Validity of Program Types
the validity of The program type model in studies of television
program diversity study
By
Doh-Yeon Kim
Doctoral Candidate
and Augie Grant
Associate Professor
Department of Radio-TV-Film
CMA 6.118
The University of Texas at Austin
Austin, TX 78712-1091
512/471-6640
Fax: 512/471-4077
email: [log in to unmask]
[log in to unmask]
Running Heads: Validity of Program Types
Paper prepared for 1996 AEJMC Convention
April 1, 1996
the validity of The program type model in studies of television
program diversity
Abstract
This study attempts to validate the program type variable
that is typically used in television diversity research by examining the
discriminability of the variable to predict audience ratings. A typology of 24
program types is tested on more than 12,000 half-hours of programming from all
broadcast and most cable networks. The results confirm the program type model,
and the manner in which that model can be applied in a variety of research is
discussed.
Introduction
Over the past decade, bBroadcasting deregulation and the
technological development of alternative delivery system such as cable and DBS
(direct satellite broadcasting) have led to a dramatic increase in the number of
television channels expansion especially in most advanced countries and newly
industrialized countries since last decade. Parallel to this trend, there have
been discussions about how diverse the actual programming is in a multi-channel,
multi-media environment (for example, De Jong & Bates, 1991; Grant, 1994). One
of the key debates in the resulting discussion of diversity is how to
operationalize this variable.
In studies of television diversity study(and, indeed, in
many studies of television programming and audiences in general), it has been a
general practice to construct a typology of different program types in which are
different inprograms having similar subject matter and format are grouped
together as a "program type" (for example, Levin, 1971; Litman, 1979; Frank &
Greenberg, 1980; Rubin, 1984; McDonald and Schechter, 1988.), which are assumed
to beThe assumption behind this practice is that each "program type" is
differently preferred by television viewers (Greenberg & Barnett, 1971).
Although this practice is criticized by some, it is widely accepted because it
enjoys a high degree of face validity.
Despite the widespread use of this practice, little research
has been conducted However, what has hardly been done is to empirically examine
whether different program types are differently preferred and watched by
television viewers. As Greenberg and Barnett (1971) pointed out, "to the extent
it is useful to classify programs into '"types,"' these should be based on
audience behavior." The purpose of this study is to investigate whether viewers
showdemonstrate different levels of preference to different program types.
AnsweringTo answer this question is valuable for at least two reasons. First,
it can test the validity of the program type model for television diversity
study. Second, it can provide a evidenceclue to buildhelp in constructing a
standardized television program typology for future studies.
Television Diversity Studyies and the Program Type Model
How to define diversity has been a central issue in
television diversity studies because diversity of television programs is by no
means a self-defining term. Conceptually, diversity has been defined as "the
degree of choice available to the viewer" from television programming (Grant,
1994, p.53). However, a question may arise if separately identified programs
can be regarded as choice. In a sense, each television program has its own
uniqueness (Cass, 1981). For example, can three different sports programs, such
as football, basketball, and baseball telecasts be regarded diverse? Or, can
three separate news programs be treated as providing the same or less diversity
thanas three different types of program (e.g. news, sports, and quiz)?
To solve this conundrum, Glasser (1984) distinguished the
concept of variety from diversity. According to himHe suggested that, variety
only offers only intra-format diversity, which is a variation of the same
cultural product, not genuine cultural diversity of the kind which relates the
diversity of what is on offered to some external standard of differentiation in
the community or society.
Glasser's (1984) distinction coincides with Greenberg and
Barnett's (1971) program type model. They argued that programs could be
aggregated into categories reflecting the relative preference placed on them by
the viewers, which was called the program type model. They explained that
diversity could be defined as differences in program type if "we can specify
program types within which we can reasonably believe that variations of quality,
advertising, timing, author, performer, suppliers and degree of competition do
not contribute to diversity" (p.90). Their expression of variation can be
interpreted as Glasser's(1984) intra-format variety. In the program type model,
each program is presumed to be categorizable into a certain program type, and
each program type provide a different aesthetic and cultural value to television
viewers.
Previous studies ofn television diversity have usuallymostly
adopted the program types model in measuring diversity. Typically,
researchersreseasrchersthey constructed a set of program type categories which
are exhaustive and mutually exclusive. They assumed that individual programs
canould be usefully subclassified among a set of program categories such that
programs across types arewere perceived by viewers as offering poorer
substitutes than did programs within the same type (Levin, 1980). In other
words, viewers are assumed to have more similar preferences for the programs of
the same type than those of other types. The principal categories were normally
defined by program packagers, ratings services, and stations owners, and
modified by the researchers for their purpose of research.
Because program types are in the first instance defined by
the program supply industries and rating services, there has been an open
question as to how homogeneous they were from a consumer standpoint. Some
researches have addressed that question. For example, Levin (1980) measured the
differences in the audience members' preferences for each program type in
network television programming statistically and confirmed that type differences
did appear to matter to audiences significantly. Youn (1995) constructed
program types based on audiences' attitude on each program. These studies
reveal that the consideration of viewer's standpoint in program type model is
necessary and can be methodologically executed. However, it should be noted
that each of the above studiesy has its own limitations in that Levin (1980)
studied onlyjust prime time (from 8:00 pm to 11:00 pm) programming of the three
commercial broadcasting networks, and that Youn (1995) measured not actual
viewing behaviors but attitudes of viewers rather than actual viewing behavior.
In existing diversity studies, the number of program types
for the measurement of diversity ranges from 9 (Litman, 1979), to 12 (Youn,
1994), 21 (Levin, 1980), 24 (Grant, 1994), and 38 (Jackson, 1986).
Inconsistencies in the categories of program types among the studies stem partly
from time gaps between studies from the early 1970s to the mid 1990s, during
which new program types were invented and introduced. At the same time, these
researchers suggest that it is very difficult to make optimal program
categories. As a matter of fact, there can not be a single correctright answer
to the question of program typology. For that reason, it has been said that the
validity and reliability of measuring diversity depends on the degree to which
valid and reliable distinctions between content categories are made (Levin,
1980).
The key question, then, is whether the conceptual program
type variable can be supported by empirical evidence. That evidence could take
many forms: audience preferences, content analysis for formal features, cluster
analysis (based upon viewer preferences), etc. Perhaps the simplest measure is
to determine the amount of variance in program rating which can be attributed to
differences across program types as opposed to differences within program types.
Any such significant difference would provide a strong measure of validity for a
specific typology of program types, as well as for the program type model in
general.
Research Question and Hypothesis
The mMain criticisms oftoward program diversity studies
which apply the program type model have been (1) program types are based upon
producers' point of view; and (2) there may be as many or more differences
perceived within a single program type, as among (or across) different types.
This study does not attempt to add another to existing television program
typologies by constructing its own. Instead of that, this study
tryattemptsatempts to test the validity of the program type model by asking
whether different program types are differently preferred and watched by actual
television viewers and testing it with an existing typology.
Greenberg and Barnett (1971) stated that, in the program
type model, audience preferences are assumed to be distributed unevenly over
program types. To test this, this study proposes a null hypothesis like the
following null hypothesis:
H0 : As many or more differences will be perceived within
program types, as among (or across) different types by
television
viewers.
Based upon the null hypothesis, the research hypothesis is
proposed as:
H1 : More differences will be perceived among (or across)
different program types than within program types by
television viewers.
Method
To test the above hypothesis, this study used the television
program typology developed by Grant (1994) as a basis. He developed thea
typology of 24 different subject matter categories[1] from those used in
previous studies such as Levin (1971) and Litman (1979). Along with 24-category
typology, this study used Grant's (1994) data base which covered 16 full days of
television programming (7 a.m. to 1 a.m.) appearing on 41 nationally distributed
cable and broadcast television networks from January 1, 1986, through June 30,
1986. Although this data base is now somewhat dated, it is useful because it
permits the analysis of inter-type differences extended to cacross a range of
channel typesable, including programming, with the ratings data offor every
program. The channels in the data base include all nationally distributed
broadcast networks (including PBS), pay cable networks, and those nationally
distributed basic cable networks (including superstations) that reached at least
10% of all U.S. cable subscribers for which program schedules and ratings data
were available. The 180-day sample period was subdivided into four quarters of
equal length, and two weekdays and two weekend days (one Saturday and one
Sunday) were randomly chosen from each quarter, to constitute the 16 sample
days. Coders familiar with television programming used program names and
descriptions to classify each 30 minute segment for each network into a single
program type. When a coder was unfamiliar with a program type, a call was made
to the program service to obtain more detail about the program in question.
Intercoder reliability was determined by having a different coder recode a
random sample of programming. For the program type variable, intercoder
reliability was computed by drawing a random sample of two periods from each
network (n=80), which were independently coded by each of the four coders after
excluding all cases that had been left blank by a coder who was unfamiliar with
the program in question. Intercoder reliability was .80 using Scott's pi
(Stempel, 1981).
To usemake the original data base fit for this study, the
day of the week, the time of the program aired, the channel type (broadcast
network, basic channel, superstation, and premium channel), the program type
(among 24-type typology), and the ratings score of Nielsen program data of each
30-minute television program unit were recoded. Data were analyzed using SPSS
for Windows, v. 6.1.
Results
To determinedo pairwise comparisons of the impacts of
program types on audiences' actual viewing, a regression analysis was executed
using a audience rating score as a dependent variable, and dummies of days of
the week, programming times, channel types, and program types as independent
variables. Table 1 shows the result of the regression analysis. In
interpreting the result (as well as the grid of pairwise comparisons in Table
2), the following background information should be kept in mind:
1. The impact of each program type is estimated by the
use of twenty three dummy variables, relative to the
situation comedy
base, (chosen becauseon the basis of its had being the type
of highest
average rating).
2. The impact of each of the three time segments (7:00
a.m.-10:00 a.m., 10:00 a.m.-8:00 p.m., and 8:00 p.m.-11:00
p.m.) is
estimated with dummy variables, relative to that of the
segment from
11:00 p.m. to 1:00 a.m.
3. Channel type impact, using dummy variables for basic
cable, premium cable and broadcast network, is measured
relative to
superstation.
4. Day-of-the-week impact, using six dummy variables, is
measured relative to Saturday.
To ensure the reliability of the final equation, the present
study executed sequential regression analysis, running the equation first with
three channel type and twenty three program type dummies, then, adding three
time dummies, and finally, adding six day-of-the week dummies. In the result,
the stability of coefficients and t-values across all three equations was
striking. The explained variance adjusted for degrees of freedom (adjusted R
square) rose from .48304 for equation 1, to .49716 for equation 2, and to
.501075 for the final equation which is shown in Table 1. In Table 1, column 2
listrepresents variable names used as independent dummy variables in the
regression equation, column 3 containstands forthe regression coefficients ofor
each variable, and column 4 and 5 indicate t-values and significance levels of
corresponding variables. With only one exception (Science Fiction/ Fantasy) the
coefficients for all program typesall other pairwise comparisons with situation
comedy reveal statisticallywere significant differences in audience size
normally at p<= .01 confidence level. Thus,at can be understood that in 22 out
of 23 tests performed, the net differences in viewersaudience rating indicate
that the pairs of program types examined are perceived as different by viewers.
who obviously do not watch both.
However, Table 1 provides only 23 comparisons of situation
comedy with the other 23 types. The extensive comparisons among program types
require additional regression analyses setting each program type as the base.
Table 2 is a full grid of 552 pairwise comparisons after the further analyses.
Each cell denotes the regression coefficient of the column program type when the
row program type is set as a base. For example, the first upper-right cell
(-.46) of Table 2 means that the regression coefficient of 'general comedy /
variety' variable, when 'situation comedy' variable is the base of program
types, is -.46 as is shown on the 18th line of Table 1. Shaded cells represent
significantly different program type pairs at least p<= .05 level. Shaded cells
with bold numbers stand for significantly different program type pairs at
p<=.01 level. Overall, 372 pairwise comparisons among 24 x 23 = 552 possible
pairs revealwere found to be significantly different (67.394%).
These complete results of all conceivable type comparisons
arecan be more briefly summarized for present purposes in Table 3, in which 24
program types are ranked according to the proportion of pairwise comparisons
that yield statistically significant differences in audience size. In that
table, column 1 and 2 represent the identification numbers aand names of
program types used in the analysis, and column 3 containmeans the percentages of
each program type as a whole. Column 4 denotes the numbers of significant pairs
in the test result, and column 5 reportstands for the proportion of significant
pairs in 23 tests. The summary indicates:
1. Eight out of 24 types have significant differences
from types in 19 to 22 pairwise comparisons (out of the 23
tests run per
type), which are above 79 percent of the basic comparisons
made.
2. Another 12 out of 24 types have significant
differences from other types in 13 to 17, or 50 to 75
percent (13 to 17)
of each set of 23 test runs.
3. Among the other four types, three have significant
differences forrom 11 types or almost 50 percent of tests.
4. In only one case (Agriculture), significant pairs are
well belowmuch fewer than 50 percent of the comparisons,
which might be
dueowing to the extremely low proportion of programs in the
Agriculture
category (.001 percent of the total programming).
Discussion
TheIn conclusion, results reported in Table 1, Table 2, and
Table 3 prudently imply that the program type is an important factor, along with
other factors like channel type, time-of-the-day, and day-of-the-week, into
determininge the actual television viewing of the audiences. MAnd more
importantly, program type differences do matter to viewers, and viewers perceive
the inter-type differences. Before discussing the implications of this finding,
it is important to note the limitations of this study.
Although the data consisted of a wide variety of channel
types, program types, and time periods (totalingtotalling almost 20,000 cases,
with each case representing one half hour of programming), all data were derived
from a single six-month time period. Since preferences for different program
types change over time (McDonald and Schechter, 1988) the pattern of results
observed for this time period might be different from other time periods. Also,
the Finally, the use of Nielsen ratings data in the analysis subjects the
results to all of the limitations related to the use of ratings analysis,
including responserespone rate and other ratings biases (for a detailed
discussion, see Webster and Lichty, 1991).
On the other hand, the data set analyzed contained almost
20,000 cases, with each case representing one-half hour of programming. The
data included the widest range of channel types, all time periods for which
ratings data were available, and an oversampling of weekend days (because there
is a substantial difference in program type availability on weekends versus
weekdays). Although the amount of programming available to the average viewer
has increased through an increase in the number of cable channels and the number
of broadcast networks, the same program types that were predominant in 1986
remain predominant in 1996.
These results clearly indicate the validity of the program
type variable as a discriminating variable related to viewing behavior.
Specifically, the set of 24 program types developed by Grant (1994) appears to
be a useful tool for the study of television diversity. Other typologies may be
just as useful, or even more useful, than this particular typology. To that
end, one of the most important contributions of this study to the literature may
be the explication of a means to validate a particular typology of television
programming.
An examination of the pairwise comparisons reported in Table
2 indicates that dramatic formats (including general drama, situation comedy,
science fiction, action-adventure, and westerns) demonstrate the highest levels
of discriminability among program types. On the other hand, close inspection of
the pairwise comparisons for a number of informational program types (including
skills, biography, health/nutrition/exercise, science, and nature) indicates
significant discriminability from dramatic program types, but not from other
informational program types. (One way to explore this question would be to
perform a network analysis of the program types, using the coefficients reported
in Table 2 as measures ofs similarity between program types.)
The primary purpose of this study was to examine the
validity of the program type variable, but it is also important to discuss the
manner in which the variable can be used in studies of diversity. Levin (1971)
identifies two types of diversity: "horizontal" and "vertical" diversity.
Horizontal diversity refers to the availability of program types across all
channels at a given time, and vertical diversity refers to the range of program
types offered by a single channel across its entire schedule.
The simplest means of measuring diversity is to simply count
the number of program types available. However, as xx suggests, this measure is
just a simple function of the number of channels available. A better measure
was suggested by Bohrnstedt and Knoke (1982), in whose formula the value of
diversity is determined by two factors: (1) the number of program type
categories (n), and (2) the distribution (Si) of each program types like the
following.
n
Div = 1 - Si 2
i=1
In the above formula, Si is the proportion of all program
types offered for the "i"th program type. If only one program type were
offered, the index has a value of 0. As the number of program types offered
increases, the index value increases, reflection the increase in diversity.
Therefore, the more program types are offered and the more balanced they are
offered, the higher the value of diversity index. Theoretically, the index has
a maximum value of 1 when an infinite number of programs are offered and each is
of a different type, though it is practically impossible.
xx DOH-YEON, CAN YOU ADD THE CITE, THE FORMULA, AND THE
EXPLANATION OF THE INDEX OF DIVERSITY HERE?
Future research should attempt a similar analysis for other
program typologies. We expect that similar results will be obtained that verify
the validity of the program type variable, and we further expect that,
ultimately, one particular set of program types will prove to be better than
others as a discriminating audience preferences. Although it is beyond this
project to do so, it may be useful to collapse similar program types into a
single value, where pairwise comparisons indicate no difference in audience
preferences.
These results cannot be applied directly to the television
systems in other countries. Differences in culture, media system, and media
traditions in each country are likely to affect the preferences of audience
members in different societies. We expect that a set of discriminating program
types will exist for each country or region, but that the particular set of
discriminating program types might differ in the same manner that the culture,
media systems, and media traditions differ.
In conclusionTherefore, the program type model, and more
clearly the 24 program type developed by Grant (1994), can obviously be a solid
tool for the study of television diversity. As television continues to evolve
and policy makers and researchers contemplate the issue of diversity, the
program type model should be considered a valuable variable to use in the
measurement of diversity. Furthermore, other areas of research ranging from
media effects to the social utility of television should be able to use the
program type variable with a greater degree of confidence.
Discussion
References
Bohrnstedt, G. W. and Knoke, D. (1982). Statistics for
social data analysis. IL: Peocock.
Cass, R. A. (1981). Revolution in the wasteland: Value
and diversity in television, Charlottevill: University Press
of
Virginia.
De Jong, A. S. & Bates, B. J. (1991). Channel diversity
in cable television. Journal of Broadcasting and Electronic
Media, 35:2,
159-166.
Frank, R. E. and Greenberg, M. G. (1980). The public's
use of television: who watches and why. Beverly Hills: Sage.
Glasser, T. L (1984). Competition and diversity among
radio formats: legal and structural issues. Journal of
Broadcasting
28:2, 127-142.
Grant, A. E. (1994). The Promise Fulfilled? An empirical
analysis of program diversity on television. The Journal of
Media
Economics, 7:1, 51-64.
Greenberg, E. & Barnett, H. J. (1971). TV program
diversity - New evidence and old theories. American Economic
Review,
61:2, 89-93.
Jackson, A. (1986). Has cable TV diversified away the
vast wasteland? Center for Telecommunications and
Information Studies,
Working Paper Series No. 270. New York: Columbia University.
Levin, H. (1971). Program duplication, diversity and
effective viewer choices: Some empirical findings, American
Economic
Review, 61, 81-88.
Levin, H. (1980). Fact and fancy in television
regulation: An economic study of policy alternatives. New
York: Russell
Sage Foundation.
Litman, B. R. (1979). The television networks,
competition and program diversity. Journal of Broadcasting,
23,
393-409.
McDonald, D. G. & Schechter, R. (1988). Audience role in
the evolution of fictional television content. Journal of
Broadcasting
and Electronic Media, 32(1), 61-71.
Rubin, A. (1984) Ritualized and instrumental television
viewing. Journal of Communication . 34, 67-77.
Stempel, G. (1981). Statistical designs for content
analysis. In Stempel, G. H. & Westley, B. H. (Eds.),
Research methods
in mass communication Englewood cliffs, NJ: Prentice-Hall.
Webster, J. & Lichty, L. (1990). Ratings analysis:
Theory and practice. Hillsdale, NJ: LEA.
Youn, S. (1994). Program type preference and program
choice in a multichannel situation. Journal of Broadcasting
and
Electronic Media 38:4, 465-476.
[1] His actual categories were twenty five, but one category
(others) are excluded for this study.
|
