The process of selecting factors that predict executive remuneration

In this article I seek to determine those organizational and individual factors which best predict executive rem _ ation. I describe how a pilot study was used to test a variety of factors and to eliminate those which were clear} uner ed. f . . A 1 . · y poor pr 1ctors o execuuve remunerauon. pp ymg muluple regression analysis to the data submitted to an executive salary survey, two. se~ of factors wer_e obtained. _Certain factors individually proved to be better predictors, but it was found that ~ co~bmat10n of f~tors yielded th~ ~1ghest coefficients of determination. The weighting of each factor was then determined m order to derive a market pncmg model for executive remuneration.


Introduction
The method of determining executive remuneration and benefits differs from that of assessing salaries of lower levels within the organization. The salary and benefits data presented in salary surveys for the latter are applicable to organizations whose size and structure may differ significantly. A survey of executive remuneration should present salary and benefits findings according to a wide possible variety of organizational and individual factors that take organizational size and structure into account. With this article I seek to outline how such factors may be selected, how their relative importance may be ascertained, and how they may be used as the basis of an executive salary survey. A limitation of this study is that its purpose was to establish a set of factors for use in a salary survey: although the database was large, it was not practical to test all the many variables suggested in the literature.
The rationale for selecting factors that predict executive remuneration in various salary universes is acknowledged in the literature. In this article the process for selecting, testing and weighting those factors which will best predict executive remuneration are discussed. The weightings of two periods are compared in order to assess how robust the factors are as market conditions change. I then describe how the final selection of correctly weighted factors can be used to develop a market pricing model for determining executive remuneration. 1

Need for detennlnlng factors as the basis of a market pricing model
The essential difference in using market salaries to determine remuneration levels for executives and non-executives is that the salaries of the latter may be compared to the 'going rate' in the market, which is largely independent of the size and structure of the organizations with which the comparisons are made. This is not the case with executives. The study by Kostiuk (1990: 90) poses the fundamental questions as to what determines the level of executive incomes and what is the sensitivity of these levels to the performance of executives. The answer lies in the extent to which organizational size and an executive's contribution to shareholders' interests are linked to executive pay (Main, 1991). Authors such as , and Jackson, Schuler & Rivero (1989) identify industry sector as another factor affecting executive pay as individual industries are constrained by comparable technology, raw materials, economic cycles, and so on.
It is generally accepted in the literature that the larger the firm (measured in a variety of ways), the higher will be the incomes of the executives employed therein.2 This is known as the managerialist explanation of executive pay. The size of organization can be measured in various ways, as will be seen later. The importance of firm size implies that a job evaluated survey of executive remuneration is only meaningful as a basis of comparison between different organizations if it takes organizational parameters into account Of the main systems of job evaluation proposed by Ellig (1983: 47), the ranking, classification and point-factor methods fail when applied to executives because they generally neglect to incorporate those factors which pertain to inter-organizational comparison: firm size and structure. One approach that does incorporate such factors is the market pricing philosophy.
Market pricing begins by taking market salary levels, and then seeks commonalities between positions in the same salary universes. Milkovich & Newman ( 1984: 257) explain market pricing as determining job worth and executive pay by reliance on the external market Within a given salary universe, the market pricing approach isolates the most important factors common to jobs. The positions within each universe are characterized by such factors, which, as directly related to the market, result in a shift in emphasis from determining the relative worth of jobs on factors intrinsic to the job, to market related factors.

Factors predicting executive remuneration
Factors commonly listed as being predictors of executive remuneration include: company sales, assets, profits, stockholders' equity, number of employees, age and length of service of incumbent (Ellig, 1983). To these could be added many others, such as: industry sector, salary and wage bill, type of firm (in the sense of holding or subsidiary company), conlribution to shareholder interests, and so on. Several of these are more commonly encountered than others, although the reasons therefor are varied. Kostiuk (1990: 95) has found age of executive to be 'irrelevant, although experience as CEO is significant'. Ellig (1983) believes that sales or turnover is the most common independent variable used in compensation regression analysis studies. For this reason more than any other, it usually has die highest correlation with executive pay. Less recent, but widely supported research by Roberts, concludes more narrowly: '(the pay of a top executive) is related to the size of the company and to virtually no other variables (in particular, not to profit) ... size of company being measured by sales turnover' (1959: 39). The literature also recognizes that a combination of factors may provide a more accurate prediction of executive remuneration. Kostiuk (1990: 98) states: 'It does seem to indicate that the factors affecting compensation interact in some complex way. ' Burgess (1981: 290) alludes to this when providing evidence that a prediction equation using a combination of variables, including sales, can predict executive pay data more accurately than sales alone. As a further example, the American Consultancy Management Compensation Services (Rock [ed.], 1984: 41, 47) use the following organizational size factors: sales (turnover}, assets and number of employees; the performance measures are: one year return on sales, assets and capital. This effectively means that organi7.ational size may be measured in absolute terms (sales, profits, assets, etc) or in relative terms (return on investment, return on equity, and so on).
Several authors, such as Lazear & Rosen (1981), Bull, Schotter & Weigelt (1987), question whether executive pay is related to individual productivity. Others, such as Main (1991), Cosh & Hughes (1987), provide evidence that executive pay is closely allied to normal (level of performance related to the market) and abnormal (level of performance over and above the reference level) increases in shareholder's wealth, as well as the number of executive and nonexecutive directors.
The relationship between pay and performance is considered by many authors, although their mindings are by no means congruenL A broad summary of findings may take the form of the observation by Jensen & Zimmerman (1985: 1.0) that 'compensation packages of the top level help align managers' and shareholders' interests'. Other authors, while not disputing this general tenet, are rather more demanding: Tosi & Gomez-Mejia have found that: ' ... executives appear to have the best of both worlds. Their basic salaries were functions of firms' size, a relatively stable factor, their long-term incomes were greater when performance was good and the scale of their organi1.Btions provided a downside hedge against poor performance ' (1989: 170). Gomez-Mejia, Tosi & Hinkin (1987) postulate that ownership is an important factor in predicting executive pay: in firms with dominant external shareholdings, performance is S.-Afr. Tydskr.Bedryfsl.1993,24(l) a significant predictor of executive remuneration; whae control lies with management of a firm, size and not per. formance predicts pay.
Another factor that may unduly promote remunerauon which is not based on performance is that of specifying differentials between management levels, as proposed by Simon (1957) and . With this anicle I acknowledge that differentials may explain an element of lbe variation in executive pay, but I do not pursue the maaer.
In this study I seek to consider further the nature of lbe prediction equation, referred to by several authors. Various forms of this equation are found in the literature. A further equation is given by Cosh (1975): log cash remuneration= 3o + a1R + a2A + e where R is the firm rate of return on assets and A is net assets.
The regression equation used in this study will be considered once the factors have been selecled.

lnftlal selection of factors
In this section I describe how a series of factors to pn,di:t executive remuneration was tested. The first stage in the selection of factors involved undertaking a pilot study: a sample of participants in a salary survey was taken as the basis for further investigation. The sample consisted of 470 executives employed in 56 organizations, in five different induslries. The basic salary, total salary (inclusive of all cash remuneration) and total employment cost (inclusive of all benefits, but excluding deferred compensation and srock options) were collected for each individual. Other details on individuals and their organi7.ations were also obtained in order to assess which factors would best predict the executives' incomes. The factors tesled in the pilot study wae selected as they have been studied, to a lesser or greater extent, in the literature. Where it was considered that it would be difficult to gather or quantify data relating to certain factors, these were not tested. One such factor is contribution to shareholder equity. The nature of the sample and the salary survey database would not permit the collec· lion of reliable data for this factor.
The initial factors were divided into three categories. 1be first category (relating to organi1.Btional size) consisted of the following: -number of employees annual turnover (sales) -value of tangible assets -annual salary and wage bill -industry sector.
The second category of factors (reflecting organi:zational perfonnance) contained the following: _ annual net profit before tax _ return on investment The third category (assessing individual factors) included the following: -level of decision making -consequence of error -age of executive -length of service . . . . -organi:zational level (meanmg pos1hon, such as CEO, ma holding or subsidiary company).
In order to enable statistical analyses to be perfonned, it was necessary to assign quantifiable values to the above factors. For the size and performance parameters, a value can be attached appropriately. In the case of descriptive factors, executives were required to score themselves according to a series of definitions and scoring instructions. All factors were initially weighted equally. (It will be seen that the weightings obtained from the regression analy~es ~ondu~ted in 1984 are retained because subsequent analysis yields httle change in weightings.)

Prellmlnary data stratification
Points were assigned to the three sets of factors as described above. During the pilot study, the salary and benefits levels for the 470 executives were gathered together with the infonnation pertaining to the factors listed in the three categories.
Because of the relatively small number of individuals in the pilot study, some normali:zation of the data was necessary. The frequently encountered survey ratio was used ~ ensure meaningful stratification of the data. A survey raho of 1.5 was selected, that is, the ratio of the highest to lowest salary in each size category was calculated. The interpretation is that, for example, in the category level Annual Turnover (Sales) RJOO-R300 million, the highest salary may be no more than 1.5 times the lowest salary. The use of the survey ratio meant that excessively high or low salaries, relative to the data as a whole, were excluded from the analysis.
A measure of capital/labor intensity has also proved useful in data normali:zation because there may be a significant difference between the remuneration of executives in highly capital industries and labour intensive industries (Kostiuk, 1990: 94). Data were excluded if the ratio of value of tangible assets (the measure of capital intensity) to number of employees lay outside the range RIO 000 to RlOO 000 per employee.

Prellmlnary assessment of factors
The significant trend towards the 'cafeteria' approach in the remuneration of senior staff has entrenched concepts such as value of remuneration package or total employment cost as the most meaningful measure of executive remuneration. For this reason, the factors which best predicted total employment cost were sought. Excluded from the total employment cost are deferred compensation and stock options, despite the fact that these can constitute a significant proportion of an executive's income.3

2S
Using regression analyses, the coefficient of determination (r-squared) was used to assess the extent to which the variation in the dependent variable (in this case, salary in the form of total employment cost) is explained by the independent variable (in this case, each of the factors being tested). The value of r-squared may be interpreted as follows: when considering the independent variable, say, annual salary and wage bill, and the dependent variable total employment cost for a CEO in a particular industry, regression analysis may yield a value of r-squared of 0.69; this means that 69% of the variation in the total employment cost of that CEO is explained by the variable annual salary and wage bill.
The results obtained from the preliminary regression analyses of all the factors investigated in the pilot study yielded widely varying results for r-squared. The variable with the lowest value of r-squared (less than 0.01) was the age of the executive. In simple terms, of all the factors considered, age is the least significant predictor of executive remuneration. This finding agrees with that of Kostiuk (1990). The highest value of r-squared (0.90) was obtained for the independent variable annual turnover (sales). Using these values of r-squared, it was decided that certain factors could be excluded as they were poor predictors of executive remuneration. Multiple regression analyses also indicate which combinations of independent variables will best predict the dependent variable. Such combinations generally produce a higher value for r-squared than individual factors. The factors chosen were those which yielded the highest rsquared on an individual basis, using the t-statistic at a 95% confidence level and infinite degrees of freedom.
Two combinations of independent variables which best predicted the dependent variable were selected: Grou~ 1 considered only directly quantifiable measures of organi:zational size and performance, while Group 2 combined s~ parameters with certain individual factors. The two combinations were: The next step was to test these factors in a larger sample, and to establish the correct weighting for each. In 1984 and1988 participants in a South African executive survey of salaries and benefits were asked to score or rate themselves using a set of explicitly defined rules, ~so su1: milling the size and perfonnance parameters o! thetr organ1-:zations. Salary and benefits data collected m the salary survey were also submitted by this ~e ~mp!e o~ near:ty 6 000 executives in over 1 200 orgam:zauons m mne different industry sectors. The executive positions extended One immediate problem in the exercise so far was that all factors were weighted equally. This is unsatisfactory because, as Sargent points out 'Experience has shown that except by rare coincidence each factor carries a different force (or weight) in determining job contenL Consequently, each factor has a different point value ' (in Rocle [ed.], 1984: 11/ 4).

Determination of factor weightings
An assessment was made of how well the various factors predicted executives' basic salary, total cash salary and total employment cost, by comparing the results of a series of regression analyses. Given in Table 1 Table 1).
In Table l, the r-squared figure for the All Group 2 Factors Combined has been assigned a value of 1.00, with other factors expressed as a proportion of this figure.
It is evident from the r-squared values in Table 1 that the best single predictor is annual company turnover (sales), and that the poorest predictor is annual company profit ( corroborating the views of Roberts and others). Both sets of factors (for each year) chosen show that the combinations of factors yield a higher value of r-squared, thereby indicating that the combinations of factors are better predictors of executive remuneration than any of the single factors alone.
It is also apparent that the factor combinations are better predictors of the CEO's remuneration than that of the Marketing Executive. It may therefore be suggested that a different set of factors should be used for each executive position. However, from a practical point of view, it would be cumbersome to assign different sets of factors (wilh different weightings) to each executive position. In order M> obtain the best overall picture for all executives in lhc sample, a series of regression analyses was conducted for the entire sample. The correct weighting for each factor was obtained from the regression coefficients: these are the ~ coefficients in the regression equation of the form: log remuneration= a+ P.x1 + Pix2 + ~X3 + ...
Assigning a weighting value of 1.00 to annual turnover, the weightings shown in Table 2 were obtained. It can be seen from Tables 1 and 2 that there is little change in the relative importance of the factors when com· paring the figures for 1984 and 1988. As a factor on its own. profit is seen to be less important in 1988 than in 1984. It remains the least significant predictor of remuneration by a considerable margin, as is evident from Table 1. The regression coefficients in Table 2 also show a reduced figure for profit. Other figures have not changed significantly. The results do not at this stage include industry sector, although it is felt that this is an important variable whose list effects should be analyzed. The reason is that industry sector classifications collected for executives in holding companies did not take into account that subsidiaries of the holding companies could be in several categories. This inconsistenCY would partially explain the spurious industry sector results.

Market pricing model
Since the Group 2 factors give a better overall prediction of executive remuneration, these factors were selected as tho basis of the market pricing model. The initial factorS in Group 2 may now be re-weighted in order to incorporate tho   Table 3. The market pricing model is thus developed by having executives score themselves, and then using the market pricing survey data to establish their market related remuneration levels. The Group factors and their definitions remain the same, except that the weightings of the factors then become those given in Table 3. Although regression analyses were conducted on the data for two distinct years, only one set of weightings is used: it was felt that the barely perceptible changes in relative importance of the factors from 1984 to 1988 do not justify a change in weightings.
As an example, assume that a CEO is employed in an organization with the following size parameters, and where the CEO's level of decision making, organization level and consequence of error are as shown in this example: Consequence of error Errors will affect survival of company, but will only become apparent some years Score 24

Level of decision making
Total score 137 (~pace does not permit the inclusion of the scoring definitions and rules whereby the above scores are determined.) The executive now needs to relate this score for a CEO to r, the ~arket value of remuneration, which would typically be supplied by the remuneration survey as shown in Table 4.
The CEO with a score of 137 will thus expect to earn be-twee~ R207 000 (lower quartile) and R312 000 (upper quartile) per annum. The percentile chosen will depend on a number of factors such as company policy, as well as the fact that 137 is the maximum in that category: it would be expected that the executive would be paid towards the upper quartile. ~ee reasons for discrepancies between predicted execuuve pay and actual pay are given by Main (1991): some companies are more aggressive in tying top executive pay to company performance (such as profit); the volatility of the market; evidence of top executive pay lagging behind company performance.

Conclusion
In this article I have described how factors were chosen that would best predict executive remuneration, using multiple regression analysis. The better combination of factors (with factor weightings also determined by the regression analysis) was then used to develop a market pricing model which takes into account organizational size and performance factors as well as certain organizational factors. The model relies on an individual's scoring the executive position using a set of rules. The process is subject to the same criticism that could be levelled at any job evaluation system (inexactitude of the factor definitions, subsequent misinterpretations and possible inaccurate scoring of individuals' positions). This study has analyzed the data at two specific points in time: the factors and weightings seem to exhibit a robust set of circumstances. After a further time period, additional work considering the changing importance of factors over a number of years may provide an interesting extension to the study.
The important outcome of a study such as this is that a simple model is produced which considers both organizational size and performance, as well as individual factors, when seeking a basis for determining market related executive remuneration.

Notes
The remuneration survey database used in this study was that 2. See, for example, Roberts (1959), Crystal (1989), Cosh (1975, Kerr & Bettis (1987), Kostiuk (1990), Oi (1983. 3. Different authors have varying approaches to the measure dyof remuneration in comparable studies. Kostiuk's (1990) stu ncentrated on total cash remuneration and also excluded :ferred compensation and stock options. Gerhart & Milkovich (1990) found that contingent pay is associated with financial performance, but basic pay is not This study follows Kostiuk as both contingent and basic pay are included in total cash remuneration.