8SmithRoodt.qxd Contemporary research (Cummings, 2001; Denison, 2001; Dove, 2001; Flamholtz, 2001; Nadler, Thies & Nadler, 2001) has shown that organisational culture is perhaps the single best predictor of organisational performance. However, poor measurement instruments of culture, would yield poor data that renders best management interventions useless. Common pitfalls in this regard are the choice of response scale formats (Nunnally & Bernstein, 1994; Schepers, 1992; Swart, Roodt & Schepers, 1999), and wording of items (Petty, Rennier & Cacioppa, 1987; Schepers, 1992). Little research has been conducted on the metrics of different response scales in the field of organisational culture instruments. The five-point response Likert scale is used in the organisational culture instruments by many authors (Ashkanasy, Broadfoot & Falkus, 2000; Church & Waclawski, 1998; Van der Post, de Coning & Smit, 1997a). In the South African context there appears to be little appreciation for the limitations posed by the design of the Likert scale, in the context of organisational culture instrumentation. The aim of the study was to identif y which response scale from the four, five or six-point scale response format is the most effective for assessing organisational culture. This research also aimed to establish which response scale format would yield the best metric characteristics for use in organisational survey instruments. The following postulates for the investigation were proposed: � Postulate 1: Four-point response scales would yield the poorest metric properties, compared to the five and six-point response scales. � Postulate 2: Six point response scales would yield the best metric properties, compared to the four and five-point response scales. � Postulate 3: Five point response scales would yield better metric properties that the four-point response scale, but worse that the six-point response scale. Culture The earliest references to culture in the literature go back as far as 1887. The concept culture represents in broad holistic terms that, are passed from successive generations (Kotter & Heskett, 1992). The following definition indicates the universality of the concept culture: “the total of inherited ideas, beliefs, values and knowledge, which constitute the shared bases of social action” (Collins Dictionary and Thesaurus, 1988, p. 237). The above definition emphasises the broad nature of the concept culture and the manner in which culture is passed to members by social rewards and sanctions. The totality of the process is emphasised by the social nature of the transmission of everything that is important to a group. Social anthropolog y has provided the framework for the development what we currently understand to be organisational culture (Denison, 1990; Hatch, 1997; Kotter & Heskett, 1992; Ott, 1989). Kroeber and Kluckhohn (1952) have reviewed the concept culture and associated definitions. The work attests to the difficulties in defining culture. While considering over 100 definit ions, none of the definit ions were evaluated as acceptable. Common denominators among the definitions are learning and the transmission of culture. National culture Hofstede (2001) has done the most significant work on national culture and its influence in multi-national organisations. He described national culture using a systems approach where those who belong to a particular group share a value system. The norms of the grouping or societ y have resulted in the development of institutions with particular functions. Hofstede’s view of national culture presents a strong case for influence on organisational culture. The society is a social entity that has specific values, rituals, heroes and symbols peculiar to a group. Similar influences act on the culture of an organisation. Organisational culture Since the early 1980’s, organisational culture has received much attention in the literature (Peters & Waterman, 1982). Many books appeared, focussing on the performance of organisations (Alvesson, 2002; Kotter & Heskett, 1992; Peters & Waterman, STAN SMITH GERT ROODT Programme in Leadership in Performance and Change Department of Human Resources Management, RAU ABSTRACT The aim of the study was to investigate which response scale, of the four, five, or six-point response scales would yield the best metric properties on the Culture Assessment Instrument. This was achieved by utilising data sets where the respective scales were used. The subjects included participants of various organisations, ages, educational levels, language and gender groups. No one scale could be identified as having the better metric properties. The lack of conclusive results is attributed to differences in education, aggregation effect, possibility of different units of measurement being measured and the manner in which Cronbach Alpha is calculated OPSOMMING Die doel van die studie was om vas te stel watter responsskaal, van ‘n vier-, vyf-, of sespuntskaal die beste metriese eienskappe sou oplewer op die Kultuurtakseringinstrument. Dit is bereik, deur gebruik te maak van datastelle waar die skale toegepas is. Deelnemers was afkomstig uit verskeie organsisasies, ouderdomsgroepe, opvoedkundige agtergronde, taalgroepe en geslagte. Geen skaal kon gëidentifiseer word met onderskeidende metriese eienskappe nie. Die gebrek aan beduidende resultate word toegeskryf aan die verskille in die opvoedkundige agtergronde van deelnemers, samevoegingseffek, die moonlikheid van verskillende eenhede van meting en die wyse waarop Cronbach Alpha bereken is. AN EVALUATION OF RESPONSE SCALE FORMATS OF THE CULTURE ASSESSMENT INSTRUMENT Requests for copies should be addressed to: S Smith, Department of Human Resource Management, RAU University, PO Box 524, Auckland Park, 2006 60 SA Journal of Human Resource Management, 2003, 1 (2), 60-75 SA Tydskrif vir Menslikehulpbronbestuur, 2003, 1 (2), 60-75 RESPONSE SCALE FORMATS 61 1982). Other books have attempted to explain the competitiveness of Japanese organisations (Alvesson, 2002; Denison, 1990; Hatch, 1997; Ouchi, 1981). The definition of Schein has received the most attention in the literature (Hatch, 1997; Ott, 1989): “A pattern of basic assumptions-invented, discovered, or developed by a given group as it learns to cope with its problems of external adaptation and internal integration – that has worked well enough to be considered valid and therefore, to be taught to new members as the correct way to perceive, think, and feel in relation to those problems” (Schein, 1985, p. 9). Schein’s levels of organisational culture Schein (1985) identified three levels of culture; artefacts and creations; values and basic assumptions. (See figure 1). Level one, according to Schein (1985) relates to artefacts. Artefacts are conscious, obvious expressions of culture. Artefacts are a visible, tangible and audible demonstration of behaviour supported by organisational norms, values and assumptions. Artefacts range from physical aspects such as architecture to forms of language and rituals. Organisational members could be less aware of organisational culture, but it is observable to the outsider (Schein, 1985). Level two, according to Schein (1985) relates to values and norms. Values represent the principles and standards valued by organisational members. Values are the foundation as to what is acceptable and what is not acceptable. Values, though not obvious, operate uppermost in members’ minds. Organisational members are able to recognise their values when challenged by others (Schein, 1985). Norms are related to values. Norms provide the unwritten rules that indicate expectations in terms of actions applicable in a number of situations. Norms within the business environment include appropriate dress codes (Schein, 1985). Values indicate what is important to organisational members and norms help to indicate what the expectations are among organisational members. The relationship between norms and values is that which is considered acceptable and can be traced to what is valued in a particular culture. Therefore, organisational members share values and conform to norms because the foundational assumptions support the norms and values. Norms and values support the manifestation of more obvious observable behaviours. Figure 1: Schein’s model of organisational culture Level three, according to Schein (1985) relates to beliefs and assumptions. Assumptions are the basis of an organisation’s culture. Where solutions to a problem work continuously, the solution is used unconsciously and becomes the way things are done by the group. The beliefs and assumptions are the foundation of an organisation’s culture. Assumptions are the basis for the manner in which organisational members think and feel. Assumptions are unconscious and are taken for granted. Assumptions are complex in the variety of assumptions that apply in a culture at a time (Schein, 1985). Domains of organisational culture Organisational culture manifests on a continuum, from being concrete and visible to being subtle and invisible. At the one end of the continuum are artefacts and at the others are basic underlying assumptions. Artefacts Artefacts include any materials, patterns that communicate information about an organisation’s technology, beliefs, values, assumptions and practices (Ott, 1989; Schein, 1999). Artefacts are not easily understood, although visible within the organisation. Artefacts inevitably provide an “image” of the organisation and its culture (Shultz, 1995; Schein, 1999). Artefacts include symbols, language, metaphors, myths, stories, legends, heroes and ceremonies. Symbols are representations of the meanings that express more than their extrinsic content. Symbols lead to represent wider patterns of meaning causing organisational members to associate consciously or unconsciously at different levels of meaning. Symbols could include anything from a flag, to a picture of the chief executive officer or leader (Ott, 1989). Language is a part of organisational culture. Language must be learnt so that organisational members can communicate effectively. Language includes words, phrases and acronyms not comprehensible to outsiders. Language serves to identif y members of a group and those who do speak the language (Ott, 1989; Shultz, 1995). Metaphors are powerful representations of organisational language because of the communication of meaning beyond the obvious context of words (Alvesson, 2002; Ott, 1989). Myths are extended metaphors. The story, as part of the myth has a part that is factually correct and focuses on the origins of beliefs (Ott, 1989). Stories relate to anecdotal events that have occurred in the past. While similar to myths, the contents of stories tend to be accurate. Often, stories communicate morals metaphorically and can be related to the core values of the organisation. Stories communicate core messages implicitly or metaphorically. Stories have a major influence on the attitudes that members have (Ott, 1989; Wilson, 2001). Sagas and legends are the stories told that relate to the history of the organisation. Sagas and legends have the capability to illustrate the distinctiveness of the organisation. Legends provide information about allegiances, commitment and emotional investment (Ott, 1989). Heroes are the leading actors in organisational life. Heroes are the embodiment of the values of the organisation and provide a mechanism to relate the strengths of the organisation. Hero’s behaviour set standards and serve as role models (Ott, 1989; Schein, 1999). Ceremonies and commemorations are celebrations of the values and basic assumpt ions held by organisat ions. Ceremonies celebrate the achievement of heroes (Ott, 1989). Rites and ceremonies are characterised by elaborate planned activities, involving social interaction. They usually benefit SMITH, ROODT62 an audience and have social consequences for organisational members involved (Trice & Beyer, 1984). Beliefs and values Beliefs and values are at the next level on the continuum of organisational culture. In the context of organisational culture, they represent beliefs, values, ethical codes, moral codes and ideologies. Shared beliefs and values are functional for the organisational members in that choices and decisions are based on the values held. Beliefs are consciously held (Ott, 1989; Schein 1999). Values are affective desires or wants and are important to people and can be associated with almost anything. Beliefs are what people believe to be true or real. Beliefs are also what is important to people. The importance of beliefs and values can be related to the influence on the patterns of behaviour and the resultant artefacts (Ott, 1989; Schein, 1999). Espoused values represent practical applications of values, or a theoretical view of values. Beliefs and values are what people admit to and are able to articulate what the values are. Basic assumptions are what people actually believe and feel which informs their actual behaviour (Ott, 1989; Schein, 1999). Norms are the prescriptions for behaviour and serve as blueprints for organisational members in general. Norms are useful for those organisational members who perform specific functions and enact roles in the organisation. Norms have the effect of providing structure and coherence. Norms stabilise organisational behaviour and provide a framework for common expectations and attitudes (Ott, 1989; Wilson, 2001). Basic underlying assumptions Organisational assumptions are conceptually at the higher level of the organisational culture continuum. Basic assumptions are those acts and behaviours that proved to be useful so often that the behaviour is no longer conscious behaviour (Ott, 1989: Schein 1985; 1999). Time Time is not included in the previous levels of culture as conceptualised by Schein (1985). Time is related to the values and beliefs that organisational members hold in terms of activities. Time relates to the complexities of the tasks that organisational members need to complete. Tasks may be completed in a linear manner (monochronic time) or simultaneously (polychronic time) and would depend on the abilities of organisational members or the requirements of the task. Organisational culture practices would require ether polychronic or monochronic time approaches in the work environment (Bluedorn, 2000). Conceptually culture has common characteristics irrespective of the context wherein culture is described. Culture, in broad anthropological terms share similar components at national and in the organisational fields. The differences are evident only in the manner in which the components manifest at different levels in organisations. The organisational culture model aims to accommodate the components at the implicit and explicit levels in their relationship to organisational outcomes. A model of organisational culture is indicated in Figure 2. Response scales and measurement The nature of response scales used in survey instruments has an impact on the statistical analysis of the data obtained (Babbie & Mouton, 2001; Welman & Kruger, 2001). Most of what has been called measurement in organisational research involves scaling. The way scales work is not that obvious on their own. For that reason, some rules are required. Rules are important in the arena of standardisation (Nunnally & Bernstein, 1994). A measure is standardised when the rules are clear, are practical to apply, does not demand exceptional skills from users, and the results do not depend on the skills of particular user (Cronbach, 1990; Nunnally & Bernstein, 1994). The features that distinguish the levels of measurement are indicated in Table 1. The higher levels of measurement accommodate characteristics of lower levels of measurement. When designing scales, the issues of having equal intervals and absolute zero points remain problematic. Where the attribute to be measured fails to comply with the basic conditions of measurement, there are limited mathematical operations available to the practitioner. The practitioners needs to consider the assumptions made about the attribute measured and thus the choice of level of measurement, and the construction of the scale (Nunnally & Bernstein, 1994; Van der Post et al., 1997). Many response scales are available to the practitioner, though, there are only a few that are not complex, expensive, and labour intensive to develop. Some formats have been more popular than others. Others have been highly regarded, but are not often used like the Thurstone and Guttman scales (Rubin & Babbie, 1997). The Likert scale and semantic differential scale have been the more popular of the scales (Rubin & Babbie, 1997). Schepers (1992) suggested the intensity scale. The metric properties of instruments are partially dependent on the scale format used (Kerlinger & Lee, 2001; Nunnally & Bernstein, 1994). Figure 2: Model of organisational culture RESPONSE SCALE FORMATS 63 The Likert scale is one of the more popular formats used in surveys and questionnaires. An example of a more popular response format is set out in Figure 3. Likert scales offer ordinal response categories where respondents are able to provide responses indicating the intensity of their responses (Swart et al., 1999). Newstead and Arnold (1989) have found that when comparing labelled and unlabelled Likert response scales, higher means were obtained on the unlabeled scales. Strongly Disagree Neutral Agree Strongly disagree agree 1 2 3 4 5 Figure 3: Likert scale (Adapted from Church & Waclawski, 1998; Fink, 1995) Intensit y scales are similar to Likert scales, though the anchoring of the scale is limited to the extreme poles of the scale. The interval qualities of the scale become redundant as soon as more than two of the interval points are anchored (Schepers, 1992: Torgerson, 1958). The intensity scale thus provides for the advantages of the interval level of measurement and continuous scales. Refer to Figure 4 for an example of an intensity scale response format. In practice, most measurement instruments are assumed to use interval-level measurements. The absolute equality of intervals of such instruments is difficult to prove (Gregory, 1996). The controversies on the usefulness of some measures have yet to be resolved (Nunnally & Bernstein, 1994). Strongly 1 2 3 4 5 6 7 Strongly agree disagree Figure 4: Intensity scale (Adapted from Schepers, 1992) In the design of instruments a logical process needs to be followed. Table 2 provides a useful framework for the design of instruments. Essential metrics of instruments Of the various forms of reliability in the context of culture instrument development, the Kuder-Richardson Formula 20 is useful for discrete or dichotomous scores (Ghiselli et al., 1981; McIntire & Miller, 2000). For continuous scores the Cronbach Alpha formula is the appropriate formula (Cortina, 1993; Ghiselli et al., 1981; McIntire & Miller, 2000). Of the various forms of validity, important to the practitioner developing instruments, construct validity is important as it measures the theoretical construct the instrument is designed to measure (Allen & Yen, 1979; Hair, Anderson, Tatham & Black, 1998). TABLE 1 LEVELS OF MEASUREMENT Classification Level of Basic operation Description Determining Permissible statistics Response scale application of scale measurement measurement level Nominal level Equality versus Use of numerals Unique numbers – Frequencies, mode. Respondents categorise or measurement inequality to identif y objects, are assigned to sort objects or events into Classification/ individuals, events objects or events. mutually exclusive and distinctness or groups. Numbers may be exhaustive sets. Sets of changes as long data are mutually as the numbers exclusive for each object assigned to objects and are sorted into one are different. set only. Ordinal level Greater than In addition to Larger numbers Including statistics Respondents rank-order measurement versus less than identification are assigned with from the previous object or events in terms Ordering of of, numerals more that one level of some property. Objects objects describe relative property measured. – Median, percentiles, or events are ranked higher characteristics When scale value is order statistics, or lower by assigned values posed by the assigned, the scale correlation. according to the property. event, individual. may be changed as Rating scales usually long as the ordering involve people by asking of the scales to indicate opinions, maintained. beliefs, feelings or attitudes. Interval level Determination Includes the Interval level have Including statistics Respondents to assign measurement of equality of properties of numbers that allow from the previous numbers to stimuli or interval or nominal and calculation and level. differences to stimuli differences. ordinal scales. interpretation of – Arithmetic mean, through direct Equal intervals Including intervals ratio interval/ variance. Pearson estimation, produces between consecutive intervals between correlation. interval scales. The points that are equal. scales. scale values produced often take the mean or median of the values obtained from many respondents. Direct estimation methods assume respondents are skilled enough to make interval judgements. Ratio level Determination of Includes the Ratio level scales Including statistics Ratio scales are produced measurement equality of ratios. properties of have numbers that from the previous using the method of direct Absolute zero nominal, ordinal allow calculation level. estimation. Respondents and interval scales. and interpretation – Geometric mean. are required to assign Has an absolute of ratios of scale numbers to stimuli zero point. values or ratio stimuli. (Adapted from: Allen & Yen, 1979; Babbie & Mouton, 2001; Gregory, 1996; McDaniel & Gates, 2001; Nunnally & Bernstein, 1994; Welman & Kruger, 2001), C a te g o ri a l sc a le s C o n ti n u o u s sc a le s SMITH, ROODT64 TABLE 2 DESIGN PROCESS STEPS Feigl 1970, p.9 Swart et al, Rust & Golombok, Organisational 1999 p.34 1999 pp,196-217, culture instrumentation development summary Postulates Define theoretical Define the purpose Identif y the foundations of of the instrument. construct – construct. culture. Primitive Identif y the Develop blue print/ Define concepts domains within specifications for organisational the construct. instrument. culture model. Defined Identif y the sub- Identif y and define Identif y the sub- concepts domains content areas. domains of culture. Empirical Identif y/develop Identif y how the Operationalise concepts behavioural construct would the sub-domains indicators. be manifested (that in behavioural is the behavioural terms. affective areas). Observations/ Develop item Develop the items. Develop the item experiences format taking format taking care care of the of technical technical details as requirements of indicated by instruments and Schepers, 1992, metrics. pp.2-7. Response styles No rational explanation can be found for the choices of respondents in terms of the choices respondents make on response scales. The choices made could be ascribed to realistic responses, ambiguity, meaninglessness or difficulty of items. Under these conditions the respondent will respond according to a particular response style (Brown, 1983). The controversy regarding response sets combined with other diffuse factors have yet to be resolved. The responses on self-report instruments are a combination of self-deception, impression management and realistic self-portrayal (Anatasi & Urbani, 1997). The practitioner needs to take cognisance of the response bias discussed below when interpreting results of instruments. Halo effect, which is characterised by a favourable or unfavourable attribute of the person, tends to rate the characteristics of the attribute in favourable or unfavourable terms of the attributes that have little relation to the attribute rated (Welman & Kruger, 2001). Leniency or stringency error refers to the respondent who rates all individual or attributes either too strictly or too leniently (Guilford, 1954; Welman & Kruger, 2001). Blacks, when compared to whites, tend to focus their responses at one end of the response scale (Bachman & O’Malley, 1984). Greenleaf (1992) reported conflicting findings in research on extreme response style. He indicated that there is evidence to suggest that the response styles are not stable and that response styles are not necessarily related to personality or demographic variables. He did report that income, education and age were associated with increased extreme response style. Logical error is similar to the halo effect (Guilford, 1954). Logical error is the tendency to rate attributes that are considered logically related in a similar way (Guilford, 1954; Welman & Kruger, 2001). Central tendency bias occurs where respondents are reluctant to rate attributes at the extremes of the scale, thus tending to rate most attributes at the centre of the scale (Guilford, 1954; Welman & Kruger, 2001). The challenge of dealing with central tendency or “don’t know” responses continues to be a problem for practitioners and the associated interpretations of the response (Fleick, 1989; Poe, Seeman, Mclaughlin, Mehl & Dietz, 1988; Duncan & Stenbeck, 1988). The origins of central tendency responses are multiple and could be ascribed to an error response in the sense that the respondents may have misunderstood the item. Secondly the respondent is ambivalent or ignorant to the alternatives available (Fleick, 1989; Sanchez & Morchio, 1992) and finally, the responses could be a “non- attitude” (Fleick, 1989; Francis & Busch, 1975; Gilljam & Granberg, 1993). Where the response options include “no opinion” or “not sure” these responses should not be interpreted as being interchangeable (Duncan & Stenbeck, 1988). Constant error occurs where respondents tend to exaggerate the difference between themselves on an attribute or those being rated (Guilford, 1954; Welman & Kruger, 2001). Guilford (1954) reported that the phenomenon may be ascribed to respondents requiring others to be similar to them in terms of the attribute and are surprised when the opposite is true. Proximity error introduces high covariances among construct measures. Error is attributed to the nearness in space and time of items in an instrument. Similar items spaced close to each other tend to inter-correlate higher than when items are spaced apart (Guilford, 1954). Statement formats and question formats have their individual effects on response st yles. In personalit y and interest instruments, use is made of questions. Use is also made of intensity scales, which are often designed as ordinal measures. At times use is also made of the Likert scale, which is also an ordinal response scale (Schepers, 1992). Where statements are used in item design, respondents are likely to respond in the affirmative without having considered the content of the item (Petty et al., 1987; Schepers, 1992) This response style is referred to an acquiescence response style (Anastasi, 1968; Jackson & Messick, 1967). To deal with the problem, questions should be asked that require the respondent to engage with the items to avoid the acquiescent response (Schepers, 1992). Questions require more thoughtful responses (Petty et al., 1987). Prestige bias occurs with the use of prestige names or intentional words that have value which are attached to the words. The effect is to confuse attitude with the evaluation of the issue at hand. Furthermore, prestige names add to the stimuli presented, thus contributing to the variance. The interpretation of the responses becomes difficult as respondents may or may not be responding to the named symbols or issue referred to (Smith & Squire, 1990). Brown (1983) offers some advice to control bias. He suggested obtaining the co-operation of the respondents by explaining the purpose of the instrument. The instrument should also have clear instructions and be well structured to eliminate ambiguity. Wording effects that affect metrics of instruments Avoid abbreviations and vaguely worded items. Only those abbreviations and terms that have commonly understood meanings should be used in instruments (Fink, 1995b; Fowler, 1992; Gaskill, O’Muircheartaigh & Wright, 1994; Sarantakos, 1998). Avoid slang and colloquial expressions. An instrument is likely to become redundant as words go out of fashion (Fink, 1995b; Sarantakos, 1998). Avoid technical terminology by using plain language in question wording wherever possible (Fife-Schaw, 2000). Avoid Intensifier words. Intensifier words tend to magnif y the meanings of words. Words like very, good, satisfied and really may influence respondents to respond in a particular direction on a scale. Intensifier words effects are dependent on the context in which the words are used. Intensifiers do not create consistent responses in all sit uations (O’Muircheartaigh, Gaskill & Wright, 1993). RESPONSE SCALE FORMATS 65 Avoid value judgements in item wordings. The views of sponsors should not be mentioned in items (Fife-Schaw, 2000). Context effects cannot be divorced from the context in which instruments are used. Questions referring to issues that are current in an organisation or other questions in the instrument will influence responses to questions (Fife-Schaw, 2000). Avoid hidden assumptions. Items should not contain assumptions where respondents are required to respond to a situation that they have not been exposed to before or are not likely to be exposed to (Fife-Schaw, 2000). RESEARCH METHOD Research design The research design followed in this study was ex post facto in nature where secondary data was used for the data analysis. The sample The sample for the study comprised of members working in different organisations ranging from service orientated organisations to information orientated organisations. The sample described below, represents the data common to all the samples utilised in the study. From Table 3 it is clear that most respondents are male, from different language groups and between 25 – 35 years old. TABLE 3 THE SAMPLE Characteristics 4-points 5-point 6-point response scale response scale response scale Gender Males 1485 2034 381 Females 36 1879 27 Missing values 150 153 442 Total 1671 4066 850 Language Afrikaans 330 1014 193 English 111 225 41 Ndebele 4 4 North Sotho 68 31 South Sotho 299 164 Swazi 84 32 Tsongo 134 10 Tsawa 174 86 Venda 23 3 Xhosa 264 124 Zulu 108 71 Shangaan 62 Other languages 47 21 1 Missing data 25 2806 28 Total 1671 4066 850 Age 24 years and younger 112 436 41 25-35 years of age 672 1808 323 36-45 years of age 549 1008 346 46 years of age 306 638 108 Missing data 32 176 32 Total 1671 4066 850 Organisations Information Financial services Mining technology Retail Postal services The measuring instrument Different response scale formats of the Culture Assessment Instrument (CAI) were used in this study. The instrument was originally developed to measure the organisational culture of a financial institution. The instrument has since been used to assess organisational culture in other South African organisations (Martins & Martins, 2001). The reliability of the five-point response scale (Cronbach Alpha) is 0,933. The internal consistency of the culture dimensions measures range from 0,655 to 0,932. Test-retest reliability is between 0,933 and 0,955 (Martins, 2000). The instrument is modelled on the work of Schein (Martins, 2000). The instrument was initially developed with five-point Likert response-scale-requiring reactions to statements in positive and negative formats. The participant must indicate whether he/she, differs or agrees. The points on the scale are marked as follows: Scale point 1, indicates strongly disagree, scale point 2 indicates differ, scale point 3 indicates uncertain, scale point 4 indicates agree, and scale point 5 indicates strongly agree. In adapting the instrument for customer requirements, use was made of the four and six-point response scales. The scales marking were: Four-point response scale were marked, 1 indicated strongly agree, 2 indicated agree, 3 indicates disagree and 4 indicates strongly disagree. Six-point response scale were marked, 1 indicated strongly agree, 2, indicates agree, 3 indicates slightly agree, 4 indicates slightly disagree, 5, indicates disagree and 6, indicates strongly disagree. The full questionnaire had 79 items. Only 38 items from the questionnaire were used that were generic to all the organisations where the instrument was used. Research procedure The steps followed in the research process are reported in Table 4. TABLE 4 METHODOLOGY PROCESS STEPS Phase Action One A database was developed after the Culture Assessment Instrument was used in different organisational settings. The settings include the application of the three response scales formats, i.e. four, five and six-point response scale format. Two Questions common to all the organisations were identified, Non- generic items were removed from the data set. Three The data sets were checked to ensure the correctness and completeness of the data. Four The data were then subjected to statistical analyses using the SPSS program to: � Obtain the descriptive statistics for the respective scales. � Establish the factor structure for the respective response scales using first and second order level factor analyses. � Establish the internal reliability, (Cronbach’s alpha) through iterative item analyses. Five The analysed information will be reported and interpreted, From the results, recommendations will be made for future research. SMITH, ROODT66 Statistical analysis With the data available, the data analysis is possible. As indicated earlier, the focus is on the data obtained from the four, five and six-point response scale format on the CAI. The purpose of the approach outlined below is to reduce the data so that conclusions may be drawn. RESULTS Results pertaining to the four-point response scale The data for the four-point response scale originated in an information technology organisation. More detail of the sample is reported in Table 3. Descriptive statistics for the four-point scale are reported in Table 5. TABLE 5 FOUR POINT RESPONSE SCALE DESCRIPTIVE STATISTICS Item Mean Median Mode Std. Skew- Std. Error Kurtosis Std. Error deviation ness of skewness of kurtosis Q2 2,86 3,00 3 0,909 -0,564 0,060 -0,392 0,120 Q3 2,85 3,00 3 0,902 -0,580 0,060 -0,337 0,120 Q4 2,67 3,00 3 0,962 -0,360 0,060 -0,803 0,120 Q5 2,39 2,00 3 0,980 0,016 0,060 -1,039 0,120 Q6 2,58 3,00 3 1,027 -0,171 0,060 -1,103 0,120 Q9 2,69 3,00 3 0,983 -0,308 0,060 -0,905 0,120 Q10 2,73 3,00 3 1,066 -0,347 0,060 -1,116 0,120 Q11 2,83 3,00 3 0,943 -0,450 0,060 -0,669 0,120 Q12 2,41 3,00 3 1,088 0,015 0,060 -1,313 0,120 Q14 2,52 3,00 3 1,017 -0,108 0,060 -1,102 0,120 Q15 2,77 3,00 3 0,972 -0,497 0,060 -0,694 0,120 Q16 2,39 2,00 3 1,038 0,026 0,060 -1,195 0,120 Q17 2,52 3,00 3 0,921 -0,234 0,060 -0,817 0,120 Q18 2,47 3,00 3 1,035 -0,078 0,060 -1,169 0,120 Q20 2,56 3,00 3 0,971 -0,246 0,060 -0,930 0,120 Q21 2,56 3,00 3 1,005 -0,221 0,060 -1,036 0,120 Q25 2036 2,00 3 0,915 0,024 0,060 -0,873 0,120 Q26 2,39 2,00 3 0,921 -0,041 0,060 -0,898 0,120 Q31 2,32 2,00 2 0,988 0,154 0,060 -1,033 0,120 Q30 2,45 3,00 3 0,995 -0,098 0,060 -1,078 0,120 Q32 2,38 2,00 3 0,987 0,010 0,060 -1,067 0,120 Q33 2,37 3,00 3 1,016 -0,101 0,060 -1,211 0,120 Q34 2,53 3,00 3 0,996 -0,198 0,060 -1,030 0,120 Q38 2,81 3,00 3 0,876 -0,597 0,060 -0,207 0,120 Q39 2,72 3,00 3 0,922 -0,450 0,060 -0,583 0,120 Q40 2,89 3,00 3 0,833 -0,685 0,060 0,146 0,120 Q42 2,79 3,00 3 0,958 -0,544 0,060 -0,592 0,120 Q43 2,58 3,00 3 0,929 -0,336 0,060 -0,764 0,120 Q44 1,85 1,00 1 0,998 0,791 0,060 -0,657 0,120 Q47 2,98 3,00 3 0,877 -0,730 0,060 -0,013 0,120 Q48 2,68 3,00 3 0,932 -0,455 0,060 -0,630 0,120 Q50 2,82 3,00 3 0,905 -0,548 0,060 -0,392 0,120 Q51 3,11 3,00 3 0,903 -0,890 0,060 0,071 0,120 Q52 2,58 3,00 3 0,903 -0,270 0,060 -0,704 0,120 Q53 2,74 3,00 3 0,907 -0,458 0,060 -0,518 0,120 Q54 2,89 3,00 3 0,863 -0,648 0,060 -0,064 0,120 Q55 2,93 3,00 3 0,898 -0,628 0,060 -0,274 0,120 Q56 2,73 3,00 3 0,948 -0,444 0,060 -0,676 0,120 N = 167 Missing Values = 0 Minimum Value = 1 Maximum Value = 4 The data set for the four-point response scale was factor analysed on two levels according to a procedure suggested by Schepers (1992) in order to determine the factor structure of the instrument. The data was analysed using the SPSS programme. To determine the suitability of the inter-correlation matrix for factor analysis, the Kaiser-Meyer-Olkin (KMO) Measure of Sampling adequacy (MSA) and Bartlett’s Test of Sphericity were conducted on the matrix. The KMO yielded a MSA of 0,953 and the Bartlett’s Test a Chi-square of 22158 (p = 0,000). The matrix is therefore suitable for further factor analysis. The items of the CAI were inter-correlated and the eigenvalues of the unreduced matrix calculated. Due to the limited space, the inter- correlation matrix is not reported here. The number of factors postulated according to Kaiser’s (1970) criterion (eigenvalues greater that unity) are reported in Table 6. These seven factors explain about 52% of the variance in the factor space. The item loadings on the seven postulated factors are presented in Table 7. The loadings are reported in bold type for each of the factors. Only items with values greater than 0,3 were included in this sorted matrix. TABLE 6 EIGENVALUES ON THE ITEM INTER-CORRELATION MATRIX (38 X 38) Initial eigenvalues Root Eigenvalue % of variance Cumulative % 1 11,357 29,887 29,887 2 1,929 5,076 34,962 3 1,524 4,011 38,973 4 1,371 3,608 42,581 5 1,356 3,567 46,148 6 1,135 2,986 49,134 7 1,101 2,898 52,032 8 0,994 2,616 54,649 9 0,940 2,474 57,123 10 0,884 2,326 59,448 11 0,837 2,204 61,652 12 0,786 2,067 63,719 13 0,763 2,007 65,726 14 0,736 1,937 67,663 15 0,729 1,918 39,581 16 0,687 1,809 71,390 17 0,654 1,702 73,110 18 0,633 1,666 74,776 19 0,625 1,645 76,422 20 0,596 1,568 77,990 21 0,582 1,532 79,522 22 0,573 1,509 81,031 23 0,545 1,434 82,465 24 0,534 1,406 83,871 25 0,522 1,372 85,244 26 0,514 1,353 86,597 27 0,490 1,289 87,886 28 0,474 1,246 89,132 29 0,450 1,185 90,317 30 0,446 1,173 91,490 31 0,441 1,160 92,651 32 0,430 1,133 93,783 33 0,426 1,120 94,903 34 0,413 1,087 95,990 35 0,408 1,072 97,062 36 0,384 1,010 98,072 37 0,371 0,976 99,048 38 0,362 0,952 100,000 Trace = 38 RESPONSE SCALE FORMATS 67 TABLE 7 ITEM LOADINGS ON SEVEN POSTULATED FACTORS Factor Items 1 2 3 4 5 6 7 48 0,568 56 0,528 52 0,494 39 0,494 55 0,487 0,329 40 0,464 0,344 42 0,433 43 0,407 53 0,609 54 0,560 51 0,325 0,558 50 0,523 47 0,411 0,471 38 0,431 18 0,399 0,393 16 0,654 12 0,544 17 0,372 26 0,306 0,348 44 0,328 15 0,317 32 0,672 33 0,324 0,494 31 0,480 0,319 34 0,342 0,432 30 0,417 10 0,636 11 0,326 0,523 9 0,489 21 0,396 14 0,322 0,335 3 0,610 4 0,505 5 0,384 0,349 2 0,381 25 0,304 0,308 0,347 20 0,268 6 0,515 Sub-scores on each of the postulated factors were calculated by adding item scores. These sub-scores were again inter-correlated and the results are portrayed in Table 8. TABLE 8 INTER-CORRELATION OF SUB-SCORES ON SEVEN POSTULATED FACTORS Factors Sub scores 1 2 3 4 5 6 7 1 1,000 0,680 0,626 0,637 0,566 0,586 0,359 2 0,680 1,000 0,566 0,493 0,598 0,525 0,360 3 0,626 0,566 1,000 0554 0,517 0,568 0,396 4 0,637 0,493 0,554 1,000 0,478 0,546 0,406 5 0,566 0,598 0,517 0,478 1,000 0,497 0,232 6 0,586 0,525 0,568 0,46 0,497 1,000 0,448 7 0,359 0,360 0,396 0,406 0,323 0,448 1,000 All correlations are significant at the p = 0,05 level. Eigenvalues were again calculated on this unreduced inter- correlation matrix. Only one factor was postulated according to Kaiser’s (1970) criterion (eigenvalues greater that unity) that accounts for about 59% variance in factor space. The results appear in Table 9. TABLE 9 EIGENVALUES ON THE SUB-SCORE INTER-CORRELATION MATRIX (7 X 7) Initial eigenvalues Root Eigenvalue % of variance Cumulative % 1 4,100 58,573 58,573 2 0,768 10,971 69,544 3 0,555 7,928 77,471 4 0,455 6,497 83,969 5 0,440 6,279 90,248 6 0,410 5,858 96,105 7 0,273 3,895 100,000 Trace = 7 One factor was extracted using principal axis factoring. The loadings of sub-scores on the single factor appear in Table 10. TABLE 10 SUB-SCORE LOADINGS ON SECOND LEVEL FACTOR Factor Loadings Communalities Sub-scores 1 1 0,831 0,691 2 0,761 0,579 3 0,758 0,574 4 0,735 0,525 5 0,728 0,479 6 0,692 0,540 7 0,508 0,258 Iterative item analyses were conducted on the single obtained scale. The item-test correlations as well as the test reliabilities (Cronbach Alpha) with the respective item deleted appear in Table 11. The obtained single scale yielded a Cronbach Alpha of 0,9345. The reliability item statistics for the four-point scale are reported in Table 11. Results pertaining to the five-point response scale The sample for the five-point response scale is reported in Table 3. Descriptive statistics for the five-point scale are reported in Table 12. The data set for the five-point response scale was factor analysed on two levels according to a procedure suggested by Schepers (1992) in order to determine the factor structure of the instrument. The data was analysed using the SPSS programme. SMITH, ROODT68 TABLE 11 ITEM RELIABILITY STATISTICS FOR THE FOUR-POINT SCALE Item Scale mean if Scale variance if Corrected item Alpha if item item deleted item deleted Total correlation deleted 2 96,8294 396,5332 0,4462 0,9333 3 96,8324 396,1647 0,4611 0,9332 4 97,0150 368,0076 0,4612 0,9332 5 97,2908 369,9693 0,3984 0,9338 6 97,1065 366,2485 0,4745 0,9331 9 96,9916 366,0766 0,5029 0,9328 10 96,9527 369,3337 0,3778 0,9341 11 96,8612 368,6059 0,4547 0,9333 12 97,2795 367,1883 0,4217 0,9337 14 97,1688 363,3248 0,5570 0,9323 15 96,9144 365,6615 0,5206 0,9327 16 97,2974 369,9193 0,5034 0,9328 17 97,1682 368,0046 0,4837 0,9330 18 97,2190 365,9951 0,4772 0,9331 20 97,1239 367,4284 0,4726 0,9331 21 97,1287 365,4415 0,5077 0,9328 25 97,3279 367,5451 0,5010 0,9328 26 97,2974 366,3815 0,5311 0,9326 31 97,3656 366,7051 0,4828 0,9330 30 97,2382 362,2917 0,5987 0,9319 32 97,3004 366,8654 0,4794 0,9331 33 97,3107 367,0634 0,4585 0,9333 34 97,1538 364,0955 0,5488 0,9324 38 96,8749 369,7670 0,4579 0,9332 39 96,9701 363,8817 0,6037 0,9319 40 96,7965 369,5322 0,4910 0,9330 42 96,8971 370,2409 0,4013 0,9338 43 97,1041 363,7173 0,6029 0,9319 44 97,8360 368,5192 0,4293 0,9335 47 96,7050 365,0620 0,6005 0,9320 48 97,0018 361,9862 0,6512 0,9315 50 96,8683 366,6749 0,5326 0,9326 51 96,5709 366,1852 0,5482 0,9324 52 97,1017 366,3238 0,5443 0,9325 53 96,9467 365,6073 0,5627 0,9323 54 96,7923 368,7742 0,4959 0,9329 55 96,7576 364,2939 0,6085 0,9319 56 96,9563 362,7831 0,6165 0,9318 No of cases = 1671 No of items = 38 Cronbach alpha 0,9345 To determine the suitability of the inter-correlation matrix for factor analysis, the Kaiser-Meyer-Olkin (KMO) Measure of Sampling adequacy (MSA) and Bartlett’s Test of Sphericity were conducted on the matrix. The KMO yielded a MSA of 0,960 and the Bartlett’s Test a Chi-square of 47436 (p = 0,000). The matrix is suitable for further factor analysis. The items of the C AI were inter-correlated and the eigenvalues of the unreduced matrix calculated. Due to the limited space, the inter-correlat ion matrix is not reported here. The number of factors postulated according to Kaiser’s (1970) criterion (eigenvalues greater that unit y) are reported in Table 13. These six factors explain about 47% of the variance in the factor space. The item loadings on the six postulated factors are presented in Table 14. The loadings are reported in bold t ype for each of the factors. Only items with values greater than 0,3 were included in this sorted matrix. TABLE 12 FIVE-POINT RESPONSE SCALE DESCRIPTIVE STATISTICS Item Mean Median Mode Std. Skew- Std. Error Kurtosis Std. Error deviation ness of skewness of kurtosis Q2 3,93 4,00 4 0,904 -1,135 0,038 1,418 0,077 Q3 3,82 4,00 4 1,092 -0,959 0,038 0,161 0,077 Q4 3,32 4,00 4 1,236 -0,443 0,038 -1,005 0,077 Q5 2,82 3,00 2 1,214 -0,114 0,038 -1,043 0,077 Q6 3,06 3,00 4 1,394 -0,148 0,038 -1,381 0,077 Q9 3,41 4,00 4 1,236 -0,462 0,038 -1,033 0,077 Q10 3,33 4,00 4 1,370 -0,375 0,038 -1,192 0,077 Q11 3,40 4,00 4 1,243 -0,430 0,038 -1,020 0,077 Q12 3,80 4,00 4 1,217 -0,988 0,038 -0,045 0,077 Q14 2,99 3,00 4 1,340 -0,084 0,038 -1,376 0,077 Q15 3,72 4,00 4 1,241 -0,899 0,038 -0,290 0,077 Q16 2,88 3,00 4 1,385 -0,011 0,038 -1,375 0,077 Q17 2,83 3,00 2 1,197 0,044 0,038 -1,045 0,077 Q18 3,01 3,00 4 1,371 -0,194 0,038 -1,307 0,077 Q20 3,58 4,00 4 1,047 -0,664 0,038 -0,160 0,077 Q21 3,03 3,00 4 1,336 -0,104 0,038 -1,257 0,077 Q25 3,14 3,00 4 1,181 -0262 0,038 -1,039 0,077 Q26 3,21 4,00 4 1,253 -0,379 0,038 -1,053 0,077 Q31 2,86 3,00 4 1,246 0,032 0,038 -1,274 0,077 Q30 2,74 2,00 4 1,323 0,123 0,038 -1,349 0,077 Q32 2,98 3,00 4 1,254 -0,137 0,038 -1,213 0,077 Q33 3,40 4,00 4 1,191 -0,625 0,038 -0,708 0,077 Q34 3,32 4,00 4 1,139 -0,450 0,038 -0,854 0,077 Q38 3,16 3,00 4 1,171 -0,272 0,038 -0,943 0,077 Q39 3,68 4,00 4 1,018 -0,827 0,038 0,035 0,077 Q40 3,30 4,00 4 1,281 -0,483 0,038 -0,985 0,077 Q42 3,23 4,00 4 1,441 -0,343 0,038 -1,322 0,077 Q43 2,81 3,00 4 1,292 -0,019 0,038 -1,291 0,077 Q44 2,59 2,00 2 1,253 0,283 0,038 -1,181 0,077 Q47 3,53 4,00 4 1,114 -0,765 0,038 -0,303 0,077 Q48 3,35 4,00 4 1,211 -0,536 0,038 -0,810 0,077 Q50 3,79 4,00 4 1,170 -0,904 0,038 -0,152 0,077 Q51 3,44 4,00 4 1,132 -0,583 0,038 -0,611 0,077 Q52 2,95 3,00 4 1,168 -0,109 0,038 -1,103 0,077 Q53 3,41 4,00 4 1,131 -0,583 0,038 -0,696 0,077 Q54 3,41 4,00 4 1,131 -0,583 0,038 -0,622 0,077 Q55 3,41 4,00 4 1,229 -0,540 0,038 -0,784 0,077 Q56 3,17 4,00 4 1,387 -0,239 0,038 -1,314 0,077 N = 4066 Missing values = 0 Minimum value = 1 Maximum value = 5 Sub-scores on each of the postulated factors were calculated by adding item scores. These scores were again inter-correlated and the results are portrayed in Table 15. A KMO test for sampling adequacy and Bartlett’s test for sphericity was performed to test the suitability of this matrix for factor analysis. Eigenvalues were again calculated on this unreduced inter- correlation matrix. Only one factor was postulated according to Kaiser’s (1970) criterion (eigenvalues greater that unity) that accounts for about 57% variance in factor space. The results appear in Table 16. One factor was extracted using principal axis factoring. The loadings of sub-scores on the single factor appear in Table 17. Iterative item analyses were conducted on the single obtained scale. The item-test correlations as well as the test reliabilities (Cronbach Alpha) with the respective items deleted appear in RESPONSE SCALE FORMATS 69 Table 18. The obtained single scale yielded a Cronbach Alpha of 0,9248. TABLE 13 EIGENVALUES ON THE ITEM INTER-CORRELATION MATRIX (38 × 38) Initial eigenvalues Root Eigenvalue % of variance Cumulative % 1 10,529 27,709 27,709 2 2,013 5,297 33,006 3 1,627 4,282 37,288 4 1,313 3,454 40,742 5 1,170 3,080 43,822 6 1,134 2,985 46,808 7 0,986 2,596 49,403 8 0,929 2,445 51,849 9 0,884 2,327 54,176 10 0,853 2,245 56,421 11 0,808 2,153 58,574 12 0,788 2,125 60,699 13 0,771 2,074 62,773 14 0,745 2,029 64,802 15 0,731 1,960 66,762 16 0,718 1,923 68,685 17 0,699 1,890 70,575 18 0,693 1,839 72,414 19 0,648 1,824 74,239 20 0,624 1,706 75,945 21 0,624 1,643 77,588 22 0,615 1,618 79,206 23 0,601 1,583 80,789 24 0,593 1,560 82,345 25 0,584 1,538 83,667 26 0,573 1,507 85,393 27 0,558 1,467 86,860 28 0,529 1,391 88,252 29 0,508 1,337 89,589 30 0,507 1,334 90,923 31 0,488 1,284 92,206 32 0,471 1,239 93,446 33 0,464 1,221 94,667 34 0,431 1,134 95,801 35 0,424 1,115 96,916 36 0,412 1,085 98,001 37 0,399 1,049 99,050 38 0,361 0,950 100,00 Trace = 38 TABLE 14 ITEM LOADINGS ON SIX POSTULATED FACTORS Factor Items 1 2 3 4 5 6 16 0,611 18 0,609 42 0,606 0,332 40 0,573 17 0,565 26 0,553 55 0,506 0,362 48 0,424 0,306 43 0,421 25 0,419 14 0,335 38 0,296 44 0,312 0,476 34 0,458 50 0,449 20 0,429 53 0,421 39 0,412 21 0,362 0,385 0,330 15 0,354 11 0,697 10 0,302 0,631 9 0,585 12 6 0,499 4 0,497 5 0,490 3 0,461 2 0,370 30 0,491 31 0,496 32 0,447 47 0,307 0,318 33 51 0,429 54 0,339 0,401 52 0,381 56 0,353 TABLE 15 INTER CORRELATION MATRIX OF SUB-SCORES ON SIX POSTULATED FACTORS (6 X 6) Factor Factors 1 2 3 4 5 6 1 1,000 0,575 0,559 0,500 0,682 0,542 2 0,575 1,000 0,493 0,475 0,574 0,389 3 0,559 0,493 1,000 0,360 0,439 0,332 4 0,500 0,475 0,360 1,000 0,517 0,340 5 0,682 0,574 0,439 0,517 1,000 0,460 6 0,542 0,389 0,332 0,340 0,460 1,000 All correlations are significant at the p = 0,05 level. TABLE 16 EIGENVALUES ON THE SUB-SCORE INTER-CORRELATION MATRIX (6 X 6) Initial eigenvalues Root Eigenvalue % of variance Cumulative % 1 3,440 57,339 57,339 2 0,694 11,566 68,905 3 0,655 10,919 79,824 4 0,488 8,139 87,963 5 0,437 7,277 95,240 6 0,286 4,760 100,00 Trace = 6 SMITH, ROODT70 TABLE 17 SUB-SCORE LOADINGS ON THE SECOND LEVEL FACTOR Factor loadings Communalites Sub-scores 1 1 0,858 0,736 2 0,789 0,520 3 0,721 0,379 4 0,617 0,380 5 0,615 0,623 6 0,577 0,333 TABLE 18 ITEM RELIABILITY STATISTICS FOR THE FIVE-POINT SCALE Item Scale mean if Scale variance if Corrected item Alpha if item item deleted item deleted Total correlation deleted 2 119,9764 566,1874 0,2836 0,9246 3 120,0821 562,6545 0,2964 0,9247 4 120,5831 547,7880 0,5165 0,9225 5 121,0876 553,2834 0,4282 0,9236 6 120,8689 547,5689 0,4553 0,9232 9 120,4889 547,2330 0,5116 0,9226 10 120,5780 546,8634 0,4754 0,9230 11 120,5010 548,5359 0,5000 0,9227 12 120,1008 555,9539 0,3793 0,9240 14 120,9129 549,9821 0,4363 0,9234 15 120,1840 559,3447 0,3120 0,9247 16 121,0189 539,0533 0,5947 0,9215 17 121,0760 543,4464 0,6151 0,9215 18 120,8947 540,7938 0,5728 0,9218 20 120,3232 561,1290 0,3423 0,9242 21 120,8756 542,3004 0,5642 0,9219 25 120,7641 546,0833 0,5750 0,9219 26 120,6904 538,8682 0,6662 0,9209 31 121,0401 547,6995 0,5138 0,9225 30 121,1596 540,0791 0,6080 0,9214 32 120,9225 547,0875 0,5208 0,9225 33 120,5049 555,2311 0,4018 0,9237 34 120,5856 555,4934 0,4177 0,9235 38 120,7469 554,3386 0,4262 0,9235 39 120,2226 546,5996 0,2806 0,9247 40 120,6058 539,9269 0,6322 0,9212 42 120,6242 543,6939 0,4974 0,9228 43 121,0935 548,9256 0,4727 0,9230 44 121,3180 549,6890 0,4755 0,9230 47 120,3684 548,4453 0,5656 0,9221 48 120,5553 544,0846 0,5956 0,9217 50 120,1178 551,6636 0,4763 0,9230 51 120,4678 548,7199 0,5506 0,9222 52 120,9501 555,2297 0,4108 0,9236 53 120,4338 551,1426 0,4883 0,9228 54 120,4934 552,9482 0,4696 0,9230 55 120,4899 542,4661 0,6156 0,9214 56 120,7381 570,5555 0,1015 0,9274 No of cases = 4066 No of items = 38 Cronbach alpha 0,9248 Results pertaining to the six-point scale The sample is reported in Table 3. The descriptive statistics are reported in Table 19. TABLE 19 SIX-POINT RESPONSE SCALE DESCRIPTIVE STATISTICS Item Mean Median Mode Std. Skew- Std. Error Kurtosis Std. Error deviation ness of skewness of kurtosis Q2 3,70 4,00 5 1,682 -0,402 0,084 -1,279 0,168 Q3 4,18 5,00 5 1,461 -0,780 0,084 -0,455 0,168 Q4 3,78 4,00 5 1,537 -0,530 0,084 -0,923 0,168 Q5 3,61 4,00 5 1,597 -0,315 0,084 -1,223 0,168 Q6 3,58 4,00 5 1,644 -0,251 0,084 -1,253 0,168 Q9 4,41 5,00 5 1,495 -0,663 0,084 -0,683 0,168 Q10 4,15 5,00 5 1,547 -0,647 0,084 -0,768 0,168 Q11 4,22 5,00 5 1,391 -0,726 0,084 -0,368 0,168 Q12 4,16 5,00 5 1,646 -0,652 0,084 -0,847 0,168 Q14 3,59 4,00 5 1,569 -0,223 0,084 -1,210 0,168 Q15 4,33 5,00 5 1,329 -0,885 0,084 0,054 0,168 Q16 3,61 4,00 5 1,649 -0,204 0,084 -1,286 0,168 Q17 3,38 4,00 4 1,469 -0,150 0,084 -1,105 0,168 Q18 3,53 4,00 5 1,695 -0,158 0,084 -1,344 0,168 Q20 3,77 4,00 5 1,454 -0,502 0,084 -0,894 0,168 Q21 3,26 3,00 5 1,595 0,089 0,084 -1,233 0,168 Q25 3,51 4,00 5 1,516 -0,251 0,084 -1,156 0,168 Q26 3,66 4,00 5 1,562 -0,329 0,084 -1,189 0,168 Q31 3,31 4,00 4 1,513 -0,063 0,084 -1,185 0,168 Q30 3,59 4,00 5 1,551 -0,358 0,084 -1,129 0,168 Q32 3,26 3,00 4 1,528 -0,041 0,084 -1,252 0,168 Q33 3,47 4,00 5 1,585 -0,213 0,084 -1,271 0,168 Q34 ,076 4,00 5 1,534 -0,362 0,084 -1,014 0,168 Q38 4,23 5,00 5 1,340 -0,869 0,084 -0,101 0,168 Q39 4,00 4,00 5 1,365 -0,587 0,084 -0,641 0,168 Q40 4,25 5,00 5 1,317 -0,845 0,084 -0,055 0,168 Q42 4,47 5,00 5 1,551 -1,062 0,084 -0,007 0,168 Q43 3,90 4,00 5 1,301 -0,644 0,084 -0,460 0,168 Q44 2,98 3,00 1 1,614 0,226 0,084 -1,335 0,168 Q47 4,33 5,00 5 1,299 -0,932 0,084 0,183 0,168 Q48 3,91 4,00 5 1,507 -0,598 0,084 -0,777 0,168 Q50 4,13 5,00 5 1,305 -0,784 0,084 -0,241 0,168 Q51 4,62 5,00 5 1,284 -1,055 0,084 0,407 0,168 Q52 3,84 4,00 5 1,367 -0,615 0,084 -0,608 0,168 Q53 4,00 4,00 5 1,469 -0,535 0,084 -0,850 0,168 Q54 4,11 4,00 5 1,421 -0,558 0,084 -0,659 0,168 Q55 3,91 4,00 5 1,483 -0,566 0,084 -0,779 0,168 Q56 3,96 5,00 5 1,461 -0,678 0,084 -0,750 0,168 The data set for the six-point response scale was factor analysed on two levels according to a procedure suggested by Schepers (1992) in order to determine the factor structure of the instrument. The data was analysed using the SPSS programme. To determine the suitability of the inter-correlation matrix for factor analysis, the Kaiser-Meyer-Olkin (KMO) Measure of Sampling adequacy (MSA) and Bartlett’s Test of Sphericity were conducted on the matrix. The KMO yielded a MSA of 0,938 and the Barlett Test a Chi-square of 12571 (p = 0,000). The matrix is suitable for further factor analysis. The items of the CAI were inter-correlated and the eigenvalues of the unreduced matrix calculated. Due to the limited space, the inter-correlation matrix is not reported here. The number of factors postulated according to Kaiser’s (1970) criterion (eigenvalues greater that unity) are reported in Table 20. These six factors explain about 52% of the variance in the factor space. The item loadings on the six postulated factors are presented in Table 21. The loadings are reported in bold type for each of the factors. Only five factors yielded significant item loadings. RESPONSE SCALE FORMATS 71 Factor five was non-determined, having only one significant loading. Only items with values greater than 0,3 were included in this unreduced matrix. TABLE 20 EIGENVALAUES ON THE ITEM INTER-CORRELATION MATRIX (38 X 38) Initial eigenvalues Root Eigenvalue % of variance Cumulative % 1 10,666 28,068 28,068 2 3,644 9,589 37,657 3 1,770 4,657 42,314 4 1,483 3,903 46,216 5 1,158 3,048 49,264 6 1,087 2,861 52,125 7 0,997 2,622 54,747 8 0,908 2,391 57,138 9 0,892 2,347 59,485 10 0,849 2,233 61,718 11 0,807 2,124 63,842 12 0,781 2,055 65,897 13 0,735 1,934 67,830 14 0,729 1,918 69,748 15 0,682 1,796 71,544 16 0,668 1,758 73,302 17 0,657 1,729 75,031 18 0,638 1,680 76,711 19 0,625 1,646 78,357 20 0,581 1,529 79,886 21 0,562 1,478 81,364 22 0,537 1,412 82,776 23 0,524 1,378 84,154 24 0,517 1,360 85,514 25 0,500 1,315 86,829 26 0,478 1,258 88,088 27 0,463 1,218 89,306 28 0,447 1,177 90,483 29 0,430 1,131 91,614 30 0,427 1,123 92,738 31 0,396 1,043 93,78 32 0,387 1,017 94,798 33 0,364 0,958 95,756 34 0,361 0,949 96,705 35 0,341 0,897 97,602 36 0,321 0,845 98,447 37 0,310 0,817 99,264 38 0,280 0,736 100,000 Trace = 38 Sub-scores on each of the five postulated factors were calculated by adding item scores. These sub-scores were again inter- correlated and the results are portrayed in Table 22. A KMO test for sampling adequacy and a Bartlett’s Test of sphericity was performed to test the suitability of this matrix for factor analysis. The KMO yielded a MSA of 0,662 and Bartlett’s Test a Chi-square of 939,17 (p = 0,000). The matrix is suitable for further factor analysis. Eigenvalues were again calculated on this unreduced inter- correlation matrix. Two factors were postulated according to Kaiser’s (1970) criterion (eigenvalues greater that unity) that account for about 70% variance in factor space. The results appear in Table 23. TABLE 21 ITEM LOADINGS ON SIX POSTULATED FACTORS (6X6) Items 1 2 3 4 5 6 55 0,727 26 0,719 56 0,669 17 0,662 33 0,658 25 0,646 16 0,642 32 0,611 18 0,567 30 0,531 0,327 21 0,527 48 0,464 0,419 4 0,463 52 0,461 0,345 42 0,444 14 0,441 44 0,441 31 0,427 0,344 0,360 20 0,336 0,302 0,087 54 0,691 53 0,652 50 0,607 51 0,591 40 0,584 38 0,563 47 0,533 39 0,328 0,439 15 0,424 43 0,355 0,380 9 0,380 5 0,686 6 0,478 3 0,472 10 0,645 11 0,375 0,574 12 0,568 2 0,342 -0,385 34 0,335 0,516 TABLE 22 INTER-CORRELATION MATRIX OF SUB-SCORES ON FIVE POSTULATED FACTORS (5 X 5) 1 2 3 4 5 1 1,000 0,522 0,516 0,026 0,342 2 0,522 1,000 0,301 0,356 0,452 3 0,516 0,301 1,000 -0,30 0,194 4 0,26 0,356 -0,30 1,000 0,313 5 0,343 0,452 0,194 0,131 1,000 All correlation are significant ate the p = 0,05 level. Two factors were extracted using principal axis factoring. The loadings on the sub-scores appear in Table 24. SMITH, ROODT72 TABLE 23 EIGENVALUES ON THE SUB-SCORE INTER-CORRELATION MATRIX (5 X 5) Initial eigenvalues Root Eigenvalue % of variance Cumulative % 1 2,273 45,450 45,450 2 1,210 24,202 69,652 3 0,620 12,401 82,053 4 0,538 10,754 92,807 5 0,360 7,193 100,000 Trace = 5 The scores were inter-correlated and the results portrayed in Table 24. Factor 2 is non-determined. In order to create an equal base for comparison, the factor analysis was forced into a single factor solution. TABLE 24 SUB-SCORES ON THE SECOND LEVEL FACTOR Factors loadings Sub-scores 1 2 1 0,889 0,278 2 0,582 0,113 3 0,564 0,662 4 0,002 0,646 5 0,368 0,547 One factor was extracted using principal axis factoring. The loadings of the sub-scores on a single factor appear in Table 25. TABLE 25 FACTOR CORRELATION MATRIX Factor Factor 1 2 1 1,000 0,272 2 0,272 1,000 Iterative item analyses were conducted on the single obtained scale. The item-test correlations as well as the test reliabilities (Cronbach Alpha) with the respective item deleted appear in Table 26. The obtained single scale yielded a Cronbach Alpha of 0,9273. TABLE 26 SUB-SCORE LOADINGS ON THE SECOND LEVEL FACTOR Factor Communalities Sub scores 1 1 0,806 0,649 2 0,686 0,470 3 0,527 0,278 4 0,493 0,243 The item reliability statistics are reported in Table 27. TABLE 27 ITEM RELIABILITY STATISTICS FOR THE SIX-POINT SCALE Item Scale mean if Scale variance if Corrected item Alpha if item item deleted item deleted Total correlation deleted 55 126,0412 700,8758 0,6000 0,9241 26 126,2941 694,7944 0,6428 0,9235 56 125,9918 704,6654 0,5593 0,9246 17 126,5682 699,8946 0,6190 0,9239 33 126,4800 695,5196 0,6239 0,9237 25 126,4424 701,5756 0,5763 0,9243 16 126,3512 696,2085 0,5889 0,9241 32 126,6859 699,5278 0,5975 0,9241 18 126,4235 696,9912 0,5623 0,9245 30 126,3612 695,7599 0,6357 0,9236 21 126,6929 701,3131 0,5484 0,9247 48 126,0424 702,9287 0,5630 0,9245 4 126,1718 704,1236 0,5355 0,9248 2 126,1141 704,5347 0,6032 0,9242 42 125,4812 712,8766 0,4213 0,9262 14 126,3612 707,3829 0,4832 0,9255 44 126,9694 707,1062 0,4714 0,9256 31 126,6376 705,8709 0,5227 0,9250 20 126,1776 712,0874 0,4635 0,9256 54 125,8376 718,1314 0,3943 0,9264 53 125,9471 720,3329 0,3509 0,9269 50 125,8224 713,2487 0,5054 0,9252 51 125,3341 715,6974 0,4779 0,9255 40 125,7012 722,9565 0,3597 0,9267 38 125,7165 716,3706 0,4462 0,9258 47 125,6224 715,6464 0,4723 0,9256 39 125,9541 709,2146 0,5379 0,9249 15 125,6212 718,9824 0,4128 0,9262 43 126,0541 711,4411 0,5337 0,9249 9 125,8129 716,4868 0,3930 0,9265 5 126,3376 715,0508 0,3813 0,9267 6 126,3753 708,5851 0,4440 0,9260 3 125,7694 721,1694 0,3424 0,9270 4 126,1859 713,1715 0,4230 0,9262 No of cases = 850 No of items = 34 Cronbach alpha 0,9273 A Cronbach Alpha 0,9273 was calculated. The Alpha values if an item is deleted are of the order of 0,92 for all the items. Cortina (1993) indicated that a Cronbach Alpha of 0,7 and more is significant. DISCUSSION Comparative findings and discussion of the four, five and six- point response scales The study set out to identif y which response scale would provide the best metric properties on the CAI. The item statistics for the respective response scales indicate that most of the items were negatively skewed indicating that most of the respondents were in agreement with the statements posed in the CAI. The factor structures are similar, with a single factor extracted in each of the respective scales, except for the six-point scale where a single factor solution was forced. Range ranges of the item statistics are reported in Table 28. RESPONSE SCALE FORMATS 73 The Alpha values of the respective response scales all have high reliabilities that exceed 0,7. Values more than 0,7 are significant (Cortina, 1993). The literature makes no further distinction regarding the significance of values that exceed 0,7 making further interpretation difficult. The column where the ranges have larger differences is limited to the ranges of corrected items and total correlation. Here the ranges are wider indicating larger differences between the lowest and the highest values in terms of the correlated values. The highest values for the three scales do not indicate major differences. The values for the four and six-point response scales are close, while the largest difference on the low score is for the five-point scale. The differences are attributed to the characteristics of the sample populations completing the CAI and the effects of the using ordinal scales. The differences in the sample populations include: � Data collected in different organisations that are not comparable in terms of core business; � Home language of most of the respondents was not English. The majority of respondents have indicated that their first language is not English. Where items are not clear or ambiguous, respondents are likely to respond in the affirmative (Greenleaf, 1992; Mda, 2000). � Standard biographical data is not available for all the response scales. In the six-point response format there was a significant number of respondents that have lower levels of qualifications. Low levels of qualified respondents are likely to respond at the either extreme ends of response scales (Backman & O’Malley, 1984; Greenleaf, 1992). � Ages of respondents varied across the data sets. Older respondent are likely to respond at the extremes of response scales (Greenleaf, 1992). � Comparisons of the reliabilities of the respective scales indicate differences, but the overall differences are not significant. The lack of significant differences between the Alphas is attributed to differences in the sample populations that completed the CAI and the effects of using ordinal scales. The response format in the CAI was a combination ordinal scale with statements. Schepers (1992) reported that a combination of ordinal scale with statements affects responses. � Another possible factor could explain the results of the Alphas. This requires a re-examination of the Cronbach Alpha formula. This formula suggests when VX is restricted (as in the case of the four-point scale) the obtained coefficient Alpha would increase. This may explain the slightly higher coefficient for the four-point scale. The fact that ordinal scales were used with all three-scale formats may be the reason for the relative small differences in overall reliabilities. � The items included in the instrument do not indicate any unusual practice in business. Most businesses engage in the practices reflected in the items. The nature of the items would be typical practice in most organisations, hence the affirmative responses by the respondents to most of the items in all three of the response scales. � In the design of items the instrument employs statement-type items on a five-point, Likert type scale. Schepers (1992) indicates that there is a high likelihood for respondents to engage in acquiescence bias where items are of the statement linked to the Likert t ype labelled response points. Participants are less likely to engage in the items where statement type items are included (Petty et al., 1987). � James (1982) and Glick (1985) argue that when doing organisational research that requires respondents to respond in terms of their opinion and perceptions, the perceptions relate to ambiguity autonomy influence, facilitation support and warmth (James, 1982). There also exists possibility that two levels of research result. At the one level is the organisation, while at another level is the respondent’s psychological level. This creates two units of measurement (James 1982; & Glick, 1985). Glick (1985) further argues that the literature makes a distinction between organisational and psychological unit concepts. The levels should therefore be treated as different levels of measures in organisational research. Organisational culture included a variety of psychological variables. The present study may have run into the dilemma of mixing the psychological with the organisational units of measurement. While the postulates proposed have not been confirmed, the study has again highlighted aspects that require attention in culture instrument results interpretation. Aspect that requires attention include: � The comparison of the respective scales would only be meaningful once organisation level units and psychological level units are clearly defined and accommodated in the research design (James, 1982). � The paucity of research relating response styles in the South African population. � In the format used, the number of items was limited to 38 items. To justif y the effort, more items need to be developed that would measures at the less obvious levels of organisational culture. The instrument should then be used where organisations are comparable in terms of core business, and sample characteristics. Once the many of the extraneous variables are controlled, the possibility exists to make meaningful comparisons of the respective scales (Klein & Kozlowski, 2000). � Instrumentsshould be developed taking a question format combined with an intensity scale. This is likely to control biased responses and improve the metric properties of the instrument (Schepers, 1992). Limitations of the research include � The effects of organisation, age, education, gender language, race and income levels were not considered in the data analysis which may have allowed for more comparable data to be extracted. � The lack of standard biographical data across all the scales did not allow for comparisons to be made or for comparisons within the samples of the respective response scales. � The results were based on an instrument of only 38 items that measure organisational culture at a superficial level. A more robust measure with multiple levels of culture (Schein, 1985) may have produced more significant results. TABLE 28 RANGES OF ITEM RELIABILITY STATISTICS FOR THE FOUR, FIVE AND SIX POINT SCALES Response Range of scale means Range of scale variances Range of corrected Range of Alpha scores Cronbach scale if item deleted if item deleted items – total correlation if item deleted Alpha Low High Low High Low High Low High Four-point 96,5709 97,9360 361,9862 396,5332 0,3778 0,6512 0,9315 0,9338 0,9345 Five-point 119,9764 120,9501 540,7938 561,1290 0,1015 0,6662 0,9212 0,9247 0,9248 Six-point 125,3341 126,9694 695,5196 718,1314 0,3424 0,6428 0,9235 0,9269 0,9273 SMITH, ROODT74 Emanating from this study, the following research is suggested: � To investigate the true effects of response formats (that is four, five and six-point response formats within similar samples and organisations. � Investigating the wording effects on response scales on the South African population. � Establish the effects of changes in wording on response scales, positive to negative and negative to positive and the relationship to response styles. � Specific effects of age and education, language, gender, income and race in response sets and styles. � The effect of changing the response positions i.e., changing the ‘don’t know’ from the middle positions to the end of the response options where Likert type response scale are required. � To research the racial differences, the effects of socio-economic and educational levels of respondents in relation to response bias. � What the effect would be on the metrics of instruments by re- designing the response scale in the intensity scale and question format. � Evaluate the Culture Assessment Instrument and the different response scales using data from similar organisations with comparable sample. ACKNOWLEDGEMENT The authors are indebted to Prof Nico Martins of Organisational Diagnostics for providing the data set for conducting this research. The 56-item CAI used in this research was abstracted from a larger item pool collected across various organisations and does not constitute the original Organisational Culture Survey of Prof Martins. Data collected from each organisation by means of the Survey was suplemented by qualitative methods. The authors would like to thank Riëtte Eiselen and her staff from Statcon at RAU for their professional service and valuable contribution regarding the data analysis for this project. REFERENCES Allen, M.J. & Yen, W.M. (1979). Introduction to measurement theory. Monterey, CA: Brooks/Cole. Alvesson, M. (2002). Understanding organisational culture. London: Sage. Anastasi, A. (1986). Evolving concepts of test validation. Annual Review of Psychology, 37, 1-15. Anastasi, A. & Urbani, S. (1997). Psychological testing. (7th edition). Upper Saddle River, N J: Prentice Hall. Ashkanasy, N.M., Broadfoot, L E. & Falkus, S. (2000).Questionnaire measures of organisational culture. In Ashkanasy, N.M. Wilderom, C.P.M. & Pearson, M. F. (Eds). Handbook of organisational culture and climate. Thousand Oaks: Sage. Ashkanasy, N.M., Wilderom, C.P.M. & Pearson, M.F. (2000). (Eds.). Handbook of organisational culture and climate. Thousand Oaks: Sage. Babbie, E. & Mouton, J. (2001). The practice of social research. (South African edition) Cape Town: Oxford University Press. Bachman, J.G. & O’Malley, P.M. (1984). Yea-saying, nay-saying, and going to extremes: black-white differences in response styles. Public Opinion Quarterly, 48, 491-509. Bluedorn, A.C. (2000). Time and organisational culture. In Ashkanasy, N.M., Wilderom, C.P.M. & Pearson, M.F. (Eds.). Handbook of organisational culture and climate. Thousand Oaks: Sage. Bradburn, N.M., Sudman, S., Blair, E. & Stocking, C. (1978). Question threat and response bias. Public Opinion Quarterly, 42 (2), 221-234. Brown, F.G. (1983). Principles of educational and psychological testing. (3rd edition). New York: Holt, Rinehard & Winston. Church, A.H. & Waclawski, J. (1998). Designing organisational sur veys. Aldershot: Gower. Collins dictionary and thesaurus in one volume (1987). London; London. Cortina, J.M. (1993). What is coefficient alpha? An examination of the theory and applications. Journal of Applied Psychology, 78 (1), 98-104. Cronbach, L.L. (1990). Essentials of psychological testing. (5th edition). New York: Harper Collins. Cummings, T. (2001). Culture and change. In: C.L Cooper, S. Cartwight & P.C. Earley, (Eds.). The international handbook of organisational culture and climate. Chichester: John Wiley. Denision, D. (2001). Organisational Culture: Can it be a key lever for driving culture change. In C. L Cooper, S., Cartwight & P.C. Earley (Eds). The international handbook of organisational culture and climate. Chichester: John Wiley. Dove, R. (2001). Response ability: The language, structure, and culture of the agile enterprise. New York: John Wiley. Duncan, O.D. & Stenbeck, M. (1988). No opinion or not sure. Public Opinion Quarterly, 52, 513–525. Flamholtz, E. (2001). Corporate culture and the bottom line. European Management Journal, 19 (3), 268-275. Feigl, H. (1970). The orthodox view of theories: Remarks in defence as well as critique. In M. Radner, & S. Winokurs, (Eds.). Analysis of theories and methods of physics and psychology. Minneapolis: Universit y of Minnesota Press. Fife-Schaw, C. (2000). Questionnaire design. In G.M. Breakwell, S. Hammond & C. Research methods in psychology. (2nd edition). London: Sage. Fink, A. (1995a). The sur vey handbook. Thousand Oaks: Sage. Fink, A. (1995b). How to ask sur vey questions. Thousand Oaks: Sage. Fleick, L. (1989). Latent class analysis of survey questions that include don’t know responses. Public Opinion Quarterly, 53, 525–547. Fowler, F.J. (1992). How unclear terms affect survey data. Public Opinion Quarterly, 56, 218-231. Francis, J.D. & Busch, L. (1975). What we know about “I don’t knows”. Public Opinion Quarterly, 39 (2), 207-218. Gaskill, G.D., O’Muircheartaigh, C.A. & Wright, D.B. (1994). Survey questions about the frequency of vaguely defined events: The effect of response alternatives. Public Opinion Quarterly, 58, 241-254. Ghiselli, E. E., Campbell, J.P. & Zedrick, S. (1981). Measurement theory of the behavioural sciences. San Francisco: WH Freeman. Gilljam, M. & Granberg, D. (1993). Should we take don’t know for an answer? Public Opinion Quarterly, 57, 348–357. Glick, W.H. (1985). Concept ualising and measuring organisational and psychological climate: Pitfalls in multilevel research. Academy of Management Review, 10 (3), 601-616. Greenleaf, E.A. (1992). Measuring extreme response style. Public Opinion Quarterly, 56, 328-351. Gregory, R.J. (1996). Psychological testing: History, Principles and Applications. (2nd edition). Boston: Ally and Bacon. Guilford,J.P. (1954). Psychometric methods. New York: McGraw- Hill. Hair, J.S., Anderson, R.E., Tatham, R.L. & Black, W.C. (1998). Multivariate data analysis. (5th edition). Upper Saddle River, N J: Prentice-Hall International. Hatch, M.J. (1997). Organisational theory: Modern, Symbolic, and Post-modern Perspectives. Oxford: Oxford University Press. Hofstede, G. (2001). Cultures and consequences: Comparing values, behaviours institutions and organisations across nations. (2nd edition). Thousand Oaks: Sage. Jackson, D.M. & Messick, S. (1967). Response styles and the assessment of psychopathology. In D.N. Jackson & S. Messick (Eds.). Problems in human assessment. New York: Robert E Krieger. James, L.R. (1982). Aggregation bias in estimates of perceptual agreement. Journal of Applied Psychology, 67 (2), 219-229. RESPONSE SCALE FORMATS 75 Kaiser, H.F. (1970). A second-generation Little Jiff y. Psychometrika, 35, 401-415. Kerlinger, F.N. & Lee, H.B. (2001). Foundations of behavioural research. (4th edition). Fort Worth: Harcourt College. Klein, K.J. & Kozlowski, S.W.J. (2000). A Multilevel approach to theory and research in organisations: Contextual, temporal and emergent processes. In K.J. Klein & S.W.J. Kozlowski (Eds.). Multile vel theor y, research, and methods in organisations: Foundations, extensions, and new directions. San Francisco: Jossey-Bass. Kotter, J.P. & Heskett, J.L. (1992). Corporate culture and performance. New York: John Wiley. Kroeber, A.L. & Kluckhohn, C. (1952). Culture: A critical review of concepts and definitions. Papers of the Peabody Museum of American Archaeology and Ethnology, (Harvard University), XLVII, No 1. Cambridge, Massachusetts: Peabody Museum. Martins, N. (2000). Organisational culture study. Unpublished manuscript. McIntire, S & Miller, L. A. (2000). Foundations for psychological testing. Boston: McGraw-Hill. Nadler, D.A., Thies, P.K. & Nadler, M.B. (2001). Culture change in the strategic enterprise: Lessons from the field. In C. L Cooper, S. Cartwight & P.C. Earley (Eds.). The international handbook of organisational culture and climate. Chichester: John Wiley. Newstead, S.E. & Arnold, J. (1989). The effect of response format on ratings of teachers. Educational and Psychological Measurement, 49, 33-43. Nunnally, J.C. & Bernstein, I.H. (1994). Psychometric Theory. (3rd edition.) New York: McGraw-Hill. O’Muircheartaigh, C.A., Gaskill, G.D. & Wright, D.B. (1993). Intensifiers in behavioural frequency questions. Public Opinion Quarterly, 57, 552-565. Ott, J.S. (1989). The organisational culture perspective. Chicago: Dorsey Press. Ouchi, W.G. (1981). Theory Z. Reading, MA: Addison-Wesley. Peters, T.J. & Waterman, R.H. (1982). In search of excellence: Lessons of from America’s best-run companies. New York: Harper & Row. Petty, R.E., Rennier, G.A. & Cacioppa, J.T. (1987). Assertion versus interrogation formats in opinion surveys: Questions enhance thoughtful responding. Public Opinion Quarterly, 51, 481-494. Poe, G.S., Seeman, I., McLaughlin, J., Mehl, E. & Dietz, M. (1988). “Don’t know” boxes in factual questions in a mail questionnaires: Effects on level and quality of response. Public Opinion Quarterly, 52, 212-222. Rubin, A. & Babbie, E. (1997). Research methods for social work. Pacific Grove, CA: Brooks/Cole. Rust, J. & Golombok, S. (1999). Modern psychometrics: The science of psychological assessment. (2nd edition). London: Routledge. Sanchez, M.E. & Morchio, G. (1992). Probing “don’t know” answers: Effects on survey estimates and variable relationships. Public Opinion Quarterly, 56, 454-474. Sarantakos, S. (1998). Social research. (2nd edition). South Yarra: MacMillan. Schein, E .H. (1985). Organisational culture and leadership. San Francisco: Jossey-Bass. Schein, E.H. (1999). The corporate sur vival guide: Sense and nonsense about culture change. San Francisco: Jossey-Bass. Schepers, J.M. (1992). Toetskonstruksie: Teorie en Praktyk. Johannesburg: Rand Afrikaans University. Schultz, M. (1995). On studying organisational cultures. Berlin: Walter de Gruyter. Smith, E.R.A.N. & Squire, P. (1990). The effects of prestige names in question wording. Public Opinion Quarterly, 54, 97-116. Swart, C., Roodt, G. & Schepers, J.M. (1999). Itemformaat, differensiële itemskeefheid en die faktorstruktuur van ’n selfvoltooingsvraelys. SA Tydskrif vir Bedryfsielkunde, 25 (1), 33-43. Torgerson, W.S. (1958). Theory and methods of scaling. New York: John Wiley. Trice, H.M. & Beyer, J.M. (1984). Studying organisational culture through rites and ceremonials. Academy of Management Review, 8 (4), 653-669. Van der Post, W.Z., de Coning, T.J. & Smit, E.v.d.M. (1997). An instrument to measure organisational culture. South African Journal of Business Management, 28 (4), 147-168. Van der Post, W.Z., de Coning, T.J. & Smit, E.v.d.M. (1998). The relationship between organisational culture and financial performance: some South African evidence. South African Journal of Business Management, 29 (1), 30-40. Welman, J.C. & Kruger, S.J. (2001). Research methodology for business and administrative sciences. (2nd edition). Cape Town: Oxford University Press. Wilson, A.M. (2001). Understanding organisational culture and the implications for corporate marketing. European Journal of Marketing, 35 (3/4), 453-567.