and Finance, 1, 29–54.
Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. The 
Journal of Finance, 23(4), 589–609.
Anandarajan, M., Lee, P. & Anandarajan, A. (2001). Bankruptcy prediction of financially stressed firms: An 
examination of the predictive accuracy of artificial neural networks. International Journal of Intelligent Systems 
in Accounting, Finance & Management, 10, 69–81.
Anderson, R. (2007). The credit scoring toolkit: Theory and practice for retail credit risk management and decision 
automation. Oxford: Oxford University Press.
Asteriou, D. & Hall, S. G. (2007). Applied econometrics: A modern approach. New Y ork: Palgrave.
Baetge, J., Beuter, H. & Feidicker, M. (1992). Kreditwürdigkeitsprüfung mit Diskriminanzanalyse. Die Wirtschaft-
sprüfung, 24, 749–761.
Bahiraie, A., Akma bt Ibrahim, N. & Azhar, A. K. M. (2009). On the predictability of risk box approach by ge-
netic programming method for bankruptcy prediction. American Journal of Applied Sciences, 6(9), 1748–1757.
Barniv, R. & McDonald, J. B. (1992). Identifying financial distress in the insurance industry: A synthesis of meth-
odological and empirical issues. The Journal of Risk and Insurance, 59(4), 543–574.
Barniv, R. & Raveh, A. (1989). Identifying financial distress: A new nonparametric approach. Journal of Business 
Finance & Accounting, 16(3), 361–383.
Bartual, C., Garcia, F ., Gimenez, V . & Romero-Civera, A. (2012). Credit risk analysis: Reflection on the use of the 
logit model. Journal of Applied Finance & Banking, 2(6), 1–13.
Beaver, W . H. (1966). Financial ratios as predictors of failure. Empirical Research in Accounting: Selected Studies, 
4, 71–111.
Begley, J., Ming, J. & Watts, S. (1996). Bankruptcy classification errors in the 1980s: An empirical analysis of 
Altman´s and Ohlson´s models. Review of Accounting Studies, 1, 267–284.
Bhattacharjee, A., Higson, C., Holly, S. & Kattuman, P. (2009). Macroeconomic instability and business exit: 
Determinants of failures and acquisitions of UK firms. Economica, 76, 108–131
Blum, M. (1974). Failing company discriminant analysis. Journal of Accounting Research, 12(1), 1–25.
Brabazon, A. & Keenan, P . B. (2004). A hybrid genetic model for the prediction of corporate failure. Computa-
tional Management Science, 1, 293–310.
Bruse, H. (1978). Die Prognosefähigkeit von Kennzahlen bei verschiedenen Maßen für das Unternehmenswach-
stum. Zeitschrift für Betriebswirtschaft, 48, 138–152.
Bryant, S. M. (1997). A case-based reasoning approach to bankruptcy prediction modeling. Intelligent Systems in 
Accounting, Finance and Management, 6, 195–214.

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 311
Burns, R. B. & Burns, R. A. (2008). Business research methods and statistics using SPSAS. London: Sage.
Casey, C. & Bartczak, N. (1985). Using operating cash flow data to predict financial distress: Some extensions. 
Journal of Accounting Research, 23(1), 384–401.
Castanias, R. (1983). Bankruptcy risk and optimal capital structure. The Journal of Finance, 38(5), 1617–1635.
Chalos, P. (1985). Financial distress: A comparative study of individual, model, and committee assessments. 
Journal of Accounting Research, 23(2), 527–543.
Chancharat, N., Tian, G., Davy, P ., McCrae, M. & Lodh, S. (2010). Multiple states of financially distressed com-
panies: Tests using a competing risk-model. Australasian Accounting Business and Finance Journal, 4(4), 27–44.
Chatterjee, S. & Srinivasan, V . (1992). Graphical analysis and financial classification: A case study. Managerial 
and Decision Economics, 13, 527–537.
Chaudhuri, A. (2013). Bankruptcy prediction using Bayesian, hazard, mixed logit and rough Bayesian models: A 
comparative analysis. Computer and Information Science, 6(2), 103–125.
Chava, S. & Jarrow, R. A. (2004). Bankruptcy prediction with industry effects. Review of Finance, 8, 537–569.
Chen, J., Marshall, B. R. Zhang, J., & Ganesh, S. (2006). Financial distress prediction in China. Review of Pacific 
Basin Financial Markets and Policies, 9(2), 317–336.
Chi, L.-C. & Tang, T.-C. (2006). Bankruptcy prediction: Application of logit analysis in export credit risks. Aus-
tralian Journal of Management, 31(1), 17–27.
Coats, P . K. & Fant, F. L. (1993). Recognizing financial distress patterns using a neural network tool. Financial 
Management, 22, 142–155.
Dambolena, I. G. & Khoury, S. J. (1980). Ratio stability and corporate failure. The Journal of Finance, 35(4), 
1017–1026.
Dawley, D. D., Hoffman, J. J. & Brockman, E. N. (2003). Do size and diversification type matter? An examination 
of post-bankruptcy outcomes. Journal of Managerial Issues, 15(4), 413–439.
Dietrich, J., Arcelus, F. J. & Srinivasan, G. (2005). Predicting financial failure: Some evidence from new Brun-
swick agricultural co-ops. Annals of Public and Cooperative Economics, 76(2), 179–194.
Doumpos, M. & Zopounidis, C. (1998). A multicriteria discrimination method for the prediction of financial 
distress: The case of Greece. Multinational Finance Journal, 3(2), 71–101.
Exler, M. W. & Situm, M (2013). Früherkennung von Unternehmenskrisen: Systematische Einteilung von 
Krisenfrüherkennungsindikatoren zu den unterschiedlichen Krisenphasen eines Unternehmens. Krisen-, Sani-
erungs- und Insolvenzberatung, 9(4), 161–166.
Fawcett, T. (2006). An introduction to ROC analysis. Pattern Recognition Letters, 27, 861–874.
Feldesman, M. R. (2002). Classification trees as an alternative to linear discriminant analysis. American Journal 
of Physical Anthropology, 119, 257–275.
Fitzpatrick, J. & Ogden, J. P. (2011). The detection and dynamics of financial distress. International Review of 
Finance, 11(1), 87–121.
Foster, B. P ., Ward, T. J. & Woodroof, J. (1998). An analysis of the usefulness of debt defaults and going concern 
opinions in bankruptcy risk assessment. Journal of Accounting, Auditing & Finance, 13(3), 351–371.

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 312
Frydman, H., Altman, E. I. & Kao, D.-L. (1985). Introducing recursive partitioning for financial classification: 
The case of financial distress. The Journal of Finance, 40(1), 269–291.
Gebhardt, G. (1980). Insolvenzfrüherkennung aus aktienrechtlichen Jahresabschlüssen. Wiesbaden: Gabler.
Gilbert, L. R., Menon, K. & Schwartz, K. B. (1990). Predicting bankruptcy for firms in financial distress. Journal 
of Business Finance & Accounting, 17(1), 161–171.
Gombola, M. J., Haskins, M. E., Ketz, E. J. & Williams, D. D. (1987). Cash flow in bankruptcy prediction. Finan-
cial Management, 16, 55–65.
Grunert, J., Norden, L. & Weber, M. (2005). The role of non-financial factors in internal credit ratings. Journal 
of Banking & Finance, 29, 509–531.
Gudmundsson, S. V . (2002). Airline distress prediction using non-financial indicators. Journal of Air Transporta-
tion, 7(2), 3–24.
Hauschildt, J., Rösler, J. & Gemünden, H. G. (1984). Der Cash Flow – Ein Krisensignalwert? Die Betrieb-
swirtschaft, 44(3), 353–370.
Ho, R. (2006), Handbook of univariate and multivariate data analysis and interpretation with SPSS. Boca Raton: 
Chapman & Hall.
Hosmer, D. W . & Lemeshow, S. (2000). Applied logistic regression (2nd ed.). Massachusetts: John Wiley..
Hol, S. (2007). The influence of the business cycle on bankruptcy probability. International Transactions in Op-
erational Research, 14, 75–90.
Hopwood, W ., McKeown, J. & Mutchler, J. (1988). The sensitivity of financial distress prediction models to de-
partures from normality. Contemporary Accounting Research, 5(1), 284–298.
Keasey, K. & W atson, R. (1991). Financial distress prediction models: A review of their usefulness. British Journal 
of Management, 2, 89 – 102.
Klecka, W . R. (1980). Discriminant analysis. Newbury Park: Sage. 
Iazzolino, G., Migliano, G. & Gregorace, E. (2013). Evaluating intellectual capital for supporting credit risk as-
sessment: An empirical study. Investment Management and Financial Innovations, 10(2), 44–54.
Kim, H. & Gu, Z. (2006). Predicting restaurant bankruptcy: A logit model in comparison with a discriminant 
model. Journal of Hospitality & Tourism Research, 30(4), 474–493.
Korol, T. & Korodi, A. (2011).  An evaluation of effectiveness of fuzzy logic model in predicting the business 
bankruptcy. Romanian Journal of Economic Forecasting, 3, 92–107.
Laitinen, E. K. & Laitinen, T. (2000). Bankruptcy prediction: Application of the Taylor´s expansion in logistic 
regression. International Review of Financial Analysis, 9, 327–349.
Lennox, C. (1999a). Identifying failing companies: A re-evaluation of the logit, probit and DA approaches. Jour-
nal of Economics and Business, 51, 347–364.
Lennox, C. S. (1999b). The accuracy and incremental information content of audit reports in predicting bank-
ruptcy. Journal of Business Finance & Accounting, 26(5/6), 757–778.
Li, H. & Sun, J. (2011). Empirical research of hybridizing principal component analysis with multivariate dis-
criminant analysis and logistic regression for business failure prediction. Expert Systems with Applications, 38, 
6244–6253.

M. SITUM | THE RELEV ANCE OF EMPLOYEE-RELATED RATIOS FOR EARLY DETECTION... 313
Lin, R.-H., Wang, Y .-T., Wu, C.-H. & Chuang, C.-L. (2009). Developing business failure prediction model via RST, 
GRA and CBR. Expert Systems with Applications, 36, 1593–1600.
Madrid-Guijarro, A., Garcia-Pèrez-de-Lema, D. & van Auken, H. (2011). An analysis of non-financial factors 
associated with financial distress. Entrepreneurship & Regional Development, 23(3/4), 159–186.
Marchesini, R., Perdue, G. & Bryan V . (2004). Applying bankruptcy prediction models to distressed high yield 
bond issues. The Journal of Fixed Income, 13(4), 50–56.
McKee, T. E. (2007). Altman’s 1968 bankruptcy prediction model revisited via genetic programming: New wine 
from an old bottle or a better fermentation process? Journal of Emerging Technologies in Accounting, 4, 87–101.
McKee, T. E. (1995). Predicting bankruptcy via induction. Journal of Information Technology, 10, 26–36.
Mensah, Y. M. (1984). An examination of the stationarity of multivariate bankruptcy prediction models: A 
methodological study. Journal of Accounting Research, 22(1), 380–395.
Metz, C. E. (1978). Basic principles of ROC analysis. Seminars in Nuclear Medicine, 8(4), 283–298.
Min, J. J. & Lee, Y.-C. (2005). Bankruptcy prediction using support vector machine with optimal choice of kernel 
function parameter. Expert Systems with Applications, 28, 603–614.
Muller, G. H., Steyn-Bruwer, B. W . & Hamman, W . D. (2009). Predicting financial distress of companies listed on 
JSE – A comparison of techniques. South African Journal of Business & Management, 40(1), 21–32.
Nam, C. W ., Kim, T. S., Park, N. J. & Lee, H. K. (2008). Bankruptcy prediction using a discrete-time duration 
model incorporating temporal macroeconomic dependencies. Journal of Forecasting, 27, 493–506.
Neophytou, E. & Mar Molinero, C. (2004). Predicting corporate failure in the UK: A multidimensional scaling 
approach. Journal of Business Finance & Accounting, 31(5/6), 677–710.
Neves, J. C. & Vieira, A. (2006). Improving bankruptcy prediction with hidden layer learning vector quantiza-
tion. European Accounting Review, 15(2), 253–271.
Ohlson, J. A. (1980). Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Re-
search, 18(1), 109–131.
Ooghe, H. & Spaenjers, C. (2010). A note on performance measures for business failure prediction models. 
Journal of Applied Corporate Finance, 10(1), 21–32.
Pacey, J. W. & Pham, T. M. (1990). The predictiveness of bankruptcy models: Methodological problems and 
evidence. Australian Journal of Management, 15(2), 315–337.
Paradi, J. C., Asmild, M. & Simak, P. C. (2004). Using DEA and worst practice DEA in credit risk evaluation. 
Journal of Productivity Analysis, 21, 153–165
Pervan, I. & Kuvek, T. (2013). The relative importance of financial ratios and nonfinancial variables in predicting 
of insolvency. Croatian Operational Research Review, 4, 187-198.
Pervan, M. & Visic, J. (2012). Influence of firm size on its business success. Croatian Operational Research Review, 
3, 213–223.
Platt, H. D., Platt, M. B. & Pedersen. J. G. (1994). Bankruptcy discrimination with real variables. Journal of Busi-
ness Finance & Accounting, 21(4), 491–510.
Pohar, M., Blas, M. & Turk, S. (2004). Comparison of logistic regression and linear discriminant analysis: A 
simulation study. Metdološki Zvezki, 1(1), 143–161.

ECONOMIC AND BUSINESS REVIEW  |  VOL. 16  |  No.  3  |  2015 314
Pompe, P . P . & Bilderbeek, J. (2005). Bankruptcy prediction: The influence of the year prior to failure selected for 
model building and the effects in a period of economic decline. Intelligent Systems in Accounting, Finance and 
Management, 13, 95–112.
Pretorius, M. (2008). Critical variables of business failure: A review and classification framework. South African 
Journal of Economics and Business, 11(4), 408–430.
Samad, F., Yusof, M. A. M. & Shaharudin, R. S. (2009). Financial distress risk and stock returns: Evidence form 
the Malaysian stock market. Journal of International Finance and Economics, 9(2), 19–38.
Shah, J. R. & Murtaza, M. B. (2000). A neural network based clustering procedure for bankruptcy prediction. 
American Business Review, 18(2), 80–86.
Shin, K-S., Lee, T. S. & Kim, H.-j. (2005). An application of support vector machines in bankruptcy prediction 
model. Expert Systems with Applications, 28, 127 – 135.
Silva, D. A. P ., Stam, A. & Neter, J. (2002). The effects of misclassification costs and skewed distributions in two-
group classification, Communications in Statistics – Simulation and Computation, 31(3), 401–423.
Situm, M. (2014). The age and the size of the firm as relevant predictors for bankruptcy. Journal of Applied Eco-
nomics and Business, 2 (1), 5 – 30.
Subhash, S. (1996). Applied multivariate techniques. New Y ork: John Wiley.
Sudarsanam, S. & Lai, J. (2001). Corporate financial distress and turnaround strategies: An empirical analysis. 
British Journal of Management, 12, 183–199.
Theodossiou, P ., Kahya, E., Saidi, R. & Philippatos, G. (1996). Financial distress and corporate acquisitions: Fur-
ther empirical evidence. Journal of Business Finance & Accounting, 23(5), 699–719.
Thomas, L. C., Edelman, D. B. & Crook, J. N. (2002). Credit scoring and its applications. Philadelphia: Society for 
Industrial and Applied Mathematics.
Thornhill, S. & Amit, R. (2003). Learning about failure: Bankruptcy, firm age, and the resource-based view. 
Organization Science, 14(5), 497–509.
Wang, W . T. & Zhou, X. (2011). Could traditional financial indicators predict the default of small and medium-
sized enterprises? International Proceedings on Economics and Finance Research, 4, 72–76.
Wetter, E. & Wennberg, K. (2009). Improving business failure prediction for new firms: Benchmarking financial 
models with human and social capital. The Journal of Private Equity, 12(2), 30–37.
Whitaker, R. B. (1999). The early stages of financial distress. Journal of Economics and Finance, 23(2), 123–133.
Wilson, N., Chong, K. W . & Peel, M. J. (1995). Neural network simulation and the prediction of corporate out-
comes: Some empirical findings. International Journal of Economics of Business, 2(1), 31–50.
Wooldridge, J. M. (2006). Introductory econometrics: A modern approach (3rd ed.). Mason: Thomson. 
Yeh, C.-C., Chi, D.-J. & Hsu, M.-F. (2010). A hybrid approach of DEA, rough set and support vector machines 
for business failure prediction. Expert Systems with Applications, 37, 1535–1541.
Yim, J. & Mitchell, H. (2007). Predicting financial distress in the Australian financial service industry. Australian 
Economic Papers, 46(4), 375–388.
Zmijewski, M. E. (1984). Methodological issues related to the estimation of financial distress prediction models. 
Journal of Accounting Research, 22(Supp.), 59–82.