Scaling and Measurement Essay Example
Scaling and Measurement Essay Example

Scaling and Measurement Essay Example

Available Only on StudyHippo
View Entire Sample
Text preview

Comparing Alternative Measures of the French and Raven

Power Bases John T. Drea, Gordon C. Bruner II and Paul J. Hensel Twenty-five multi-item scales used to measure power sources anchor exercised power in a sales or channel setting are reviewed. The procedures for assessing the reliability and validity (convergent, discriminant, and nomological) of each scale are discussed and reviewed. Findings indicate adequate reliability for all ofthe scales. Problems in the establishment of validity for some ofthe coercive and legitimate power measures are noted.

Recommendations are made regarding the use and improvement of power scales in future research. The measurement of power is central to understanding the behavior of organizations and individuals. Power is commonly defined as "the ability of one individual or group to control or influence the behavior of another" (Hunt and Nevin 1974). Power has served as an important construct because of

...

its hypothesized relationship to other variables such as satisfaction, role performance, and conflict. However, despite its importance, measures of power have received relatively little critical comparison and analysis.Additional research is needed into creating and testing valid and reliable multiitem scales for measuring power (Podsakoff and Schriesheim 1985).

The purpose of this article is to examine alternative Likert-type summated ratings scales purported to measure the French and Raven (1959) bases of social power. The convergent, discriminant, and nomological validities of each ofthe power constructs will be examined using the criteria proposed by Peter (1979,1981), ChurchiU (1979; and Peter 1984), Gerbing and Anderson (1988), and Bagozzi and Yi (1991).Conclusions are drawn regarding the reliability and validity of each scale, and recommendations are made regarding the most sound measures as well as future scale development. Methodology A

View entire sample
Join StudyHippo to see entire essay

review was conducted of more than 600 articles published in Journal of Marketing Research, Journal of Marketing, Journal of the Academy of Marketing Science, and Industrial Marketing Management for a ten year period from 1980 to 1989. These journals were selected for thorough review because they were the most well known sources of scholarly research investigating the power construct during the decade of the 1980s.

Additionally, selected influential articles on the power construct which appeared in other sources (Journal of Applied Psychology and the Journal of Personal Selling and Sales Management) were also collected. As a result of this search process, seven articles which used multi-item power source scales based on French and Raven (1959) were identified. Some ofthe studies measured the perception of power sources using the five original sources of power (John 1984; Comer 1984; Michie and Sibley 1985; Kohli 1989; Hinkin and Schriesheim 1989), while others measured the exercise of power (Gaski and Nevin 1985; Gaski 1986).Each of the power scales reviewed used a multiitem scale to measure one of the five sources of power delineated by French and Raven (1959).

Measures excluded from the present analysis included Journal of Personal Selling & Sales Management, Volume Xin, Number 4 (Fall 1993), John T. Drea (DBA candidate. Southern Illinois University) is the Assistant Dean of Instruction at John Wood Community CoUege in Quincy, Illinois. Gordon C. Bruner II (Ph. D. , University of North Texas) is Associate Professor of Marketing at Southern Illinois University and Director of the Office of Scale Research. Paul J. Hensel (Ph.D. , University of Houston) is Professor of Marketing at the University of New Orleans.

74 Journal of Personal Selling &

Sales Management all single item measures (see Podsakoff and Schriesheim 1985 for a review of these), and measures which separated anchor combined the sources of power differently than in the original five bases of social power (e. g. , Schul, Remington, and Berl 1990). Results Reliability In the present research, reliability is defined as "the degree to which measures are free from error and therefore yield consistent research" (Peter 1979, P-6). Procedures for Assessing Reliability.Of the 25 power scales reviewed, 24 reported evidence of reliability through the use of Cronbach's alpha.

This procedxire provides a relatively simple and well accepted assessment of the internal consistency of a given measure. Potential Influences on Scale Reliability. Churchill and Peter (1984) have indicated that sample size, the number of scale items, and the number of scale points can potentially infiuence the reliability of a given scale. Among the articles reviewed, sample sizes (where Cronbach's alpha was also reported) ranged from 147 to 251, with the exception of Hinkin and Schriesheim (1985), who used one sample with 375 subjects.

The number of scale items ranged between four and ten, with the only exception being a 15 item scale used by Gaski (1986; Gaski and Nevin 1985) to measure perceived and exercised reward jjower. The number of scale points ranged from four to eleven. The relatively narrow range of sample sizes, scale items, and scale points used with each scale indicates that these factors alone are unlikely reasons for any given scale to perform more reliably in comparison to any other scale.To verify this, a correlational analysis of the 24 scales reporting reliability coefficients was conducted to examine any potential correlations between

sample size, nvimber of scale items and number of scale points. The results indicated very weak and non-significant.

Convergent and Discriminant Validity

Churchill has defined convergent validity for a given measure as "the extent to which it (the measure) correlates highly with other methods designed to measure the same construct," while defining discriminant validity as "the extent to which the measure is indeed novel and not simply a refiection of some other variable" (Churchill 1979, p. 0). Procedures for Assessing Convergent and Discriminant Validity. One of the common techniques used to assess convergent and/or discriminant validity among the scales examined was exploratory factor analysis (John 1984; Comer 1984; Michie and Sibley 1985; Hinkin and Schriesheim 1989; Kohli 1989).

While exploratory factor analysis has proved useful for reducing the number of scale items and examining the relatedness among items, it lacks an explicit test of the unidimensionality provided by confirmatory factor analysis (Gerbing and Anderson 1988).Of the scales reviewed, only Hinkin and Schriesheim (1985) and Gaski (1986) appear to have used a confirmatory procedure to assess convergent or discriminant validity. The scales used by Gaski (1986; Gaski and Nevin 1985) made use of portions of the multitraitmultimethod (MTMM) approach to assess convergent and discriminant validity as outlined by Campbell and Fiske (1959). As an additional check of discriminant validity, Gaski (1986) used LISREL in a procedure similar to one suggested by Bagozzi and Phillips (1982) to examine whether the five power source constructs were perfectly correlated.Convergent validity was assessed for the reward and coercive power measures by correlating each with an alternative measure of the same construct.

Confirmatory factor analysis was also used by FaU1993 75

Hinkin and Schriesheim (1989) to

assess the imidimensionality and convergent validity of each the five scales. Exploratory factor analysis was used to assess discriminant validity by determining whether items from the five pnjwer bases would cross load on non-jx;gt;wer base items (satisfaction and commitment). Exploratory factor analysis was similarly used by John (1984) to assess convergent and discriminant validity.The Attributed Power Index (API) scales, originally developed by Holzbach (1974), were subjected to the full multitrait-multimethod validation procedure by Comer (1984). Correlations were calculated between each of the multi-item measures; between each of the single item and alternative method measures; and between the single and multi-item measures.

Reliabilities were also calculated for each multi- item measure. Selection of Alternative Methods. A typical problem encountered when attempting to establish convergent validity involves the selection of appropriate "alternative" methods.Use of the MTMM matrix requires that methods must be maximally different to avoid the problems of shared method and method- trait variance (Peter 1981).

While estimates of method variance can be calculated using procedures such as path analysis or confirmatory factor analysis, Peter (1981) has ? u-gued that "different forms of paper and pencil self-rating scales are clearly not maximally different methods" (p. 137). The use of such similar methods is more likely to produce an assessment of alternative form reliability than convergent validity (Cohen 1979).Both Comer (1984) and Gaski (1986; Gaski and Nevin 1985) made use of alternative measures. The alternative scales used by Comer (1984) were single item power measures developed originally by Busch (1980). These single item scales were administered in the same manner as the multi-item API measvire: a paper and pencil survey mailed to sales representatives.

The alternative measurement instrument used by

Gaski and Nevin (1985) consisted of a series of Likert-type items which were similar to the primary power measures.The alternative measures used by Hinkin and Schriesheim (1989) appear to be a rejjeated use of a single measure with multiple samples, rather than true alternative measures. Claims of convergent validity for these power scale measures should be viewed with caution. Convergent and Discriminant Validity Results.

Gaski (1986; Gaski and Nevin 1985) claimed evidence of discriminant validity for each of hia five power source scales. The alpha coefficient for each measure was greater than the correlation of the measure with other measures, leading to the claim of discriminant validity.The confirmatory factor analysis procedure used by Gaski (1986) also indicated that the five power source constructs were not perfectly correlated with one another, and that discriminant validity for each of the five power source measures was achieved. Convergent validity was asserted by Gaski and Nevin (1985) for their measures of reward and coercive power on the basis of positive and significant correlations with alternative measures of the same construct. However, no alternative measures were reported for the legitimate, referent, and expert power source measures by Gaski (1986).As indicated earlier.

Comer (1984) used the full MTMM procedure to assess validity. His findings failed to establish convergent and discriminant validity for the coercive, reward, and legitimate power source scales but it was largely satisfied for the referent and expert power source scales. The claim of discriminant validity for the scales used by Hinkin and Schriesheim (1989) was supported by factor loadings, indicating that conceptually distinct measures (satisfaction, commitment, and power) loaded on different factors, with no inappropriate loadings.The results provided some support

for a claim of convergent validity since most items loaded on a single factor. However, using ChurchiU's definition, the testing of a scale through the use of multiple samples (as opposed to comparing alternative measures of the same construct) appears to be more a measure of reliability than convergent validity.

John (1984) provided limited evidence of convergent and discriminant validity. Factor analysis was used to determine a one factor model fit to the data for each scale. However, factor loadings were not reported.Studies by Kohli (1989) and Michie and Sibley (1985) provided rather limited evidence of convergent and discriminant validity for referent and expert power (Kohli) as well as coercive power (Michie and Sibley). Exploratory factor analysis loadings were uniformly high in each study, . 59 or greater for each item in the scale, with no item loading 76 Journal of Personal Selling & Sales Management higher than 32 on any other factor. However, the measures used in both studies lacked alternative measures which could have provided more meaningful assessment of convergent and discriminant validity. lso suggested that there should be a negative relationship between coercive power and expert, referent, and legitimate power. The table presents the reported correlations between each of the power sources. Nomological Validity Results.

Claims of Nomological Validity The assessment of nomological validity "entails investigating both the theoretical relationship between different constructs and the empirical relationship between measures of those different constructs" (Peter 1981, p. 135). Procedures for Assessing Nomological Validity. Nomological validity was specifically investigated in 22 of the 25 power scales reviewed.Of these 22 investigations, all provided some positive evidence of nomological validity by correlating the given power measure with another

measure related by theory.

The most common procedure for assessing nomological validity among the scales reviewed was to correlate one or more measures of satisfaction with each power source measure (Comer 1984; Gaski and Nevin 1985; Gaski 1986; Hinkin and Schreisheim 1989). While satisfaction has been commonly used as a construct for assessing the nomological validity of power scales, the operationalization of the satisfaction construct has taken many different forms.For example, Gaski and Nevin (1985) defined satisfaction in terms of the degree to which a dealer is pleased to do business with a supplier. In contrast. Comer (1984) measured seven dimensions of satisfaction for sales representatives.

Since these meastires and others assess different interpretations of satisfaction, a different means for assessing nomological validity was needed. For this reason, correlations between the power sovirce measures were collected and einalyzed.The expected pattern of such correlations has been previously suggested by Raven and Kruglanski (1970), who reported that there should be positive relationships between the variables of expert, legitimate, and referent power sources. The basis for these propositions is that expert, referent, and reward power sovurces are considered to rely on the same mediating processes. Subsequent models of power by Himt and Nevin (1974) and Gaski (1986), while differing in some important respects, have included these three power soiux«s as dimensions of a larger, underlying trait. Raven and Kruglanski (1970) have omological vedidity are supported for most of the measures, though not tdl.

The referent, expert, and legitimate power scales used by Comer (1984), John (1984), and Gaski (1986) correlated with one an" * other in a predictable pattern, with all correlations being positive and strongly significant. The consistency of these

findings supports the claims of nomological validity for each of these scales. Most of the reward power source scales were also found to be correlated in a predictable pattern. Each reward scale correlated positively and significantly (r= . 20 to . 46) with the other scales.

The exception was the reward scale used by Comer (1984), which was correlated more strongly with referent power (r=. 73) and expert power (r=. 57). With regard to coercive power measures, Gaski (1986; Gaski and Nevin 1985) has noted that the exercise of a coercive power source produces a different effect on reward power than on other power sources. Specifically, they pointed out that the correlation between reward power sources and coercive power sources (both perceived measures) is expected to be positive, while a negative correlation is expected between reweird power sources and exercised coercive power sources.Accordingly, the effects of the exercised coercive power source measure on reward power soxirces must be viewed as a different relationship than the effect of one power source on another.

The coercive power measure used by Comer (1984) is a part of Holzbach's (1974) API index and performed as expected when correlated with reward (r=. 22) and legitimate (r= -. 12) power sources. How- ^ ever, the API-based coercive power measure displayed a very low and positive correlation (r=. O2) with expert power source, when a negative correlation as expected.

In addition, the coercive power measvire used by Comer (1984) was correlated positively with the referent power soiurce (r=. 25), when a negative correlation was expected. Accordingly, the nomological validity claim of this coercive power scale cannot be confirmed on the basis of correlations with other

power sovirce scales. # The legitimate power source scale reported by John (1984) correlated as expected with reward, referent, and expert power sources. Sure displayed a very low and positive correlation (r=.Ol) with a coercive power soiurce measure, when a negative correlation was expected. Given this, the nomological validity claim of this particular legitimate power scale should be viewed cautiously. As expected, the legitimate power scale used by Hinkin and Schriesheim (1985) correlated positively with reward, referent, and expert power and with satisfaction, thereby supporting the nomological validity of the scale. However, the legitimate power measure did not correlate as expected with coercive power, yielding significant pwsitive correlations in three samples (r =. 20, 16, and . 1; pected pattern, including the particular dimensions measvired by both measures as well as the reliability of the measures. For these and other reasons, it has been suggested that evidence of nomological validity should include correlations with several measvires that have been previously established as being construct valid and not just one measure (Churchill 1979). The scales used by Comer (1984), John (1984), Gaski (1986; Gaski and Nevin 1985), and Hinkin and Schriesheim (1989) used correlations with several different measures to support claims of nomologic? tl validity.

A pattern of correlations with other established measures would increase the support of nomological validity for these scales. Discussion In general, all of the scales were found to possess at least acceptable levels of reliability. All of the reviewed scales used to measure reward, referent, and expert power had high internal consistency (0. 77 or gn*eater), while all of the scales used to measure legitimate and coercive power sources were generally lower,

though still acceptable. Most of the power scales addressed either directly or indirectly the issues of convergent and discriminant validity.

Some cautions are noted, including the lack of a test of unidimensionality in exploratory factor analysis; the use of alternative measures for validation purposes when such measures are not maximally different; and the use of only part of the total MTMM matrix. The presence of such weaknesses does not refute the convergent and discriminant validity of the scales, but concurrently, indicates that strong evidence of validity is still lacking. Finally, nomological validity was discussed either directly or indirectly for twenty-one of the twenty-five power source scales.Each of these twenty-one scales reported some evidence of nomological validity by correlating the scale with another theoretically related measure, such as satisfaction, power, or other power sources. A subsequent examination of the correlations between power source scales, provided support for claims of nomological validity for the measures of reward, referent, and expert power by each author.

However, the measure of coercive power used by Comer (1984) as well as the legitimate power measures use by John (1984) and Hinkin and Schriesheim (1989) did not relate to other power source measvires in a predictable fashion.The measurement of legitimate power is a specific area of concem. First, scale reliabilities for legitimate power were generally lower in comparison to measures of other bases of power. Second, it is noteworthy that the legitimate power scales used by Comer (1984) and the legitimate power scale originally used by Kohli (1989 ) were both found to possess a bi-dimensional structure.

Kohli later dropp;gt;ed three of the five scale items and re-named the scale "informal legitimate power" (p. 57).

The legitimate power scales used by John (1984) also did not correlate in the expected manner with a measure of coercive power.Finally, the legitimate Fall 1993 79 power scale used by Hinkin and Schriesheim (1989) correlated positively (r=.l6 to . 40) with a measure of coercive power, when a negative correlation was expected. Additional research is needed to re-specify the construct and develop better scales. Bagozzi and Yi (1991) have suggested and reviewed alternative measures of achieving convergent and discriminant validity, including MTMM and confirmatory factor analysis. Among the power measvires reviewed, only Comer (1984) made use of the full MTMM approach. A reduced ersion of the MTMM procedtire was used by Gaski (1986; Gaski and Nevin 1985).

Only Hinkin and Schriesheim (1989) and Gaski (1986) appear to have used confirmatory factor analysis as a primary means for assessing discriminant validity. and Nevin 1985) are the set most recommended for use at this time. Convergent and discriminant validities were claimed and seem to have been supported for each scale. Nomological validity was successfully assessed through the use of confirmatory factor analysis, and inter-correlations among the set of measures were in the direction and magnitude expected.Admittedly, these positive aspects are tempered by the fact that the reliabilities were low but acceptable.

Finally, the use of the modified MTMM approach and alternative measures that were not maximally different suggests that claims of strong construct validity are premature and must still be firmly established.

References

  1. Bagojzi, Richard P. and Lynn W. Phillips (1982).
  2. "Representing and Testing Organizational Theories: A Holistic Construal," Administrative Science Quarterly, 27 (September). 459-489. nd Youjae Yi (1991)
  3. "Multitrait-Multimethod Matrices in Consumer Research," Journal of Consumer Research, 17

(March), 426-439.

  • Burke, R. J.and D. S. Wilcox (1971), 'Bases of Supervisory Power and Subordinate Job Satisfaction," Canadian Journal of Behavioral Sciences, 3, 183-193. Busch, Paul (1980)
  • "The Sales Manager's Bases of Social Power and Influence on the Sales Force," Journal of Marketing, 44 (Summer) 91-101. Campbell, Donald R. and Donald W.Fiske (1959).
  • "Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix," Psychological Bulletin, 56, 81-105.Churchill, Gilbert A. Jr. (1976)
  • Marketing Research: Methodological Foundations, Hinsdale, IL: Dryden. (1979)
  • 'A Paradigm for Developing Better Measures of Marketing Constructs,' Journal of Marketing Research, 16 (February), 64-75. and J. Paul Peter (1984)
  • "Research Design Effects on the Reliability of Rating Scales: A Meta-Analysis," Journal of Marketing Research, 21 (November), 360-375. Cohen, Joel B. (1979)
  • Get an explanation on any task
    Get unstuck with the help of our AI assistant in seconds
    New