OCCUPATIONAL CHARACTERISTICS OF ARTISTS: A STATISTICAL ANALYSIS Gregory H. Wassail and Neil O. Alper Labor market studies of occupational groups are common in the labor economics literature. Usually, the researcher faces no problem in identifying people who belong to a particular occupational group. In using U.S. Census or Bureau of Labor Statistics data, one finds that occupational categories are already established. In conducting one's own survey, either the occupation of survey respondents is known in advance, or it can be readily obtained via an appropriate question on the survey instrument. When conducting a labor market study of artists, however, several difficulties are encountered. One difficulty involves the use of Census data. First, the occupations that the Census categorizes as "artists" do not correspond to those used by researchers in the field.(1) A second difficulty, indicated in our research (Wassall, et al. 1983) and that of others (Kingston, et ai. 1981; Ruttenberg, et ai. 1978), lies in the fact that artists typically hold other jobs while working as artists. Although the reasons for holding other jobs are diverse, many artists who hold them find they spend more time and earn more money at them. However, the Census questionnaire asks that the respondent report only one occupation, and that it be the one at which the most time was spent.(2) This criterion may be desirable for allocating persons to occupations in general, but raises serious questions when applied to those who work as artists. At the very least, the implications of the Census taxonomic procedure should be explored. Thus, there exist greater incentives for researchers wishing to study artists to conduct their own surveys of the relevant population than there might for studies of other occupational groups. However, a problem that arises when studying artists via a direct survey approach is selecting an appropriate rule that excludes from the sample to be analyzed those who don't "qualify" as artists. Because the paths to artistic success are diverse, literally 13

any person working in an artistic field can, and often does, describe himself as an artist without facing a challenge to this designation. In other professions, there exist natural screening devices, such as the M.D. degree and C.P.A. designation, which are generally accepted as determining who should be counted as members of those professions. No such natural screening device exists for artists. Social scientists who have surveyed artists have often devised screening criteria to exclude survey respondents lacking the desired qualifications. For example, (Ruttenberg, et al. 1978) limited their survey of performing artists to only those who belonged to performing arts unions. Also, (Kingston, et al. 1981) limited their study of authors to those who had published at least one book, and (Felton 1978) limited her study of composers to those who had at least one composition published, recorded or performed. To our knowledge, however, the implications of applying screening devices to a survey of artists have not been explored. In this paper, the implications of using screening devices are explored. Several different screening devices, or filter rules, are applied to a broadly defined sample of 3,027 New England artists who responded to a mail questionnaire in 1980 and 1981..(3) The filter rules are then compared for consistency with one another. Next, 35 socioeconomic characteristics of the artists who pass through and fail to pass through each filter are analyzed. This analysis is made first by evaluating the mean values of the socioeconomic characteristics of the groups accepted and rejected by each filter rule, and testing for statistically significant differences between means. Second, a linear discriminant function is estimated for each filter rule. By requiring the discriminating variables which enter each function to possess a minimum level of significance, redundant information found in the comparison of means may be eliminated. Also, the effectiveness of each of the filter rules in segmenting the sample of artists into two groups, either possessing or not possessing the desired requirement, can be compared and evaluated using discriminant analysis. It is hoped that this analysis will shed some light on the implications of imposing filter rules on a population or sample of artists. 14

Further, the effects of the Census method of identifying artists can be indirectly explored.

The Filter Rules Before discussing the filter rules, it is important to note several aspects of the survey conducted by the authors. It is an occupationally comprehensive survey of artists. Each respondent was classified into one of nine occupational categories: dancer; musician; actor; theater production; writer; choreographer, composer, and playwright; visual artist; media artist; and craft artist. Since one survey instrument was administered to all artists, it was not possible to collect the kind of information that some of the authors cited above collected and subsequently used as screening devices, such as number of books or compositions published or the number of paintings sold. Also, the number of respondents rejected from the sample prior to the application of the filter rules was kept at a minimum. Only those respondents who indicated that they did not work as artists at all during the year for which the survey data were collected were deleted.(4) Five binary variables are tested for their performance as occupational filters, Two are income-related. One, INCOME, takes a value of 1 if the artist earned any income from his art during the year of the survey, 0 if he did not. The other, IRS, applies essentially the same test that the Internal Revenue Service uses in judging whether an artist is pursuing an occupation or a hobby, That is, IRS equals 1 if artistic income exceeds artistic expenses and 0 if it does not. A third filter variable, UNION, applied solely to performing artists (i.e., dancers, musicians, and actors), has a value of 1 for artists who indicated membership in one or more performing arts unions. A fourth variable, ONLYART, has a value of i if an artist held only artistic jobs during the year. This variable has a rough correspondence to the Census criterion for choosing one's principal oceupation. However, ONLYART is more restrictive. To be enumerated as an artist in the Census, one has to identify his artistic occupation as his "chief job activity or business last week,"(5) not as his sole job for the entire year. ONLYART also permits an examination to 15

be eondueted of any possible dilution of an artist's work effort due to the holding of non-artistie jobs, for whatever reason. For comparative purposes, a fifth variable, ARTIST, is also analyzed. ARTIST equals 1 if the questionnaire respondent indieated that his artistic occupation was his "prineipal profession," 0 if he did not. The percentage of artists who pass through each filter and the pairwise relationships among the filter rules are shown in Table 1. The most liberal of the rules is INCOME, with 80.4 percent of the sample having reported earning some income from artistic work during the prior year. ARTIST is not far behind, with 74.1 percent indieating that their art is their principal profession. The most stringent rule is ONLYART; only 24.1 percent worked solely as artists during the year. Also, it is clear that artistie expenses play a major role in determining financial sueeess as an artist. Although o v e r 80 percent earned income as artists, only 47.7 percent passed through the IRS filter, i.e., had artistic income whieh exceeded the expenses incurred while earning it. If this group of artists had completed the Census questionnaire a t the same time that they participated in this survey, those who would identify their occupation as t h a t of artist on the Census form would constitute less than the "/4.1 percent who simply claimed artist as their occupation but more than the 24.1 percent who worked only as artists during the year. Whatever this unknown percentage aetuaUy is, it is clear that a significant number o f persons responding to this survey would be enumerated as something other than artists by the Census. Leaving the appropriateness of the Census definition aside, information about persons on the "fringe" of the artistie p r o f e s s i o n , whieh may be relevant to both researchers and polieymakers, is lacking in Census statistics. Some information about this group may be gleaned by examining the characteristics of those rejected by the filter rules, espeeially ARTIST and ONLYART. An examination of the pairwise cross-tabulations shows that the filter rules are positively related to each other. A corrected ehi-squared test applied to each crosst a b u l a t i o n shows that all but one are significant at the .999 level. The sole exception, UNION vs. ONLYART, is 16

Table I Pairwise ~ i s o n s of F i l t e r Rules (all figures in percentages) ,

,,

,,

L~ICB 1

INCQqE

Y~s

Nd

i 63.7

I0.8

39.3

35.4

16.9

8.7

8.2

17.2

'~

ONLYART NO

YES ARTIST NO

42.9

25.7

22.5

5_I.6

74. I

16,6

14.8

,,_1..?

2~1

2s.?

YES

47.4

n.2

54.6'

33.3

21.6

58.9

80.4

NO

0.0

19.5

2.7

9.4

9 2,0

17,,6

19.~

YES

42.1

NO

15.6

20.3

16.8

30:8

47.4

22.4

6.1

46.6

52.6_

YES

12,8

44.5

57.5

NO

5.1

37. 6

.42.5

[

IRS

DNI(I~ YES ONLYART NO

' at.t ~.'ou-cem.d.ati~'~ ' ~ r~r

ar

}~r pe~"~.'mlwg a.--~ioU cwd.y,

1'7

significant at the .998 level. The phi statistic, a measure of the strength of each relationship, ranges from a low of .135 for UNION vs. ONLYART to a high of .466 for INCOME vs. IRS.(6)

T-Tests of the Socioeconomic Variables In this section, the mean values of a variety of socioeconomic attributes of the artists are compared, using t-tests, for the "yes" and "no" groups classified by each filter. Throughout the paper, the yes groups will be r e f e r r e d to as more "committed" or more "successful." This terminology is used for taxonomic purposes only and is not meant to imply a lack of dedication among artists classified into a no group. For ease of reference, all the variables are defined in Table 2. Essentially, these variables can be grouped into four areas: human capital, work effort, income, and artistic costs. Human capital variables describe the amount of education and experience each artist possessed at the time of the survey. Education is expressed only in general terms, i.e., by years of schooling and by highest degree attained. In the survey, information was also collected on whether or not artists participated in specialized artistic training, both as part of and in addition to their general education. Because the inclusion of discrete discriminating variables in a diseriminant function violates one of the underlying statistical assumptions of the model, no discrete variables are analyzed in this section or used in diseriminant functions in the following section.(7) It can be noted, nevertheless, that 84 percent of the surveyed artists received some form of specialized artistic training as part of their formal education and 74 percent received such training outside their formal education. The work effort variables include weeks worked in artistic, arts-related and nonarts-related jobs.(8) Also, artistic time worked, in terms of hours per "typical week," is broken into seven areas. The first four, hours spent performing, rehearsing, practicing, and looking for work, apply to performing artists only. The last three apply to nonperforming artists only. A significant number of artists reported working in both performing and nonperforming

18

Table 2

Definitions of Socioecon~c V a r i a b l e s

~_-~inition

yarS~ble

l a a ~ e r o f y e a r s o f formal schooling. h i ~ e s t degree e a ~ , co~ed m m ~ r ! ~ n y . ~ e at which a~.Ist~c training began. Present age.

Weeks Weeks Weeks ~oors E ~P~C

~CgG

N~ZNC

S;~JSI~C

per per per per

Hours per ~ours Hours Mours Hours Ho~s

Per per per per per

y~r year year ~k

~ k i n g as an artist. ".~cking i n arts-related jobs. working in non-arts-related jobs. m ~ m t perfoming. week spent z,~eazsing. wee~ spent practicing. week spent locking for artistic work. week spent c~eating a r t works. week spent taking lessons. w e ~ s~ent mnketi.g art wo~s.

Weeks in the year In ~ artistic ~ was earned. A . m a l artistlc income. Annual income fz~m arts-related work. Annual income fz~m non-ar~-relabed work ~ Annual n o n - ~ income received by artist. ~ m u a l inco~ earned by spouse and other family n ~ b e r s . Annual income ~ . i v e d ~ b a r t e r i n g a r t works.

~ P _ t s t i c Cost

D~L RENT ~CI~STR

on or~dnlzed s ~ . A n m m ~ a r t i s t i c travel expenses. mmual a ~ t ~ c ~ l i ~ L n g e x ~ . a r t i s t i c dL1plica~lon expenses. insm-ance o f art: ~ r k s o r implements. ~ t o f artistic space. and repair costs of i n - - i s . on cosOmes or ~ ~ts. spending on a g a r s ' f e e s and a:smlssions. ~ting on co1~right fees. spem~tng on books and research m t e r L ~ l s . ~mal on v i s u a l a r t m a t e r i a l s . A m w a l ~ - ~ l i n g on v i s u a l a r t t o o l s . Annual o t h e r a r t i s t i c spendir~.

19

fields. For example, almost 40 percent of choreographers, composers and playwrights indicated that they worked as performing artists as well. The income variables include labor income earned by artists, broken i n t o the same categories as weeks worked. They also include nonlabor income, income earned by other family members, and ineome from bartering art works or services. Further, the number of weeks for which artistie income was earned is included in this group of variables. The artistic eosts group includes 14 categories of artistie spending. Some are general in nature, but many represent costs likely to be incurred by members of a specific artistic oeeupation or group of occupations. T h e mean values of each of the soeioeeonomie variables, elassified by eaeh of the filter variables, are shown in Table 3. Before analyzing differences between means, several characteristics of the artists in the sample should be noted. Common to this and other studies of artists there is found an unusually high educational level eoupled with very low earnings within the artistic profession. New England artists possess on average slightly less than a Master's level education, but earned only $6,420 in artistic ineome in 1981. These artists, however, are not poor. Their personal income from other sourees summed to $15,644 anal t h e i r total household ineome averaged $27,297. Clearly, income from other jobs and from other household members is an important factor in sustaining artistic careers. The apparent neeessity for holding nonartistie jobs is emphasized when it is noted that summing weeks worked in all types of jobs leads to a number greater than 52. Obviously, nonartistie jobs are held not only when no a r t i s t i c work is available or possible, but simultaneously with artistie work as well. The necessity of nonartistie income is emphasized by noting the magnitude of artistie expenses. Total artistic expenses averaged $3,554, or 55 percent of artistic income. It can be seen in Table 1 that less than half the artists in the survey could pass the aforementioned IRS test of earnings exceeding expenses. Although nonartistie work and income must be viewed as neeessities for most artists, one would expeet that a filter rule designed to distinguish more from less

20

TABLE 3 Comparison o f Hearts o f Socioeconomic V a r l a b l e s Under t h e F l l t e r s I

W

c~prllpL

~

N

iI

I

'i

16.6a 3:'114.5~J 39.3a 39.0 a 16.5 s $.3 a) 2.0~t 2.18 2.4c Jo.st) 123.5'a11.0c 4.5a

BO

9 15.5

17.1 4.2

41.7 24.2

Hi.3

22.2

2.2

2.4

1.9

}0.3 [10.6 i 0.8

t6.~' 3.~" 14.~ 39.~ 3s.'~ ty.~ lo.sa 2.~' 2.~' =.~' 0.~' ~.l~ ~0

:27.1 4.2

21i.S 39.4 7,3.3

19.1

18.6

0.4

1.0

1.3

O.l

140

~7 91

4."~. 15,4 ~ . 3

IIm 1,.6r !3., /lo., 4o.,' 3.9 ;]J..l 37.4

2.0

I 31.4 20.? 16.1

,.~

~,.~ , . , '

28.3

2S.6 15.6

0.? l.;tJ 2.0 0.4 18.9 "Jl.2 3.3

Lo.~'ly., ,Jl~ b 2.~ -

'

UllZ~ 4 16.8

4.1a

L5.? 1.3

1

1.9

ls;O 3.4" 17.0 43.0 43.,~ ~ ,.r

0.~

5.3

.9

7.3 t 6.2

l.s

1.6

Ii ~

1.r 0.3 27.," 0.8

.

:IIL'/N~ NO

,,.o ,.1 t4-6i".'1"3.4t".L ,s.,,,.,(,.o

,., o.6_,,.,

;.o ,.4

l L e v e l 8 o f s J g n i f i c s n c e a r e d e s i g n a t e d by: a - s i g n i f i c a n t a t t h e .99 ] e v e 1 ; b - s i g n i f i c a n t s t t h e 195 l e v e l | and e - 8 i R n l f l c a n t a t t h e .qo l e v e l . using t-te~ts 2Annual income v a r i a b l e R e r e i n t h o u s a n d s o f d o l l a r s . 3Annual a r t i s t i c

cost variables

a r e i n hundreds o f d o l l a r s .

4For d a n c e r s , m u e l e l a n 8 end a c t o r s o n l y .

21

Table

,,~ ,~.V!~,.~,,/. ,i~ ~...~,

3 Continued

,~,~i~'i,~'l,~o~ ~-~o.**

~l, ~, .,~ ,~1 o.~ , ~ , ~ , ~

,~t,~, ~1,..o~t3~~

I

i

o o o o ~31,.,i2,

j

i ~,.o:~.,'~.~,-'~,.~

,,,io,

.~

,~

o,

.71

,,

351,,

,o

!

o~l,,

o~ o ,

2.).a 3.0a ].,2 *o-~'io.~-~.~'~." 2.4 111

46

,.,1,,.~ ,.,.

o,

..'t,? .~'~-~

o,

oo o ,

I ,11,,

0.6`3 4.? a 0.1 1.3

I

7.S 3.2 a 2.1

15

,. ,., io., ,., ,.~,.o o.,~,.,I,.o i-.~,.,'t ..~. o.,~I,.~'1 o.~'io-,.~" /1 ,/ ,, o.9"

. . . . .

iJ IT.II

22

o.1 I

,-, t~,.§ 4. ~'13.'~

committed and/or successful artists would show that the more committed/successful ones would have less work time in and less income from nonartistie sources. With one exception, each filter rule shows significantly less work effort and income from non-artistic sources among the no group. Obviously this conclusion must be true for ONLYART, since this is the basis of differentiation between the yes and no groups. In examining variables that signify a positive commitment to artistic work effort, it can be seen that for o.11 five filter rules the artists passing through the filter worked significantly more weeks as artists. In all cases, artistic income and weeks earning artistic income are significantly greater as weU for artists passing the filter test. Since artistic income is the criterion for the INCOME filter, ARTINC and WKSEARN are zero for the no group of artists. The results are less consistent for the variables describing hours per week devoted to artistic work. With o.ll filters except UNION, the difference between means wiU be reduced somewhat because performing and nonperforming artists are grouped together in these computations. The UNION filter test is applied to performing artists only; t-tests are not made for non-performing hours. Regardless, under the ARTIST, INCOME, IRS and, naturally, the UNION filters, all four types of performing hours are greater for the yes groups. A n but one comparison is significant at the .95 level or higher. For the ONLYART filter, three o f the four performing hours categories are less for the yes group. The one that is greater for the yes group is not significantly greater. This result is caused in part by the dilution of performers with nonperformers before computing these statistics. But if differences i n performing hours are compared for performers only under the ONLYART filter, it is found that HRPRAC and HRLOOK are less for the yes group, and HRRHRSE, though greater, is not significantly so. Performing artists passing through the ONLYART filter a r e thus different from performing artists who pass through other filters in that the former devote more of their time 9 to performing and less time to rehearsing, practicing and looking for work. This may very well be a necessary 23

condition of earning all one's labor income from artistic work. No data on nonperforming hours are presented for the UNION filter rule, as only performing artists are analyzed under this rule. Among the other four filter rules, both HRCREATE and HRMKTG show significantly higher values for the yes groups. In three of four cases, the HRTRAIN variable is lower for the yes groups. For artists in this survey, greater time spent in receiving education and training is associated with an earlier career stage. This in turn is likely to imply less commitment to or success from one's art, as judged by the filter variables. Although one might expect that artists passing through any of the filters would be likely to incur greater artistic costs, this might not be the case when one considers that several of the artistic cost variables arc relevant only to some of the artistic occupations in the survey. Though differences between means for the yes and no groups for relevant occupations might be significant, lumping all artistic groups together in the actual calculation can reduce o r blur such differences. However, there is little evidence from the data that this occurs. What is obvious is that the IR$ filter rule highlights differences in expenditure behavior of artists not d e t e c t e d by the other filter rules. Under t h e IRS rule, four of the fourteen artistic cost variables are less for the yes group and two are equal for both groups. Under the other filters, no more than t w o artistic cost variables are ever less for the yes group. With respect to other variables, the characteristics of the yes and no groups under the IRS filter appear to be similar to those under o t h e r filters. However, the artist who passes through the IRS filter, i.e., earns artistic income in excess of artistic costs, does so by both maximizing income and minimizing artistic costs. The findings that conform least well to theoretical expectations are found i n the human capital variables. Specifically, both measures of formal education, YRSCHL and HIDEGREE, are less for artists passing through the filters in all but one instance, One might argue that because these two variables measure the amount of general education, they do not capture the amount of artistic human capital acquired by the artists. What may be more 24

relevant is the extent of specific training received by artists, either inside or outside of the formal educational process. Though not reported in Table 3, it can be noted that artists passing through each of the filters exhibited a lower incidence of training outside the formal educational process. For artists receiving artistic training as part of the formal educational process, the results are quite mixed. Artists passing through the ARTIST and UNION filters were more likely to have had this training; artists passing through the INCOME, IRS and ONLYART filters were less likely to have had i t . Thus, the evidence reported above is consistent with the conclusion that more successful and/or committed artists are less well educated than their less successful and/or committed counterparts. Lest this conclusion seem to turn human capital theory on its head, it should be added that although the simple correlation coefficient between years of school and artistic income for these artists is negative and significant, the simple correlation coefficient between years of school and arts-related income is positive and significant, and between years of school and nonarts-related income, it is positive but not significant. Apparently artists with more education receive differentiaUy greater financial rewards from their education when working in nonartistic as opposed to artistic fields.(9) The remaining two human capital variables, ARTAGE and AGE, capture to some degree the experience component of human capital. Obviously, it is impossible to quantify completely an artist's job experience. Because an artist's career consists of a continuous process of training, practice and production, the length of the career and the age of the artist can serve as proxies for experience. In fact, for all filters but ONLYART, ARTAGE is less for the yes group. For all filters except ARTIST, AGE is greater for the yes group. Also, one can create a variable by subtracting ARTAGE from AGE. This variable estimates the length of the artistTs career. Though not reported in Table 3, this variable is significantly greater for the yes group at the 95 percent level of above for all five filter rules. 25

Diseriminant Functions for the Filter Variables For each of the five filter variables, a linear discriminant function is estimated. E a c h function is created by permitting any of the 35 variables in Table 3 to enter, in stepwise fashion, by passing a criterion of having a t-statistic of at least 1 . 0 . The functions appearing in Table 4 are the ones generated by this process with the highest canonical correlation. T h e n the population is broken into the nine occupational groups, and linear discriminant functions are estimated for each of the filter variables and each of the occupational groups except dancers.(10) These functions are not tabulated here, but they are compared in Table 5 in terms of their ability to classify artists correctly into the yes and no groups. Although only those discriminant functions with the highest canonical correlation are shown in Table 4, two functions are shown for the INCOME and ONLYART filters. In estimating a discriminant function for the INCOME filter, it is obvious that the filter will be strongly correlated with the ARTINC discriminating variable, since the INCOME filter rule was formed by collapsing the ARTINC variable into dichotomous form. Such a strong correlation will also exist between the INCOME filter and WKSEARN. Thus a discriminant function is shown for INCOME with the ARTINC and the WKSEARN variables both admitted and excluded. Similarly, because of the obvious correlation between ONLYART and WKARTREL, WKNONREL, ARINC and NARINC, discriminant functions are estimated admitting and excluding these variables. All functions in Table 4 are significant at the .999 level, using an F-test. In general, the discriminating variables predict membership in the yes and no groups in the same manner as they did in the t-tests. (In the functions, a yes response is coded as 1; a no response is coded as 0.) Thus, a positive sign for a discriminating variable implies that larger values predict membership in the yes group; similarly, larger values of a variable with a negative sign predict membership in the no group.(ll) Among the discriminant functions in Table 4, there are more similarities than differences in the pattern of variables which enter 26

I

I

I

~

.

..--~ ~ ,'~ ~ ~ ~

~,~i~,~,, 9

;

,"

9

~

i I

w - I ,

, ~ } 9' ~

;

s

r

;

9 .

,:

i

i ~.

I.

:

9

:

j,.i 2?

.0p

I

I

I

I

I

gzu

I

I

I

'ill

i.

I

I'~

I

I

I

'l'l'i'

I

I

'lll~ll

, I

1 <1 ,IP

I

I

t

.I

I

I

1

U~o ;

I

.,"

;

I

I"

i

I~

I~

I

I

I .'i-

I I I

t

:::1

I

~o "l~

I

I

~

"~

I

I ~ ' ~

'

lt~

7

i"

U

,'.

i

=llIsi'~

--:'~"

~llI . I i

,"

I

I

t

~

II

I

, I" ....

,

I

~

h"

~i o

J

9

,

~at

I

28

I

them. The variables which measure financial success as artists, ARTINC and WKSEARN, enter every function (except where they have been excluded) and in all but one ease with a positive sign. YRSCHL enters every function with the expected negative sign. The variables describing nonartistic weeks worked enter with the expected negative sign, but this is not always the ease for the variables describing nonartistie income. Where they enter, the hours worked variables behave in essentially the same way as they did in the tests of means. The artistic cost variables exhibit different patterns than they did in the tests of means. All the artistic cost variables that enter the IRS diseriminant function are negative, as are almost all that enter the ONLYART(a) and (b) functions. In the INCOlVlE(a) and (b) functions, artistic cost variables rarely enter. When the diseriminant functions formed for the filter rules are compared, it is seen that they correctly classify between 74 and 88 percent of the artist population into the yes and no groups (using, in each ease, the b e t t e r of the two functions for INCOME and ONLYART). The discriminatory power of the ONLYART and INCOME filters naturally is less when highly correlated discriminating variables are dropped from the functions. The relative effectiveness of the discriminant functions, as measured by the percent of eases correctly classified, is not the same when applied to the eight occupational classes individually. This comparison may be seen in Table 5. At the bottom of this table is tabulated the average rank that each filter rule attains, relative to the others, over all occupational classes. Using this comparison, it can be seen that the ONLYART(a) filter performs best, correctly classifying the highest percentage of artists for six of the eight occupational classes. Performing less well, in decreasing order, are the INCOME(a), IRS, ARTIST, ONLYART(b), INCOME(b), and UNION filters. Thus, when applied on an individual occupational basis, the INCOME(a) filter performs considerably better and the UNION filter performs considerably worse. As a rule, the filters classify a higher percentage of artists correctly when applied to each occupational class individually than when applied to all 29

i' ~

I~

9Ei

R

~.

,

~

~I

,

0

1) S U

+|~

~.~

t~

~

,~ p.9

~

t~

)

30

r~

Q

I~,

c0

o

M

artists. One may conclude that there is greater homogeneity within occupational groups than among them, despite significant numbers of artists in the survey possessing multiple occupational skills. Conclusion In this paper it has been demonstrated that occupational filter rules can successfully partition artists into groups signifying more or less success in and/or commitment to their profession. The filter rules were applied to a broad sample of artists, and to the same sample broken down into eight occupational classes. Attributes of the artists that were found to be most relevant in distinguishing among the two groups of artists created by each of the filter rules were those associated with artistic income and weeks worked, both as artists and as members of other occupations, hours worked per week as artists, the amount of formal education and, to a lesser extent, costs incurred while working as artists. The level of these attributes possessed by the groups passing through and not passing through each filter conformed to a priori expectations with the exception of the variables describing the amount of formal education. Artists passing through each filter had less formal education than those not passing through. The analysis in this and other work by the authors suggests that greater amounts of education increase the financial rewards to artists in nonartistic work but do not increase rewards from artistic work. The filter rules which performed best, both for all artists combined and for artists broken into occupational groups, distinguished 1) those that worked solely as artists from those that didn't, and 2) those who had artistic income in excess of artistic costs from those who hadn't. There were some differences in the relative efficiency of the filters when they were applied to all artists combined in contrast to when they were applied to each occupational group. In general, any particular filter rule was more effective in distinguishing between the two groups of artists for an individual occupation that it was for all artists combined. Northeastern University 31

This paper was made possible by financial support from The National Endowment for the Arts. An earlier version of t h i s p a p e r was presented at the 1983 Conference of the New England Business and Economics Association. FOOTNOTES .

While the Census definition encompasses most artistic occupations, it excludes one occupation used in this study: craft artist. The Census also includes architects, radio and television announcers, acrobats and translators in its artist category. All of these are excluded from our study.

.

Specifically, the 1980 Census long form questionnaire asked that the respondent "describe clearly this person's chief job activity or business last week. If this person had more than one job, describe the one at which this person worked the most hours," The wording was exactly the same in the 1970 long form questionnaire.

.

The results of this survey are reported in (Wassail, et al. 1983). Artists who were sent the mail questionnaire were chosen by a stratified random sample from mailing lists from diverse sources such as art school alumni rosters, performing artist union membership lists, performing organizations, lists of participants in art and crafts shows and fairs, membership l i s t s of artist organizations, and lists maintained by the state and regional arts councils. The sampling distribution was controlled by Census estimates of the proportion of all artists in New England in each major occupation and in each state.

.

The survey year was actually two years: 1980 for Massachusetts artists and 1981 for all other New England artists, as the survey was administered in two stages. For most data collected, the fact that they refer to two different although adjacent years makes 32

little difference. However, income and cost data for 1980 for Massachusetts artists were adjusted upward, using Massachusetts personal income data and the Boston Consumer Price Index, to make them consistent with the 1981 data collected from all other artists. 5.

Cf. t h e discussion in f o o t n o t e 2.

.

The phi statistic corrects for the fact that the value of chi-square is proportional to the number of cases by adjusting the ehi-square value. Mathematically, it is 0 = (X/N),

and takes on a value of 0 when no relationship exists and a value of +1 when the variables are perfectly related, which would occur when all cases fall only on the main or minor diagonal. .

.

.

See, for example, the discussion in Goldstein and Dillon (1978). In the questionnaire, arts-related jobs were defined as teaching or coaching in an artistic discipline, arts administration, and jobs that might be closely connected to the arts, such as management or clerical work in an artistic organization, usher, ticket taker, etc. Nonarts-related jobs were defined as any job other than artistic or arts-related. In this paper a "nonartistic" job is one that is either arts-related or nonarts-related. These relationships between education and earnings remain substantially the same when estimated in a fully-specified earnings function. See A l p e r and Wassall (1981) for more detail.

I0. The number of dancers in the survey was too small to permit estimation of a separate discriminant function, 11. When the discriminant function is created, the sign of any discriminating variable shows its relationship to 33

the values of the two group centroids formed during the estimation process. The numerical values of the centroids thus formed may be inversely proportional to the 0, 1 values arbitrarily assigned the no and yes groups. Whenever that occurred, the signs of the discriminating variables were reversed so that all of the discriminating variables in each function in Table 4 could be consistently interpreted with respect to the 0(no) l(yes) assignment. REFERENCES

Alper, Neil O., et al. Artists in Massachusetts: A Study of Their Job Market Experiences. Boston: Massachusetts Council for the Arts and Humanities, 1981. Alper, Neil O., and Gregory H. Wassall. "Determinants of Artists' Earnings: A Comprehensive Study." Presented at the Southern Economic Association Annual Meeting, New Orleans, November 1981. Felton, Marianne Victorius. " T h e Economics of the Creative Arts: The Case of the Composer." Journal of Cultural Economics 2, 1 (1978), 41-61. Goldstein, Matthew, and William R. Dillon. Discrete Discriminant Analysis. New York: John Wiley and Sons,

1978. Kingston, Paul W., et al. The Columbia University Economic Survey of American Authors. New York: Columbia University Press, 1981. Ruttenberg,

et

al.

Surve~ of

Employment, Under-

empl0.yment and Unemp.loyment in the Performing Arts. Human Resources Development Institute, AFL-CIO, 1978. Wassall, Gregory H., et al. The Arts and the New England Economy. Cambridge: New England Foundation for the Arts, 1980. 34