THE ROLE OF ACTORS AND ACTRESSES IN THE SUCCESS OF FILMS: HOW MUCH IS A MOVIE STAR WORTH?

Transcription

1 THE ROLE OF ACTORS AND ACTRESSES IN THE SUCCESS OF FILMS: HOW MUCH IS A MOVIE STAR WORTH? W. Timothy Wallace, Alan Seigerman, Morris B. Holbrook A bedrock issue in film and video research concerns the use of empirical studies to guide production or program design. Such design decisions exert a direct effect on financial success. And, for the film industry in general and the Hollywood movie in particular, financial performance is dearly the name of the game (Vogel 1990). Indeed, as noted with understatement by Donahue (1987), "it soon becomes very apparent that if one wants to create motion pictures in 'Hollywood,' one needs to understand and accept the fact that filmmaking is a business" (p. 283). From this business-oriented perspective, it follows that the production of Hollywood films aims at pleasing the largest possible audience of moviegoers (Austin 1989, Jowett 1976; Jowett and Linton 1989; Vogel 1990). In this connection, researchers have often studied the factors associated with a film's audience-attendance figures or box-office receipts (Austin 1989). Success at the box office translates directly into rental income for the film distributor (i.e., the fees paid by the theaters for showing the film) as a primary indicant of commercial success (Jowett and Linton 1989, p. 24). Thus, as discussed by Austin (1989), research has addressed such issues as the influence of external economic factors (p. 35), the motivations for moviegoing (p. 50), the effect of advertising (p. 67), the impact of critical reviews (p. 69), the role of word-of-mouth (p. 72), and reactions to the MPAA parental

2 guidance ratings (p. 111). (See also Cameron 1986; Donahue 1987; Eliashberg and Sawhney 1991; Hirschman and Pieros 1985.) Further, as noted by Jowett and Linton (1989), a substantial but scattered body of research has investigated the effects of Oscar Awards on box-office success (Dodds and Holbrook 1988; Litman 1983; Smith and Smith 1986; Sommers ). Yet - with the exception of an occasional study on the role of genre or on such production elements as the director, producer, and screenwriter (Austin 1989, p. 75; Kindem 1982, p. 90) - relatively little research appears to have explored questions associated with the design of the movie itself. With respect to this latter issue, perhaps the most widely held and devoutly cherished belief found within the motion-picture industry concerns the role of actors and actresses as key design components responsible for attracting a large audience of loyal fans (Donahue 1987, p. 34, p. 191). Based on this faith in "star power," Hollywood has traditionally relied on what Vogel (1990) refers to as "bankability" or "clout" and what Powdermaker (1950) describes as a "star system" that "provides a formula easy to understand and has made the production of movies seem more like just another business" (p. 228). That "formula" has resulted in payments to film stars such as Tom Cruise, Arnold Schwarzenegger, Eddie Murphy, and Sylvester StaUone of $9, $12, $13, and $20 million, respectively, for their appearances indays of Thunder, The Terminator II, Beverly Hills Cop, and Rocky V (Corliss 1991; Duffy 1991; Fabrikant 1990; Greenwald 1990; Grover 1991; Newcomb and Schlfrin 1990). Drawing on the work of Rosten (1941), Jowett and Linton (1989) summarize the rationale for this star system as follows: As Rosten...has pointed out, "Hollywood learned that pictures with stars make money, and those without stars do not - or do not make as much as they would if they featured popular personalities... The star system was hailed as the foundation of movie prosperity." Today, the star system (or perhaps "star-cult" would be a better term) continues to be an important cornerstone in the production of movies, and no major production can be contemplated until a guaranteed successful box office personality has agreed to star in it (p. 88). Based on such convictions and beginning as early as 1929, Hollywood has relied on various estimates of audience appeal in making design-related casting decisions. For example, during the 1940s, George Gallup's Audience Research Incorporated (ARI) performed

3 surveys to estimate the box-office attraction or "marquee value" of different movie stars. The use of such measures responded in part to the results of "ARI research indicating that a star's marquee value accounted for some 16 percent of the variance in movie rentals" (Austin 1989, p. 11). Somewhat paradoxically, both the review by Austin (1989) and that by Jowett and Linton (1989) emphasize that what little empirical research exists on the effects of star power tends to cast-doubt on the monetary value of big-name movie stars as design elements in the production of t~lrn~. Corliss (1991) has recently presented a similar conclusion in the popular press. But perhaps the most comprehensive overview of this contentious issue appears in a chapter by Kindem (1982) on "Hollywood's Movie Star System." Kindem (1982) begins by acknowledging that...the Hollywood star system is a business strategy designed to generate large audiences and differentiate entertainment program~ and products, and has been used for over seventy years to provide increasing returns on production investments (p. 79). However, Kindem also examines the empirical evidence bearing on this claim. In this connection, he cites findings from a study by the Gallup Organization that supported "the validity of the star system in the 1940s by suggesting that marquee values accounted for about 26 to 27 percent of the variation in movie box-office success" (p. 84). Further, Kindem reports that his own regression of fdm revenues on the combined marquee values of their stars produced an R 2 of 0.23 with a highly significant effect of marquee value (p ) on a sample of 87 film.~ from the 1940s. By contrast, Simonet (1978, 1980) concluded that the causality ran in the opposite direction on the grounds that "for nine players, marked rises in marquee value occurred after - not before - release of a top-grossing film" (quoted by Kindem 1982, p. 87). With similarly skeptical implications, Garrison (1971) found that - for 62 movies from the late 1960s - the box-office ratings given by theater owners were related to box-office success in opposite directions for actresses (positively) as opposed to actors (negatively). This finding caused Garrison to conclude that "it would appear that the money spent on highly inflated salaries for stars was not productive" (quoted by Austin 1989, p. 77). 3

4 Yet Kindem (1982) has suggested reasons for mistrusting the results of the studies by Simonet and Garrison. Foremost among these is the charge that "neither Garrison nor Simonet isolates consumer demand for movie stars as an independent variable" (Kindem 1982, p. 92). In other words, both these authors used some measure of past performance (e.g., past box-office success or ratings by theater owners) to compose an aggregated one-variable representation of actor or actress appeal for the fdm as a whole. Hence, the contributions of separate film stars did not appear in the models that they tested. Nor did their small samples of only 56 to 73 relatively ancient films attain enough size or recency to warrant confidence in the reliability or validity of their results. Thus, as Kindem (1982) concludes, "it is...difficult to refute the star system with analyses of data that do not properly measure the significant construct of interest, namely consumer demand for [specific] movie stars" (p. 92). Clearly, then, the choice of what actors and actresses to include in a film remains a potentially crucial aspect of movie design in the motion-picture industry. Yet, thus far, we lack applications of sound methods for determining the contributions to market success made by the inclusion of various stars. Toward that end, the present study proposes a method for investigating this issue and applies this approach illustratively to a large group of over one hundred actors and actresses from a large sample of over sixteen hundred recent films. To preview briefly, we shall examine the questions of which control variables explain the market success of these fdms in general and which actors or actresses tend to enhance or detract from f'tim revenues in particular. In this connection, we shall describe and illustrate a method that regresses rental incomes on various control variables and on dummy variables coded to represent the performance of various film stars as well as the manner in which that performance has changed over the courses of the stars' careers. Method In general, we selected a group of popular movie stars and built a data base of information on all the films in which they had appeared. For each of these films, we collected data on various control variables (such as production costs, country of origin, genre, and length in minutes). Entering these control variables into a stepwise regression to explain rental income (that portion of box-office receipts that is returned to the film distributor), we derived a best-fitting equation to control for the effects of extraneous factors (those not directly con- 4

5 netted with the movie stars of interest). Working with the residuals from this model based on control variables (i.e., estimated minus actual rental incomes), we then performed further stepwise regressions to include dummy variables representing the presence or absence of each actor or actress as well as the linear and quadratic interactions between the star and the year in which the movie was released. These dummy variables, when significant, indicated the incremental contributions of various stars to rental revenues and - in some cases- showed significant linear or curvilinear changes in the stars' contributions over the courses of their careers. Sample Our sample of film~ consisted of 1,687 motion pictures released between 1956 and These movies were chosen by virtue of including at least one star from among a set of 111 actors and actresses that had appeared in Quigley's annual poll of the top box-office draws, as reported by Screen World (1989) for the years from 1970 to Specifically, Quigley (1990) surveys the motion-picture exhibitors each year to determine the current top money-making stars. Hence, the annual Quigley report gives a fair account of those actors and actresses who appear to possess "star power" at a given moment. Stars To preserve relevance to the current market for motion pictures, the set of stars chosen for study included only those actors and actresses who were alive at the time of the data collection and who had made at least seven films during the courses of their careers. This group of movie stars contained the following 111 names.( Table 1) For purposes of data analysis, these actors and actresses were represented by 111 zero-one dummy variables, with each movie coded "1" for those dummy variables indicating the stars who appeared in that particular film and coded "0" for those pertaining to all the other stars. These star-related dummy variables served as the bases for the key independent variables of interest in explaining our major dependent variable, rental income. Dependent Variable: Rental Income We used Rental Income as our measure for the market success of each film. This dependent variable was measured in millions of dollars, adjusted in some important ways described later. As noted earlier, Rental Income refers to that portion of box-office receipts that is

6 Table 1. List of Movie Stars Used in the Analysis Allen Alda Allen Arkin Warren Beatty Jacqueline Bisset Charles Bronson John Candy Cher Glenn Close Kevin Costner Danny DeVito Faye Dunaway Sally Field Harrison Ford Ted Garr Whoopi Goldberg George Hamilton Katherine Hepburn William Hurt Diane Keaton Jessica Lange Rob Lowe Dean Martin Walter Matthau Dudley Moore Bill Murray Leonard Nimoy Ryan O'Neal Gregory Peck Richard Pryor Burt Reynolds George Segal Brooke Shields James Stewart Donald Sutherland John Travolta Jon Voight Gene Wilder Woody Allen Dan Akroyd Robbie Benson Marion Brando George Bums Dyan Cannon Tommy Chong James Cobum Tom Cruise Michael Douglas Clint Eastwood Jane Fonda Michael J. Fox Richard Gere Louis Gossett, Jr. Tom Hanks Charlton Heston Timothy Hutton Michael Keaton Tom Laughlin Shirley Maclaine Steve Martin Bette Midler Roger Moore Paul Newman Nick Nolte Tatum O'Neal Michelle Pfeiffer Robert Redford A. Schwarzenegger Tom Selleck Sissy Spacek Meryl Streep Patrick Swayze Kathleen Turner Sigoumey Weaver Robin Williams Julie Andrews Kevin Bacon Candice Bergen Matthew Broderick James Caan Chevy Chase Jill Clayburgh Sean Connery Robert De Niro Richard Dreyfuss Peter Falk Peter Fonda James Garner Mel Gibson Gene Hackman Goldie Hawn Dusfin Hoffman Glenda Jackson Kris Kristofferson Jack Lemmon T. "Cheech" Marin Marsha Mason Liza Minnelli Eddie Murphy Jack Nicholson Chuck Norris A1 Pacino Sidney Poitier Christopher Reeve George C. Scott William Shatner Sylvester Stallone Barbra Streisand Elizabeth Taylor J.-M. Vincent Raquel Welch Debra Winger 6

7 returned to the film distributor. For the 908 movies with rental incomes of $3 million or greater, this figure represented total revenues received by the film's distributor from domestic theater owners, as reported by Variety (1990a). For the 779 poor-performing films with domestic revenues of less than $3 million (below Variety's cutoff), Rental Income was set at an approximation of $1.5 million (before the adjustments described later). (Notice that, because rental incomes varied over a range of over $168 million before adjustments, there is no reason to fear that this approximation at the very low end would make any important difference in the empirical results - especially not after the adjustments explained in what follows.) On average, domestic rental income represents 45 percent of domestic box-office receipts (Donahue 1987, p. 34). However, this figure does not include subsequent revenues from foreign, network-television, pay-cable, homevideo, and other ancillary markets (Donahue 1987; Vogel 1990). Control Variables In addition to data on the presence or absence of the movie stars for each of the fdms in which any one of them had appeared, we collected information on each of the movies included in the sample with respect to a number of control variables. These control-variable data came from various sources such as CineBooks (1989), HaUiwell (1987), Quigley (1990), and Variety (1990b). We shall describe each of these control variables briefly so as to clarify their operational def'mitions. Year:. The year of the film's domestic release (CineBooks 1989; Halliwell 1987). Rating: The film's rating by CineBooks (1989) on overall quality from zero ("without merit") to five (***** = "masterpiece"). Parental Guide: A parental recommendation in which CineBooks (1989) evaluates the style and content of each fdm according to its suitability for children on a six-position continuum that ranges from "good for children" (1) to "objectionable for children" (6) (used in place of the MPAA rating, which was not available for fdms released during the first third of the years covered). Country: The film's country of origin- coded into three categories indicated by two zero/one dummy variables to represent the United States, other English-speaking countries, and non-english-speaking countries, respectively (CineBooks 1989). 7

8 Minutes: The film's length of running time in minutes (HaUiwell 1987; Quigley 1990). Genre: A set of zero-one dummy variables representing the Cinebooks (1989) classification of films' genres into 25 categories - Action, Adventure, Biography, Children's, Comedy, Crime, Dance, Disaster, Documentary, Drama, Fantasy, History, Horror, Musical, Musical Comedy, Mystery, Prison, Religion, Romance, Science-Fiction, Spy, Sports, Thriller, War, or Western. Cost: The total costs and expenses involved in acquiring and producing a film (in $ millions), including all overhead and interest charges, as reported by Variety (1990b); for fills too low in cost to be covered by Variety, average production costs as reported by Quigley (1990). Data Adjustments To achieve the best possible statistical fit, we applied a number of theoretically appropriate adjustments to the data. We shall describe each of these briefly. Normalizations All continuous variables were normalized by subtracting their means across the 1,687 movies. This adjustment helps to reduce multicollinearity between linear and quadratic terms but does not affect the shapes or overall R 2 for the linear and curvilinear relationships of interest. Quadratic Terms To check for curvilinear effects, each continuous independent variable was represented both by its simple normalized value and by this value squared. Significance of the quadratic term would indicate a reliable departure from linearity in a given relationship. Deflator Because the value of the dollar has changed considerably over the years from 1956 to 1988, we adjusted all monetary figures to constant 1989 dollars by means of a deflator computed to reflect the average movie-ticket prices prevailing in a given year. Therefore, in what follows, we shall report the results for the ticket-price-deflated measures in constant 1989 dollars. Annual Means For each f'dm, rental income (in constant 1989 dollars) was further adjusted by subtracting the mean rental income of all movies in the sample released during the same year. This adjustment has the effect of correcting for extraneous sources of variation due to general economic trends or attributable to the fortunes of the

9 movie industry as a whole. In other research, such extraneous determinants of industry performance might hold considerable interest in their own right. Here, however, they were viewed as potential sources of error and were controlled for by subtracting the annual means in measuring Rental Income. Star X Year and Star X Year z interactions. Finally, for each actor and actress separately, we tested for changes in market success over the course of that star's career by including multiplicative interaction terms to represent the Star X Year and Star X Year 2 interactions (where Star is a zero/one dummy variable and Year is a normalized continuous variable). Inclusion of these extra interaction terms permitted a clear interpretation concerning possible trends in the effect of an actress' or actor's star power. Interpretation. A significant positive (or negative) regression coefficient for Star would mean that, taking into account the fee paid to the actor or actress as part of a fdm's cost, that particular star has tended to contribute a significant additional increment (or decrement) to market success, as measured by an incremental contribution to Rental Income beyond that earned by the typical actor or actress in our sample of t'rims. A positive effect for Star would mean that, compared with the norm for other actors and actresses, fdm producers would have had an economic justification for paying that star an additional fee up to the amount represented by the coefficient for that term (in millions of 1989 dollars). A negative effect for Star would mean that, compared with the market performance of others, he or she has tended to be overcompensated by an amount equal to the relevant regression coefficient. Meanwhile, a significant Star x Year coefficient would indicate a linear trend in the incremental financial contribution of that particular actor or actress over the course of his or her career, while a significant Star x Year term would show that a curvalinear or possibly a nonmonotomc trend has appeared in the relevant star's market value. We tested for such effects of interest by means of regression analyses. Analyses Data analyses employed stepwise ordinary least-squares regression. Here, in general, stepwise procedures appear well-suited for our present purposes because we were ultimately concerned with obtaining estimates for the effects of the three star-related variables just discussed (i.e., identifying those with significant impacts on market suc- 9

10 cess in the past) and not with developing a model for purposes of predicting future performance (which would have been subject to the danger of search bias associated with the generalization of stepwise results to new samples of data). Accordingly, we began by using stepwise regression with a selection criterion based on a significance level of p 0.10 to fred the best model for explaining variance in Rental Income using only the aforementioned control variables considered alone. After finding the best-fitting model based on just these control variables, we performed a second stepwise procedure on the residuals from this first equation to select the Star, Star x Year, and Star x Year 2 terms that explained significant incremental variance in residual Rental Income. Here, for completeness and to provide a base line against which to compare other effects, Year and Year 2 were included in the equation as a starting point. After that, to permit fully determining the shape of the relationship for each relevant actor and actress, we followed the rule that if Star x Year 2 is significant, both Star x Year and Star must also be included in the equation and, similarly, that if Star x Year is significant, Star must also be included. Graphical Displays In addition to the tabular presentation of these results, we also obtained graphical displays for those stars with significant regression coefficients. The resulting figures plot the Estimated Residual Rental Income against Year. Each graph provides a base line (at the level where the Estimated Residual is zero) to represent the market performance of the typical actors and actresses whose contributions (after adjusting for Year and Year 2) do not depart significantly from the norm (associated with the aforementioned control variables). Each actor or actress with a significant Star, Star x Year, or Star x Year 2 coefficient appears as a flat, inclined, or curvilinear row or band of estimated points - appropriately labeled - whose positions show the effect of that star's market performance as compared to the norm established by the base line for the remaining actors and actresses. Results First Regression. The first regression model explained about a quarter of the variance in Rental Income. Specifically - based on the control variables considered separately and including only those that entered at a level ofp 0.10 or better - the stepwise regression produced a fit of R = 0.52 (F(15,1671) = 42.26, p ). Here, in one form or another, all the types of control variables mentioned earlier entered 10

11 the equation, usually with highly si~cant contributions. These statistically significant effects appear in Table 2. As shown in Table 2, some of the important effects of control variables are linear and therefore easy to envision. Thus, Parental Guide contributes in the negative direction (p 0.05), with the implication that every successive level of offensiveness along the six-point continuum from "good for children" to "objectionable for children" subtracts about $1.8 million from the market success of a movie. By contrast, producing a fdm in the United States (as opposed to a non-english-speaking country) tends to add about $5.6 million to its Rental Income (p 0.005). And eight Genre categories show positive effects on market value: Adventure ($5.8 million), Comedy ($6 million), Disaster ($31 million), Horror ($14 million), Musical ($8.8 million), Religion ($17.5 million), Science Fiction ($25.6 million), and Spy ($7.8 million) (all significant at p 0.05, with three significant at p ). Further, as might be expected, the Cost of a fdm is positively related to its market success (p ). However, when controlling for the other factors included in the equation, the Rental Income of a fdm tends to recapture only about half (45 cents) of an additional dollar spent on production costs (a finding to which we shall return later for discussion). Two of the variables shown in Table 2 appear to exert curvilinear effects on Rental Income, as indicated by the presence of quadratic terms (both significant at beyond p ). Because the variables have been normalized, these can be tricky to interpret without actually plotting the curves in question over the range of the relevant data. Such plots (omitted here to save space) suggest that the effect of Rating on Rental Income is U-shaped. At the lowest rating level, an extra rating point actually subtracts about $6 million from the market success of a fdm. In other words, it appears that a bad movie has something to gain from being as trashy as possible. At higher levels of quality, Rental Income begins to increase with the movie's Rating until the last rating point (i.e., the shift from **** to *****) is worth about $24 million in terms of market success. Thus, for a good movie, it apparently pays to strive for even greater excellence. A simpler relationship pertains to the effect of Minutes on Rental Income. Over the range in our data, this effect is rather flat for movies short in duration. But, as their running time in minutes increases, the effect of additional length appears to accelerate such that - at the top 11

12 Table 2. Results for Stepwise Regressions Using Control Variables to Explain Rental Income Independent Regression Variable Coefficient t-value p-level %ntercept Rating Rating Parental Guide Country - U.S Minutes Minutes Genres: Adventure Comedy Disaster Horror Musical Religion Science Fiction Spy Cost

13 of the range (for the movies longest in duration) - an extra hour's length contributes an incremental $25 million to Rental Income. Second Regression The second regression model - incorporating the effects of actors and actresses - explained about a quarter of the remaining variance. Specifically, the use of Year, Year 2, Star, Star x Year, and Star x Year 2 to explain residuals from the preceding model (based only on control variables) produced a multiple regression fit ofr = 0.47-(F(55,1631) = 8.57, p ). Here, in effect, the control variables from the preceding model (already accounted for in Table 2) plus the Intercept, Year, and Year ~ terms (presented at the top of Table 3) provide a base line (applicable to all films) again.~t which to compare the performances of those stars whose market success has departed from the norm. The outcomes of these comparisons concerning the market success of the various actors and actresses appear in the si,~nificant star-related regression coefficients shown in Table 3. This table shows that twenty-four actors and actresses contributed to rental incomes in patterns that, for many cases, varied over the courses of their careers. By comparison with the norm for actors and actresses in general, the impact of these 24 stars on a fdm's Rental Income and how this impact has changed over their careers need to be taken into account. The manner in which these effects have occurred appears most dearly in the visual representations to which we now turn. Graphical Analysis Guides to interpreting the coefficients shown in Table 3 appear in the graphs presented in Figures 1 through 4. These graphs plot the estimated departures from the base line just described (vertical axis) against the year of the fdm's release (horizontal axis) for each of the 24 actors and actresses with significant star-related regression coefficients in Table 3 (as identified by the legend that accompanies each figure). Here, the base-line performance of the typical actor or actress appears as a horizontal line across the middle of each display (at zero on the vertical axis). Meanwhile, the incremental contributions of the 24 stars listed in Table 3 - and the manners in which these have changed over the courses of their careers - appear as individually labeled rows or bands of points that can easily be compared with the norm established by the base-line performance. These comparisons reveal some interesting f'mdings. 13

14 Table 3. Results for Stepwise Regressions Using Year, Year2 Star, Star X Year and Star X Year2 to Explain Residuals from the Model Based on Control Variables Independent Regression Variable Coefficient t-value p-level Intercept Year Year Julie Andrews Andrews X Year Andrews X Year Kevin Bacon Bacon X Year Bacon X Year Marlon Brando Brando X Year James Coburn Coburn X Year Tom Cruise Cruise X Year Robert De Niro De Niro X Year De Niro X Year Danny DeVito Richard Dreyfuss Dreyfuss X Year Dreyfuss X Year Clint Eastwood Eastwood X Year Harrison Ford Ford X Year Ford X Year

15 Table 3 - Continued Independent Variable Regression Coefficient t-value p-level Charlton Heston Heston X Year Heston X Year Dustin Hoffman Jessica Lange Lange X Year ~ I Tom Laughlin Laughlin X Year Laughlin X Year Bette Midter Midler X Year Eddie Murphy Murphy X Year Murphy X Year I " Bill Murray Paul Newman Newman X Year Newman X Year " At Pacino Pacino X Year " Robert Redford Redford X Year Redford X Year " Burt Reynolds Sylvester Stallone Stallone X Year Barbra Streisand John Travolta Travolta X Year "

16 As shown in Figure 1, the presence of some stars in a film (Burt Reynolds, Danny DeVito, Dustin Hoffman, Barbra Streisand, and Bill Murray) has tended to exert a positive impact that has remained relatively constant above the base line over the period of the stars' careers (ranging from about $11.5 million for Reynolds to about $22.5 million for Murray). Others shown in Figure 1 have followed a steady downward trend over their careers (AI Pacino, Jessica Lange, and John Travolta). In all three cases, these stars began at from $30 to $70 million above the base line, but then regressed back toward the norm as their careers unfolded. Figure 2 shows six actors and actresses who have achieved a steady upward trend in market power during the span of their activity in the movies (Marion Brando, James Coburn, Tom Cruise, Clint Eastwood, Bette Midler, and Sylvester Stallone). Some, for example, have started near the base line and then risen to levels of incremental contribution well above the $30 million mark (Brando, Stallone, and Cruise). Others have begun at substandard levels but have progressed to contributions somewhat above the norm (Coburn and Midler). The stars presented in Figure 3 have pursued nonmonotonic patterns of recovery (Julie Andrews, Kevin Bacon, Robert De Niro, and Charlton Heston). These have involved declines during their early years, followed by resurgences in the later stages of their careers. For example, both Charlton Heston and Julie Andrews began their careers at over $90 million above the base line, sank to $20 or $30 million below the standard, and then rebounded to a level comfortably above the norm. Both Kevin Bacon and Robert De Niro have pursued similar but less dramatically sweeping paths of decline and recovery. Finally, Figure 4 features six actors whose market success appears to have peaked and then subsided (Richard Dreyfuss, Harrison Ford, Tom Laughlin, Eddie Murphy, Paul Newman, and Robert Redford). Some of these began near or below the base line, achieved peaks well above the norm, and then fell back to standard or substandard market performance (Laughlin, Newman, and Redford). One began high, but has subsequently peaked and declined to below the norm (Dreyfuss). Two appear to have peaked, but still remain well above the standard set by the base line (Ford and Murphy). Explanatory Power An important remaining question concerns the relative degrees of explanatory power contributed by (1) the adjustment for annual mean 16

17 FIGURE i PLOTS OF ESTIMATED RESIDUAL RENTAL INCOME VERSUS YEAR WHEN USING STAR AND STAR X YEAR TO PREDICT DEVIATIONS FROM THE BASE LINE: SIMPLE AND NEGATIVE LINEAR EFFECTS ESTIMATE " lo YEAR NAME... ABase o o O DeVito ~ $ ~ Hoffman D [] Lange ~ ~ ~ Murray.', L ~ Psoino ~ Reynolds ~ -- -" Strelsan c ~ Travolta 17

18 FIGURE 2 PLOTS OF ESTIMATED RESIDUAL RENTAL INCOME VERSUS YEAR WHEN USING STAR AND STAR X TEAR TO PREDICT DEVIATIONS FROM THE BASE LINE: POSITIVE LINEAR EFFECTS ESTIMATE ' 0-10" -20" -3O ~ YEAR NAME... ABase : ~- Eastwood o o o Brando ; ~ ~ Coburn ~ Cruise L ~ & Midler ~ ~ ~ Stallone 18

19 FIGURE 3 PLOTS OF ESTIMATED RESIDUAL RENTAL INCOM~ VERSUS YEAR WHEN USING STAR, STAR X YEAR, AND STAR X YEAR-SQUARED TO PREDICT DEVIATIONS FROM THE BASE LINE: NONMONOTONIC RECOVERY EFFECTS ESTIMATE Ii0 1 i00] 9O 2 80" 70" 60' " 30" 20" lo" \ " -20" " r-~ YEAR NAME... ABase C 3 C Andrews o o o Bacon 0 r~ DeNiro = ~ ± Heston 19

20 FIGURE 4 ESTIMATE ' ] PLOTS OF ESTIMATED RESIDUAL RENTAL INCOME VERSUS YEAR WHEN USING STAR, STAR X YEAR, AND STAR X YEAR-SQUARED TO PREDICT DEVIATIONS FROM THE BASE LINE: NONMONOTONIC PEAKING EFFECTS 7o.t 60" 56' 40' 30' 20' 10" -10' -20' -3Olr-.~_~---r--r-, YEAR NAME... ABase Murphy o o o Drey uss ~ ~ ~ Ford ~ ~ Newman ~-9-~ Redford Laughlin 20

21 Rental Income, (2) the effects of the control variables, and (3) the additional contribution of the star-related variables in accounting for the residuals from the model based on the control variables. Conclusions on this issue appear when regressing Rental Income on (1) a series of dummy variables representing the years in which the films were released (which has the effect of removing the annual means), (2) the year dummies plus the control variables, and (3) the year dummies plus the control variables plus the star-related variables. The variances explained by these three regressions are (1) R 2 = 0.03 (F(32,1654) = 1.68, p 0.01); (2) R 2 = 0.32 (F(47,1639) = 16.38, p ); and (3) R 2 = 0.47 (F(100,1586) = 14.23, p ). This means that, overall, our models explain close to half the variance in Rental Income (47 percent). Of this, the adjustment for annual means accounts for only 3 percent. The control variables explain an additional 29 percent of the variance. The star-related variables contribute another 15 percent to the overall variance explained. Thus, the star-related variables account for close to a third of the total explained variance. [Viewed differently, the variances explained in the residuals after the effects of the preceding variables have been removed are R z = 0.28 for the control variables (Table 2) and R 2 = 0.22 for the star-related variables (Table 3), respectively. This means that the star-related variables account for about 22 percent of the variance in Rental Income that remains after the influence of the other variables has been removed.] Discussion of Results and Limitations As in any correlational study of market performance, our findings are subject to certain limitations that suggest some important caveats. First, it is possible that our results might have differed if we had drawn upon a different group of actors and actresses or a different set of films. We do believe that our large samples of 111 stars and 1,687 movies lend assurance that the findings possess some degree of reliability and stability. Nevertheless, other film and video researchers might want to see what would happen to the results if stars or movies of particular relevance to their own interests were included in the analysis or if directors and producers were added as explanatory variables. Second, the set of control variables that we used could conceivably have omitted some film-specific factors of importance. For example, it is possible that movie-advertising expenditures, special merchandising promotions, press coverage, word of mouth, time of the year at 21

22 which a motion picture was released, degree of compatibility among the people working on the film, or other uncontrolled variables might influence market performance. We had no access to such data and must therefore assume that, if included in the analysis that led to Table 1, they would not have markedly affected the regression results for the star and star x year interaction terms of interest in Table 3. However, the exploration of such potential biases remains an important topic for future research. Third, our measure of market success does contain some imperfections. As previously mentioned, it was necessary to estimate the rental incomes of films below $3 million as $1.5 million (in current dollars before making the adjustments described earlier). This assumption could introduce minor inaccuracies in some instances. However, the $1.5 million estimate anchors the low end of a performance continuum that extends across a range of over $168 million. With such a wide-ranging dependent variable, a possible inaccuracy of no more than about $1 million at the low end of the range would not tend to make much of a difference in the overall results. Fourth, as also mentioned earlier, our measure of rental income did not account for performance in other ancillary markets (network broadcasts, home video, pay cable, foreign distribution, etc.). Possibly, some f'dms that perform poorly (well) in national theater distribution might succeed (fail) when aimed at other audiences. Indeed, this seems a likely explanation for how the coefficient on production costs could fall below 1.0. Specifically, the incremental contribution of Cost to Rental Income in Table 2 appears to be only about This may validate a piece of conventional wisdom in the film industry to the effect that many f'dms overspend on production costs and consequently lose moneyin their domestic theater release but recoup these losses in other markets such as network television, cable TV, videocassette sales, and foreign distribution. Due to limitations in the available data, the present study could not capture the effects of these ancillary markets. Fifth, under some circumstances, stepwise procedures of the type used here can inject search bias into the results of regression analyses. For example, this problem becomes especially serious when one intends to use the regression equation for purposes of prediction on a new sample. However, our purpose here was not prediction but rather estimation of the effects for stars with significant incremental impacts on market success. Stepwise regression appears well-suited for the latter purpose. 22

23 Conclusions Subject to these limitations, our proposed method appears to provide a viable approach in the illustrative example to the problem of estimating the market values of movie stars by identifying how their incremental contributions to rental income have diverged from the base line for other actors and actresses and by indicating how those contributions have changed over the courses of individual careers. Using just the annual means and other control variables (rating, parental guide, country, minutes, genres, cost, etc.), our model explained 32 percent of the variance in f'dm rentals. Including the stars and the star x year interaction terms increased explained variance to 47 percent. This increase of about 15 percent due to the presence of the actors and actresses themselves accounts for about a third of the overall variance explained and happens to coincide quite closely with the figure derived by the George Gallup Association many years ago (Austin 1989; Jowett and Linton 1989) but subsequently disputed in more recent work by, among others, Simonet (1978, 1980) and Garrison (1971). Hence, with respect to this controversy, we tend to agree with Kindem (1982) in concluding that certain movie stars do make a demonstrable difference to the market success of the films in which they appear. Further, our results suggest that this incremental contribution of stars to the market success of films in which they appear can be measured using something like the method proposed and illustrated here. Extensions Our findings permit other derivations that might be of interest to students of the film industry. The pattern of findings shown in Table i supports the emerging concern that many films tend not to earn back their exorbitant production costs during their initial release in domestic theaters. Apparently- after accounting for critical appraisal, length in minutes, country of origin, genre, and other potentially confounded factors - incremental production costs generate only about 45 cents of rental income on the dollar. This finding might encourage one to adopt a more conservative stance in comparison with the blockbuster mentality that has often swept away Hollywood producers (Fabrikant; Greenwald; Landro). Further, where financial return (rather than artistic merit) is the primary objective, one might lean in the direction of producing horror, religion, disaster, and science-fiction films (which contribute $14, $17.5, $25.7, and $31 million, respectively) while avoiding some of the 23

24 omitted categories (such as biographies, dramas, and romances) like the plague. However, one might also keep in mind that - for highquality films near the top of the scale - even a one-point rating increase (which would presumably reflect the excellence of a fdm's direction, scripting, music, technical production, etc.) appears to be worth enough (about $25 million) to counterbalance the poor performance of a noncommercial genre. Finally, based on the kind of analysis illustrated here, those concerned with problems of film design in motion-picture production might reach helpful conclusions concerning the justification of additional fees paid to certain actors or actresses (Chisholm 1990). For example, as indicated in Figure 1, Bill Murray, Barbra Streisand, and Dustin Hoffman qualify as bona fide financial "superstars" whose market appeal would merit payments above the level of those already included in the production costs of the films in which they have appeared. Similarly, as suggested by Figure 2, the contributions to a film's success associated with Marion Brando, Sytvester Stallone, and Tom Cruise have established the commercial "star power" of these movie actors. By contrast, Figure 1 shows that - during the period covered by our analysis - the market value of some stars such as AI Pacino, Jessica Lange, and John Travolta appears to have plummeted from its once lofty perch. Declines of comparable magnitude have also befallen the stars featured in Figures 3 and 4, where the difference concerns a comparison between the patterns of peaking and those of recovery. Thus, in Figure 3, both Charlton Heston and Julie Andrews appear to have enjoyed resurgences after long but precipitous periods of decline. By contrast, in Figure 4, Tom Laughlin, Harrison Ford, and Eddie Murphy all appear to have peaked spectacularly only to hit the skids in their later films. Summary In sum, the proposed approach to estimating the worth of a movie star in terms of his or her incremental contribution to a f'dm's market success appears to have yielded some plausible and potentially useful illustrative findings. For example, while no model can guarantee an accurate prediction of tomorrow's results, the approach proposed and illustrated here does appear to provide a helpful procedure for determining whether a given star has been under- or over-compensated in the past. 24

25 Apparently, in 1989, one would not have wanted to base the next salary for A1Pacino, Jessica Lange, Paul Newman, or Richard Dreyfuss on what they had made in their last few movies because the model suggests that, relative to the base line, they had tended to be overpaid. Conversely, one might have justified additional payments to Bill Murray, Dustin Hoffman, or Barbra Streisand on the basis of their historical tendencies to contribute positively to a fdm's market success. Thus, as one potential input to cinematic product design, we have proposed and illustrated a method for estimating the incremental market value of film stars. Besides indicating differences in the contributions to market success made by various actors and actresses, our results show that their relative degrees of star power have tended to change over the courses of their careers. This demonstration of our approach suggests that it can contribute to the better design of fdms, can provide information useful in justifying the salaries of movie stars, and might thereby help to rationalize the troubled economics of the motion-picture industry. References Columbia University Austin, Bruce A., Immediate Seating: A Look at Movie Audiences, Belmont, CA: Wadsworth Publishing Company, Cameron, S., "The Supply and Demand for Cinema Tickets: Some U.K. Evidence," Journal of Cultural Economics, 10, (June 1986) pp Chisholm, Darlene C., "Optimal Contract Design: A Theoretical and Empirical Investigation of Profit-Sharing in the Motion-Pictures Industry," Working Paper, University of Washington, CineBooks, CineBooks Movie List, Evanston, IL: CineBooks, Corliss, Richard, "Do Stars Deliver?" Time, (26 August 1991) pp Dodds, John C. and Morris B. Holbrook, "What's An Oscar Worth? An Empirical Estimation of the Effects of Nominations and Awards on Movie Distribution and Revenues," in Current Reo search in Film: Audiences, Economics, and Law, Vol. 4, ed. Bruce A. Austin, Norwood, NJ: Ablex Publishing Corporation, 1988, pp

26 Donahue, Suzanne Mary, American Film Distribution: The Changing Marke(place, Ann Arbor, MI: UMI Research Press, Duffy, Susan, "A Slasher in Loose on Paramount's Lot," Business Week, (28 January 1991) p. 53. Eliashberg, Joshua and Mohanbir S. Sawhney, "Modeling Goes to Hollywood: Predicting Individual Differences in Movie Enjoyment," Working Paper, The Wharton School, University of Pennsylvania, Fabrikant, Geraldine, "The Hole in Hollywood's Pocket," The New York Times, (10 December 1990) p. D1. Garrison, Lee Cedric, Jr., Decision Processes in Motion Picture Production: A Study of Uncertainty, unpublished doctoral dissertation, Stanford University, Greenwald, John, "Shooting the Works," Time, (21 May 1990) p. 64. Grover, Ronald, "The World Is Hollywood's Oyster," Business Week, (14 January 1991) p. 97. Halliwell, Leslie, Halliwell's Fihn Guide, Sixth Edition, New York, NY: Charles Scribner's Sons, Hirschman, Elizabeth C. and Andrew Pieros, Jr., "Relationships Among Indicators of Success in Broadway Plays and Motion Pictures," Joumal of Cultural Economics, 9, (June 1985) pp Jowett, Garth, Film: The Democratic Art, Boston: Focal Press, Jowett, Garth and James M. Linton, Movies as Mass Communication, Second Edition, Newbury Park, CA: Sage Publications, Kindem, Gorham, "Hollywood's Movie Star System: A Historical Overview," in The American Movie hzdustry: The Business of Motion Pictures, ed. Gorham Kindem, Carbondale: Southern Illinois University Press, 1982, pp Landro, Laura, "Paramount in the Dark Before Don," Wall Street Journal, (11 December 1990) pp. B1, B7. Landro, Laura, "Specter of Shrinking Earnings Leaves Hollywood Home Alone with a Ghost," Wall Street Journal, (4 January 1991) p. B1. Litman, Barry R., "Predicting Success of Theatrical Movies: An Empirical Study," Joumal of Popular Culture, 16, (Spring 1983) pp

27 Newcomb, Peter with Matthew Schifrin, "Golden Boys and Girls," Forbes, 146, (1 October 1990) pp Powdermaker, Hortense, Hollywood, the Dream Factory: An Anthropologist Looks at the Movie-Makers, Boston: Little, Brown and Company, Ouigley, International Motion Picture Almanac, 61st Edition, ed. Jane Klain, New York, NY: Quigley Publishing Company, Rosten, Leo C., Hollywood: The Movie Colony, The Movie Makers, New York: Harcourt Brace Jovanovich, Screen World, Vol. 40, ed. John Willis, New York, NY: Crown Pubfishers, Simonet, Thomas, "Performers' Marquee Values in Relation to Top- Grossing Films," paper presented at the Society for Cinema Studies Conference, March, Philadelphia, Simonet, Thomas, Regression Analysis of Prior Experiences of Key Production Personnel as Predictors of Revenues from High-Grossing Motion Pictures in American Release, New York: /M'no Press, Smith, S. P. and V. K. Smith, "Successful Movies: A Preliminary Empirical Analysis," Applied Economics, 18, (1986) pp Sommers, P. M., "ReelAnalysis,"Joumal of RecreationalMathematics, 16, 3, ( ) pp Variety, "MIFED Issue," (15 October 1990a). Variety, "Tenth American Film Market Edition," (21 February 1990b). Vogel, Harold L., Entertainment Industry Economics: A Guide for Financial Analysis, New York: Cambridge University Press,

28 本文献由 学霸图书馆 - 文献云下载 收集自网络, 仅供学习交流使用 学霸图书馆 ( 是一个 整合众多图书馆数据库资源, 提供一站式文献检索和下载服务 的 24 小时在线不限 IP 图书馆 图书馆致力于便利 促进学习与科研, 提供最强文献下载服务 图书馆导航 : 图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具