Who Decides What to Watch on TV at Home?

Insights
from People-Meter Data in Mexico: Measuring Co-
Viewing and Preference Influences To Help
Broadcasters Promote Programming
José-Domingo Mora, Robert Krider and Jason Ho
Source: Journal of Advertising Research, Vol. 55, No. 1, 2015
Downloaded from WARC

How do household members influence one another’s television-viewing behaviors, and how can these
behaviors affect new programming? The current study offers a method to separate two different
sources of interpersonal influence among television viewers in the same household: what the authors
call “social co-viewing” and the intrinsic preferences of another viewer independent of co-viewing.
Applying the method to people-meter data from Mexico, the researchers found that, ultimately, wives
were more influential than husbands in building audiences. The authors believe their method can be
applied to any people-meter data, providing insight for programmers promoting new shows and for
advertisers choosing programs to sponsor in the upfront market.

José-Domingo Mora

University of Massachusetts, Dartmouth

Robert Krider and Jason Ho

Simon Fraser University

Management slant

Television programmers and advertisers can use people-meter data to identify the influences individuals
have on one another's television viewing behavior.
In Mexico, "co-viewing" of television is the largest source of mutual influence between husbands and wives.
Future viewing of programs by Mexican husbands may result from the positive influence of their wives,
once the effect of co-viewing is separated out, but the influence of the husband on the wife is negative after
the effect of co-viewing is controlled for.
 This people-meter data focused method may be applied universally. Broadcasters may optimize scheduling
of program advertisements by accounting for the positive effects of co-viewing.
Based on these findings the authors recommend that advertisers review their practices regarding their
focus on excessive entertainment in advertising.

INTRODUCTION

The failure rate of new television programs is consistently high. In the U.S. 2013–2014 season, only 31 percent
of the new non-reality television programs launched by the five largest broadcast programmers were renewed
for another year.1 This high failure rate represents a challenge for television programmers and advertisers, most
notably in the upfront market, where most advertising spending is committed before the programs even start.

To date, a significant part of broadcast-industry practice has involved evaluating television programs on the
basis of the independent preferences of individual viewers of different demographic profiles. The authors of the
current study believe, however, that such an emphasis on independent individual preferences overlooks an
important aspect of television viewers’ program choice: the influences that household members have on one
another’s television viewing.

Interpersonal influences overall are important to individual spouses’ television program choice (Yang, Narayan,
and Assael, 2006). This model of total interpersonal influences implied that a person could play the role of both
“influencer” and “influenced” at the same time and that the influence process is ongoing.

The authors of the current study acknowledged this was an important model for understanding television-
program adoption, but that it could be further expanded by

separating interpersonal influences into two components—

the influences of early viewers’ preferences on those of their family members or friends (the
“preference” influence), and
co-viewing occasions (the “social co-viewing” influence);
and applying this new model to people-meter data from Mexico to reveal novel insights on television-
viewing behaviors.2

Identifying and separating the components of interpersonal influences could be helpful to television
programmers and advertisers as they build audiences for new television programs. Although the model and
measures were applied specifically to the Mexican television market, the authors believe they can be applied to
people-meter data from anywhere in the world to elucidate market-specific relationships among viewers.

Decomposing Interpersonal Influences

In the course of their research, the authors identified and described two components of interpersonal influences
as the following:

the preference influence: 3 The authors’ model assumes the influencer and the influenced live in the
same household. Through numerous verbal and/or nonverbal interactions, the influencer’s preference
eventually changes the influenced’s internal preference for a specific television program. The influenced
may even end up enjoying the program when watching it alone.
 the co-viewing influence: This component is based on the social context of viewing. Television viewing
at home often is a group activity (e.g., Lull, 1980b, 1982; Lee and Lee, 1994; Schmitt, Woolf, and
Anderson, 2003; Chaney et al., 2013). A high correlation between the influencer’s and influenced’s viewing
of a specific television program could be due to a couple’s propensity to use a television program as a
means to enhance family relationships (Lull, 1980a, 1980b), and experience heightened positive affect due
to the company of others (Csikszentmihalyi and Kubey, 1981).

Individual viewing habits do not necessarily reflect an individual’s internal preferences. Indeed, in the absence of
the influencer, the influenced may not watch the program she or he might be expected to view.

To separate the preference influence from the potential co-viewing influence, the current study used an Index of
Shared Viewing, (Mora, Ho, and Krider, 2011) to characterize a couple’s propensity to co-view a specific
program. Including that measure, the authors proposed, would partial out the co-viewing influence from the high
correlation between a couple’s viewing of specific television programs and thus reveal the couple’s true
preference influence.

The current model was demonstrated using the Mexican data. In addition to distinguishing between preference
influence and co-viewing influence, the authors explored how the two types of interpersonal influences relied on
program genre and viewer characteristics in the Mexican market.

In fact, the model can be applied to people-meter data to understand the detailed structure of interpersonal
influences in any particular market. The authors applied it to the Mexican data to address six research questions
regarding the extent to which the general findings in interpersonal influences in the literature applied to the
Mexican television market.

RESEARCH QUESTIONS

To the authors’ knowledge, the existing literature on interpersonal influences in television viewing is quite
limited. Scholars have considered either overall influences, without separating out co-viewing and preference
influence components, or only co-viewing influence. Observations in the communications literature (e.g., Lull,
1982; Schmitt et al., 2003), an important motivation for the current research, suggest that the co-viewing
influence is strong. And, in one scenario, overall influence of one individual on another was positive (Yang et al.,
2006). Still other research has demonstrated that the presence of other family members has a consistently
positive effect on individual consumption of television programs of all individuals in the household (Yang, Zhao,
Erdem, and Zhao, 2010).

The current authors’ first objective, then, was to separate and compare interpersonal influences in television
viewing:

RQ1: What is the relative impact of co-viewing influence and preference influence on viewing preferences?

The way family members interact when making family decisions (Chaffee, McLeod, and Atkin, 1971; Lull, 1980b)
actually depends on a variety of household characteristics, such as socioeconomic status (McLeod, Atkin, and
Chaffee, 1972) and geography/local culture (Rose, Boush, and Shoham, 2002a; Rose et al., 2002b; Kim, Yang,
and Lee, 2009).

Family structure (i.e., the presence of children) may affect the size and composition of groups of viewers.
Parenthood alters couples’ needs and attitudes and likely also will alter television-viewing. Thus, the authors
addressed the following questions with regard to household heterogeneity:
RQ2: How do co-viewing and preference influences interact with socioeconomic status of households?

RQ3: Are there geographic variations in the co-viewing and preference influences between husbands and
wives?

RQ4: Do co-viewing and preference influences between husbands and wives change in the presence of
children?

As with other product categories, different program genres likely will elicit different levels of involvement across
viewers (e.g., Kuenzel and Musters, 2008); hence, the balance of mutual influences also likely will be affected
by such variables. Thus:

RQ5: How do co-viewing and preference influences among wives and husbands interact with program genre?

The literature demonstrates that similarity of personality profiles of wives and husbands is a predictor of quality
of couple relationships and long-term satisfaction with marriage (e.g., Gattis, Bern, Simpson, and Christensen,
2004; Gonzaga, Campos, and Bradbury, 2007; Markey and Markey, 2007). The evidence of this earlier
research, however, was not conclusive in terms of which psychological construct provided a better explanation.
Similarity in religious and political attitudes, for instance, was significantly higher among actual, stable couples
than among men and women matched at random, but the average similarity in several personality inventories did
not differ between these two groups (Luo and Klohnen, 2005).

This effect of similarity or dissimilarity of various psychological measures on quality and stability of relationships
likely also should be observed in the mutual influences on television viewing. Therefore:

RQ6: How does psychological distance between wives and husbands interact with co-viewing and preference
influences on individual consumption of television programs?

THE MODELS

Measuring Preference and Co-Viewing Influences

The authors of the current study operationalized interpersonal influences as the impact that one individual’s
enjoyment (“utility”) of a specific program has on another individual’s enjoyment of that program. In other words:
“Will I enjoy the program more if my partner enjoys it too?”

Model I

Model I captures the overall interpersonal influences between a husband and a wife—the two family roles that
are the focus of the current study.

In Model I, the total influence from the husband’s utility on the wife’s utility is captured by the parameter, ω*h→w
and the influence from the wife’s on the husband’s is ω*w→h. To distinguish empirically the influence parameters
ω*h→w and ω*w→h, the researchers needed to identify

influences other than the wife’s utility that shifted the husband’s utility but not the wife’s, and
influences other than the husband’s utility that shifted the wife’s utility but not the husband’s.

For brevity, these influences are referred to as focal variables. As shown later in this study, the authors
empirically found a number of drivers of viewer utility (variables parameters) necessary to estimate the model
(See “Other Factors Affecting Viewer Utility” and Table 1).
Model I has the same structure as the model of Yang et al. (2006; i.e., two mutually determined dependent
variables predicted as well by two sets of drivers). This model has not just one equation, as most regression
models do, but two. This is because husbands and wives influence one another’s choices of programs, but they
are individually influenced, at the same time, by drivers such as television program genres, geography, and
socioeconomic status. Those two dependent variables—predicting one another simultaneously while being
predicted by other drivers—form what in elementary algebra is called a system of two equations. That is, x and z
are functions of z and x, respectively, and from several other unknowns. In econometrics, a regression model
with these characteristics is called a system of simultaneous equations (See Figure 1).

Together with the drivers common to both husbands and wives (e.g., television program genres, geography, and
socioeconomic status), these husband/wife-specific drivers are the exogenous instrumental variables in the
terminology of econometrics literature (e.g., Greene, 2011).

Model II

Model II, the updated model introduced in the study, decomposed the overall interpersonal influences into the
preference influence and co-viewing influence (See Figure 2). The main difference between Models I and II is
the addition of the couple’s propensity to co-view the television program.

Although knowledge of the effect of co-viewing on individual-viewing utility—as captured by parameters ωh and
ωw—was valuable on its own, inclusion of co-viewing propensity in Model II allowed the researchers to partial
out co-viewing from total interpersonal influences.

This means that the two interdependence parameters ωh→w and ωw→h (without the star superscript) measured
personal influence, the husband’s (wife’s) inherent preference influence on the wife’s (husband’s) viewing utility,
net of co-viewing influence.

Mathematical Formulas

The mathematical formulas corresponding to the above-discussed models are specified as follows:

Since Models I and II have the same sets of drivers, except for the inclusion of co-viewing propensity, only the
two equations corresponding to Model II for ease of exposition are presented in this study. Model II is
characterized by a system of simultaneous equations for the husband’s and wife’s utilities derived from viewing
program p in household i , y*h,i,p and y*w,i,p:

where ωh→w, ωw→h, and ηw measure the preference influence and co-viewing influence.

Co-Viewing: The Index of Shared Viewing

Capturing the couple’s propensity to co-view program p, ISVwh,i,p is the Index of Shared Viewing (ISV), an
intuitive measure introduced specifically to study the antecedents and consequences of co-viewing in people-
meter data (Mora et al., 2011).

People-meter data provide the granular viewership tw and th for the wife and husband in a household within a
15-minute interval of a specific program. The ISV for the couple for the 15-minute program segment is thus
defined as:

Intuitively, the denominator captures the opportunity for co-viewing, and the index captures the proportion of
actual co-viewing to that opportunity. Note that if both people have zero viewership within the time segment,
ISVwh is zero, and the measure ranges between zero and one. As the focus of this study was at the program
level p, the appropriate measure was the average over all 15-minute segments and episodes of program p for a
couple. Thus, each household had a unique husband–wife co-viewing propensity measure for each program.

Other Factors Affecting Viewing Utility

Xw,i,p and Xh,i,p are vectors of drivers of the wife’s and husband’s utility, respectively. As discussed earlier, some
of the drivers in the current study were common in both vectors, but each vector included husband/wife-specific
drivers. This was necessary to empirically separate the preference influence parameters, ωh→w and ωw→h.

The researchers classified the drivers into three types (See Table 1):
 Psychographics and Demographics of Couple and Household. The psychographic categories for
individuals in the household, available in the Mexican dataset, allowed the researchers to construct a
measure of psychographic distance between a husband and wife, providing a unique opportunity to study
the effects of psychographic similarity of the couple on interpersonal influences. Family members’ roles and
ages and household socioeconomic status and geography where the homes were located were available in
the dataset.
Indicator Type of Content Variables. This category comprised a set of indicator variables representing
13 types of content including 11 national free television program genres and two other options. In addition
to the usual genre classification in television studies, the researchers included a genre unique to Mexican
or Latin American television markets—telenovelas. Telenovelas are melodramatic series (soap operas),
transmitted Monday through Friday, with durations ranging between 80 and 120 episodes of 60 minutes.
According to BBC News, telenovelas are the most popular form of audiovisual entertainment in the world
(Lizarzaburu, 2006).
Propensity and Earlier Preference. This category specified the husband’s or wife’s overall television
viewing propensity (operationalized as the overall amount of viewing prior to the estimation period) and the
husband’s or wife’s earlier program preference (operationalized as the amount of viewing of specific
programs prior to the estimation period). The authors included these drivers, because past viewing
behavior is an important driver of current viewing behavior (e.g., Rust and Alpert, 1984; Shachar and
Emerson, 2000) and, therefore, network investment pre-launch to build an audience early can pay off in the
long term through this lock-in effect. Note that these particular drivers differed for husbands and wives,
serving as the husband/wife-specific drivers essential to the empirical separation of the interdependence
parameters, as discussed earlier.

Estimation Technicalities

There were two technical issues when the authors estimated equations (1a) and (1b):
 The authors did not observe directly the husband and wife utilities, y*h,i,p and y*w,i,p, and, thus, needed to
relate them to observed behaviors—namely, the individual’s viewership of program p. Viewership data
contained many zeros as many programs were not watched, which the researchers managed with a Tobit
model to avoid biased parameter estimates (biased estimates are systematically above or below the true
values). It was important to avoid this issue as wrong parameter estimates would lead to wrong
conclusions, wrong predictions, and potentially a loss of money (e.g., Yang et al., 2006; see Appendix 1 for
a description of the estimation detail of the Tobit model).
The second technical issue related to the first one. As shown in equation (2), the ISV also was defined by
individual’s viewership of program p, creating a potential endogeneity issue well documented in the
econometrics literature (e.g., Greene, 2011). The authors controlled for this by using a set of predictors of
ISV as exogenous instrumental variables (See Appendix 2 for details).

DATA AND SAMPLES

People-Meter Data

The dataset for the current study was produced by IBOPE-AGB Mexico (IAM), which has used people-meter
technology to provide measurement ratings in Mexico since 1996 (and is a previous employer of the primary
author of this study). The sample design was probabilistic-stratified with over-representation of upper
socioeconomic strata. The total number of households in the panel exceeded 2,500, which ensured polling of
the daily minimum required by the inter-industry committee representing IAM’s clients. Four major geographical
areas were represented: Mexico City, Guadalajara, a composite of 25 small cities, and Monterrey (See Table 2).

A small screen on the people-meter set-top box asked, “Who’s watching?” every other minute when the
television set was turned on, and each viewer responded using a remote. The people meters recorded each
viewer’s time watched and channel and program codes.

As the researchers confined their interest to the interpersonal influences between a wife and a husband, the
current study focused on nuclear families and childless couples, omitting alternate family structures such as
single parents, female heads of households, or same-sex couples. IAM’s dataset reported gender and two
possible roles for adult individuals: the head of the household, and the wife. The study sample comprised
households where an adult man was the head and an adult woman was the wife, with or without children, but
with no other adults age 19 years or older. Those households with missing values in such predictors as
socioeconomic characteristics were excluded from this sample.

The resulting sample size was 668 households, which were polled during primetime from 7 p.m. to 11 p.m.,
Monday to Friday, for 11 weeks, between February 13 and April 28, 2006, for a total of 55 days. A separate prior
5-week sample from January 2 to February 10, 2006 (25 days) was used to calculate the earlier preference
variables.

The dataset reported the number of minutes consumed by each individual in the panel during each 15-minute
interval each day. Of all intervals observed in the sample,

17 percent had only co-viewing by the couple;

10 percent were the husband alone; and
17 percent were the wife alone.

The researchers considered 71 programs from free national television, taking into account the average
viewership for husbands and wives and the average co-viewing for the couple. Three other distribution systems
(satellite and cable subscription television channels, local free television channels, and video games and video
players) were treated as three separate viewing options.

The correlations among focal variables and drivers were calculated in the actual study sample; the results
showed high, similar values for spousal viewership (preference influence) and co-viewing influence (See Table
3). This was in stark contrast with the findings in the regression model (Model II), where co-viewing influence
estimates became between six and seven times larger than preference influence estimates (See Table 4). Thus,
the relationships uncovered by the proposed model are non-trivial and likely will go undetected by simpler
methods.

As the unit of analysis was the primetime viewing option by individual, program, and 15-minute interval, the total
number of cases used to estimate Models I and II was 47,052, with an average of 70.44 options watched by
each of the 668 households (the difference with respect to the total of 71 options resulted from daily fluctuations
in household polling).
Psychographic Data

One feature of IAM’s dataset is a psychographic classification of the Mexican consumers. Each individual in the
dataset was placed in one of five groups using gender-specific questionnaires having 44 items in common.
Agreement was evaluated by IAM on a 5-grade Likert scale for typical statements such as “You like to try new
products before other people do” or “Less emphasis should be given to family union and more to the individual.”
The measure of psychographic distance for a couple is the Euclidean distance over the mean values of these 44
items.

RESULTS

Interpersonal Influences in Mexico

Estimation results for Models I and II, respectively, considered the effects of interpersonal influences with and
without controlling for the co-viewing influence (See Table 4).

In Model I, the ISV was not considered. Thus, the coefficients of yspouse, referring to husbands and wives,
corresponded to the overall interpersonal influences, ω*w→h and ω*h→w (See Table 4). In Model II, as the ISV
was introduced to partial out co-viewing influence, the coefficients of yspouse now captured preference influence,
ωw→h and ωh→w. The coefficients of ISV are ηh and ηw, the influences of co-viewing propensity on the husband
and wife respectively (See Table 4).

Although Model I was consistent with previous research (Yang et al., 2006) in that husbands and wives
positively influenced one another’s viewing behavior, the incorporation of co-viewing propensity, ISV in Model II,
reduced the estimates of the interdependence parameters (ω*w→h versus ww→h: 0.19 to 0.07; ω*h→w vs. ωh→w:
0.14 to –0.10). The importance of this observation can be understood in the context of the first research
question, RQ1.

Together with ISV having large positive influences on wife’s viewership (0.61) and on husband’s viewership
(0.49), it was clear that the external, situational effect of co-viewing contributed substantially to the positive
overall interpersonal influences between husbands and wives, addressing the first research question:

RQ1: What is the relative impact of co-viewing influence and preference influence on viewing preferences?

Model II, in fact, revealed an unprecedented pattern: After controlling for co-viewing, the preference influence of
wives on husbands (ωw→h) remained positive; however, the preference influence of husbands on wives (ωh→w)
was negative. In other words,

the more (less) a Mexican husband preferred a program, the less (more) his wife would watch it when he
was not present; however,
when he was present, the more (less) his wife would watch (due to the larger effect of co-viewing on
wives). In contrast,
the more (less) a wife liked a program, the more (less) her husband would watch it, whether or not she was
present (See Figure 3).

The preference-influence coefficients were significant not only in a strict statistical sense but practically as well.
Because they related two standardized variables, they could be interpreted easily. For example, the husband
preference influence on wife of –0.10 meant that an increase (decrease) of viewing minutes by one standard
deviation by a husband resulted in a preference-influence viewing decrease (increase) on the wife by one-tenth
of standard deviation. The magnitude in minutes was similar as the standard deviations were nearly identical.

Moderators of Interpersonal Influences In Mexico

To study the potential moderating effects of five factors—

socioeconomic status of the household;

geographical locations of the household;
family structure of the household;
types of television programs; and
psychographic distance of the couple.

—on preference influence and co-viewing influence, Model II was estimated on five sets of subsamples split by
each of these factors (See Table 5), with the following considerations.

The socioeconomic status of households had a strong moderating effect on the preference influence,
addressing the second research question:
RQ2: How do co-viewing and preference influences interact with socio-economic status of households?

Specifically, wives’ positive-preference influence on husbands was stronger in the upper and upper-middle socio
economic classes than in the lower-middle and lower socioeconomic classes. By contrast, husbands seemed to
have no preference influence on wives in the high socioeconomic classes and a negative influence on wives in
the lower-middle and lower socioeconomic classes.

Geography also showed strong moderating effects on the mutual influences, addressing the third research
question:

RQ3: Are there geographic variations in the co-viewing and preference influences between husbands and
wives?

Mexico City is a large metropolitan area with 19.2 million inhabitants in 2005 (Conapo, 2010), many of
whom are first- and second-generation migrants from different regions of the country. Mexico City showed
the same preference influence pattern as in the overall sample: Wives avoided programs their husbands
liked while husbands liked what their wives liked (See Table 5b).
Guadalajara is the second largest city in Mexico but considerably smaller than the capital, with 4.1 million
inhabitants (ibid.). Guadalajara is known for its conservative values and exhibits the lowest consumption of
television in the sample. Preference influence was directionally the same but not significant, indicating
greater heterogeneity in preference-influence behavior in Guadalajara.

Family structure was another important moderating factor of preference influences (See Table 5c), addressing
the fourth research question:

RQ4: Do co-viewing and preference influences between husbands and wives change in the presence of
children?

Husbands’ preferences were influenced positively by wives in nuclear families but not in the case of childless
couples. Wives showed no significant preference influence from husbands in nuclear families, and negative
preference influence in the case of childless couples.

Programs watched also changed preference influence (See Table 5e), addressing the fifth research question:

RQ5: How do co-viewing and preference influences between wives and husbands interact with program genre?

Telenovelas showed the same pattern as the overall sample, while the pattern changed for cartoons and
newscasts: Preference influence tended to be less significant, indicating more heterogeneity in influence.

The authors interpreted the fact that there was a co-viewing effect only in cartoons, possibly because neither
wives nor husbands really cared about cartoons. If they had wanted to watch cartoons in primetime, they might
have done so because they wanted to spend time with their kids. This is consistent with earlier findings in the
United Kingdom, which had demonstrated that, although children watch children’s channels more than other
channels, a significant numbers of adults also watch kids’ channels, too (Sharp, Beal, and Collins, 2009).

Based on past research (e.g., Lull, 1990; Gantz and Wenner, 1991) and popular sports-audience composition,
the currentauthors asked: Is sports television enjoyed by men and disliked by women? And, if so, do such
programs contribute strongly to the asymmetry in personal influence in Mexico?

To answer these questions, the authors estimated the proposed models on a sample where all sports programs
episodes, only eight in total, were dropped. The resulting estimates were almost identical to those reported for
the total sample, meaning that sports in Mexican television are not responsible for wives not liking programs that
husbands like. One alternative explanation to the observed asymmetry is the use of television by Mexican
women to “resist male dominance” (Wong, 2004) in occasions other than co-viewing or, conversely, that the
power imbalance at home manifests itself only during co-viewing.

The researchers also examined the moderating effects of psychographic distance on mutual influences,
addressing the sixth research question (See Table 5d):

RQ6: How does psychographic distance between wives and husbands interact with co-viewing and preference
influences on individual consumption of television programs?

The split to define the “high” and “low” categories was set close to the median of the distribution of the
psychographic distance, PSY. Couples who were more similar (small values of PSY) followed the overall pattern
of influence, but in less similar couples (higher values of PSY), there was no significant preference influence of
husbands on wives, while there was a strong preference influence of wives on husbands.

IMPLICATIONS IN MEXICO

Target the Wives to Win the Husbands

In the study sample, wives’ preferences positively influence husbands’ whereas husbands’ have a negative
effect on wives’ (i.e., they are asymmetric). The authors, therefore, suggest that Mexican television-
programming managers should focus their promotions of new programs on wives, making every effort to have
stories and plots relevant to this target audience at least during the introduction and growth stages of their
offerings. The proposed statistical model, in fact, may help establish how critical these and/or other drivers of
new program viewership are in different geographies.

Identifying Drivers of Individual Utility

Because of the asymmetric personal influence between husbands and wives, identifying drivers with differential
effects on husbands and wives was found to be important for building audiences of new programs. This insight
came unexpectedly to the current researchers and initially seemed problematic, as most drivers presented the
same sign for wives and husbands whenever significant (See Table 4). The earlier preference drivers, however,
separately measured for husbands and wives, had strong positive effects on both the husbands’ and the wives’
utilities and, still, there were important differences between those effects (See Table 4: Model II).

Other Drivers of Individual Utility

Content type and geography also constitute drivers of individual utility although with converging effects on the
husbands and wives (See Table 4). This leads to a conclusion that may help shape programming decisions:
Generally, a program type that husbands preferred also was preferred by wives. One exception to this overall
tendency was that husbands preferred magazine-styled programming (e.g., “60 Minutes”) to contests (e.g.,
“Jeopardy”), whereas wives preferred contests to magazines. Thus, different from all other genres, by
considering the negative personal influence of husbands’ utility on wives—and the positive personal influence of
wives’ utility on husbands, as the program season progresses—all else equal—content managers should expect
contest programs to build audience more easily by default than magazines.

As for advertisers in the Mexican upfront market, the same data led to a comparable finding: New contest-format
programs generally may be more attractive exposure builders than magazine-themed programs.
Geographic Implications

The current study revealed differences between the different markets:

Wives in Mexico City seemed to be more responsive to new programs than wives in Guadalajara.
Husbands were more responsive to such shows in Guadalajara than those in Mexico City.
Because of the negative personal influence of husbands’ utility on wives and the positive personal
influence of wives’ utility on husbands, Mexico City is not only a larger market but may be a more cost-
efficient location for television programmers to build audience than Guadalajara.

Predictors of Co-Viewing Propensity

The authors believe that another way to increase individual viewing utility of a program may be by increasing co-
viewing of that program (See Figure 2). As part of the analysis, the authors also estimated a model relating
potential predictors to co-viewing (See Table 6 and Appendix 2).

Although these predictors are not obvious marketing-mix variables, understanding their effect may be useful
(See Table 6). For instance, viewers in Mexico City tend to co-view more than viewers in Monterrey and
Guadalajara, whereas viewers everywhere tend to co-view local comedies and newscasts more than any other
genre.

CONCLUSIONS AND FUTURE RESEARCH

Breaking down interpersonal influences in television viewing into preference influence and co-viewing influence
can provide useful insights to television programmers and advertisers. The current study developed a method to
disentangle two influence types. Application of the method to Mexican people-meter data revealed what the
authors believe to be previously undetectable and unexpected patterns. Furthermore, they believe that the
findings demonstrate the potential for comparable research to uncover useful insights outside Mexico.

The current study revealed a surprising and robust asymmetric pattern of preference influence among couples in
Mexico. Specifically—after partialing out the situational factor of co-viewing—although husbands would still
prefer the same programs preferred by their wives, wives actually would avoid television programs their
husbands preferred (or watch programs their husbands avoided). Such asymmetry in preference influence has
important implications to Mexican programmers and advertisers, as discussed in the previous section.

The method also uncovered a robust positive influence of co-viewing propensity on the individual viewing utility
of a program in Mexico (See Table 6). The mediating role of co-viewing in viewership as demonstrated by Model
II suggests that even if two programs have the same ratings, the ratings of one may be more strongly mediated
by co-viewing than the ratings of the other. That is, the individuals are there more for the social context of
viewing in one case than in the other. This particularly is interesting in view of earlier investigations that
demonstrated differential memory effects depending on the social context (Puntoni and Tavassoli, 2007).

Mexican comedies are an example of a genre in which the social context is very important: Comedies and
newscasts have the highest co-viewing (See Table 6). After controlling for co-viewing, however, comedies have
much lower impact on individual viewing than newscasts. The authors believe that advertisers who buy time on
comedy-focused programming in Mexico would find a welcome audience for spots that feature themes of social
desirability.
In addition to managerial implications, the current study also demonstrated the potential of several future
research directions in the advertising and media industries:

The methods in the current study can be applied to any people-meter data. Future research can focus on
another geographic market or compare several markets of diverse cultures. Cross-cultural analysis can
examine the extent to which the Mexican results can generalize to other markets. For example, it would be
interesting to see whether the asymmetrical personal influence among couples in Mexico would remain
among Latin American couples outside Mexico or among couples in a different culture.
A single market study could investigate other differential drivers for husband-and-wife viewing and
investigate their impact on long-run viewership as input to an optimal resource allocation model.
 The models should be separately estimated by type of family (e.g., in multi-generation homes, in single-
parent homes) as family role structure may affect how television viewing decisions are made.
The method also should be extended to interpersonal influences among other household members. For
example, would teenage children avoid a program if their parents liked it?
The current authors studied a limited set of predictors of individual utility and co-viewing; further research
on other predictors would be useful. For example, television co-viewing has been posited as a tool to
facilitate social functions within a family (Lull, 1980a, 1980b, 1982). One type of the social uses of
television is “affiliation/avoidance”: Watching television is a way to “reduce conflict,” to “elicit verbal
contact,” as a “family relaxant,” and as a way of promoting “family solidarity.” Factors that would promote or
suppress these goals are potential predictors of co-viewing.
Further research should focus on the social processes of preference influences. The asymmetrical pattern
in preference influences is particularly interesting.
Although the current authors used standard econometric techniques to control for endogeneity, the effects
of personal influence and co-viewing influence were studied using field data (Greene, 2011). Experiments
are needed to eliminate completely all possible alternative explanations associated with the use of field
data.
Finally, further research is needed to study the generalizability of the current findings to different product
categories. Would other product categories, such as restaurants and movies, also exhibit personal
influence and co-viewing influence?

José-Domingo Mora is an assistant professor of marketing at the Charlton College of Business at the University
of Massachusetts, Dartmouth. He specializes in media audiences, interpersonal influence, and group
consumption. A native of Caracas, Mora also has worked as an audience researcher and content manager for
television, media planning, and audience-measurement companies in Latin America—RCTV, OMD Venezuela,
and IBOPE-AGB México. His work has been published in the Journal of Broadcasting and Electronic Media.
Email: [email protected]

Robert Krider is professor of marketing at the Beedie School of Business at Simon Fraser University in
Vancouver, BC. His research specialties are in the entertainment industries, retailing, and tourism. His work has
been published in such journals as Marketing Science, Journal of Marketing Research, and the Journal of
Retailing. Email [email protected]

Jason Ho is an associate professor at the Beedie School of Business at Simon Fraser University. His research
interests are in empirical models of movie and entertainment product marketing, viral marketing, and managerial
applications of quantitative methods. His work has appeared in the International Journal of Research in
Marketing and Journal of Business Research. Email: [email protected]

Acknowledgments

The authors thank IBOPE-AGB México for providing the data for this research; SSHRC Canada for their
financial support; and R. M. Altman for valuable methodological advice.

References

Chaffee, S. H., J. M. McLeod, and C. K. Atkin. “Parental Influences on Adolescent Media Use.” American
Behavioral Scientist 14, 3 (1971): 323–340.

Conapo. “Proyecciones de Población (2010).” (Spanish: “Population Projections”). Retrieved from

http://www.conapo.gob.mx
Csikszentmihalyi, M., and R. Kubey. “Television and the Rest of Life: A Systematic Comparison of Subjective
Experience.” Public Opinion Quarterly 45, 3 (1981): 317–328.

Foote, N. “Television: Watch with Mother and Everyone Else.” Financial Times, FT Creative Business,
November 30, 2004, p. 4.

Gattis, K. S., S. Berns, L. E. Simpson, and A. Christensen. “Birds of a Feather or Strange Birds? Ties among
Personality Dimensions, Similarity, and Marital Quality.” Journal of Family Psychology 18, 4 (2004): 564–574.

Gonzaga, G. C., B. Campos, and T. Bradbury. “Similarity, Convergence, and Relationship Satisfaction in Dating
and Married Couples.” Journal of Personality and Social Psychology 93, 1 (2007): 34–48.

Greene, W. H. Limdep 9.0: User’s Manual , vol. 2. Plainview, NY: Econometric Software Inc., 2007.

Greene, W. H. Econometric Analysis, 7th ed. Upper Saddle River, NJ: Prentice Hall, 2011.

Kim, C., Z. Yang, and H. Lee. “Cultural Differences in Consumer Socialization: A Comparison of Chinese–
Canadian and Caucasian–Canadian Children.” Journal of Business Research 62, 10 (2009): 955–962.

Kuenzel, J., and P. Musters. “Social Interaction and Low Involvement Products.” Journal of Business Research
60, 8 (2007): 876–883.

Lee, B., and R. S. Lee. “How and Why People Watch TV: Implications for the Future of Interactive Television.”
Journal of Advertising Research 35, 6 (1995): 9–18.

Lizarzaburu, J. (2006). “How Telenovelas Conquered the World.” BBC News, Americas. Retrieved from
http://news.bbc.co.uk/2/hi/americas/4842220.stm

Lull, J. T. “The Social Uses of Television.” Human Communication Research 6 (1980a): 197–209.

Lull, J. T. “Family Communication Patterns and the Social Uses of Television.” Communication Research 7, 3
(1980b): 319–333.

Lull, J. T. “How Families Select Television Programs: A Mass-Observational Study.” Journal of Broadcasting 26,
4 (1982): 801–811.

Luo, S., and E. C. Klohnen. “Assortative Mating and Marital Quality in Newlyweds: A Couple-Centered
Approach.” Journal of Personality and Social Psychology 88, 2 (2005): 304–326.

Markey, P. M., and C. N. Markey. “Romantic Ideals, Romantic Obtainment, and Relationship Experiences: The
Complementarity of Interpersonal Traits among Romantic Partners.” Journal of Social and Personal
Relationships 24, 4 (2007): 517–533.

McCullagh, P., and J. A. Nelder. Generalized Linear Models, 2nd ed. Monographs on Statistics and Applied
Probability 37. Chapman and Hall/CRC, 1989.

McLeod, J. M., C. K. Atkin, and S. H. Chaffee. “Adolescents, Parents and Television Use: Self-Report and
Other-Report Measures from the Wisconsin Sample.” In Television and Social Behavior, G. A. Comstock and E.
A. Rubinstein, eds. Washington, DC: Government Printing Office, 1972.

Mora, J. D., J. Ho, and R. E. Krider. “Television Co-viewing in Mexico: A Study on People-Meter Data.” Journal
of Broadcasting and Electronic Media 55, 4 (2011): 448–467.

National Institute of Statistics and Geography Research. (2007). Retrieved from http://www.inegi.org.mx

Puntoni, S., and N. T. Tavassoli. “Social Context and Advertising Memory.” Journal of Marketing Research 44,
2 (2007): 284–296.

Rose, G. M., D. Boush, and A. Shoham. “Family Communication and Children’s Purchasing Influence: A Cross-
National Examination.” Journal of Business Research 55, 11 (2002a): 867–873.

Rose, G. M., V. Dalakas, and F. Kropp. “A Five-Nation Study of Developmental Timetables, Reciprocal
Communication and Consumer Socialization.” Journal of Business Research 55, 11 (2002b): 943–949.

Rust, R., and M. Alpert. “An Audience Flow Model of Television Viewing Choice.” Marketing Science 3, 1
(1984): 113–127.

Schmitt, K. L., K. D. Woolf, and D. R. Anderson. “Viewing the Viewers: Viewing Behaviors by Children and
Adults during Television Programs and Commercials.” Journal of Communication 53, 2 (2003): 265–281.

Shachar, R., and J. W. Emerson. “Cast Demographics, Unobserved Segments, and Heterogeneous Switching
Costs in a Television Viewing Choice Model.” Journal of Marketing Research 37, 2 (2000): 173–186.

Sharp, B. How Brands Grow: What Marketers Don’t Know. South Melbourne: Oxford, 2010.

Sharp, B., V. Beal, and M. Collins. “Television: Back to the Future.” Journal of Advertising Research 49, 2
(2009): 211–219.

Smith, R., and R. Blundell. “An Exogeneity Test for a Simultaneous Equation Tobit Model with an Application to
Labor Supply.” Econometrica 54, 3 (1986): 679–685.

Szalai, G. (2012). “Analyst blasts ‘Disaster of a year’ for new broadcast TV shows.” The Hollywood Reporter.
Retrieved from http://www.hollywoodreporter.com/news/analyst-blasts-disaster-a-year-196257

Thinkbox. (2009). “It ain’t what you view, it’s the way that you view it.” Retrieved from
http://www.thinkbox.tv/server/show/nav.854

Wedderburn, R. W. M. “Quasi-Likelihood Functions, Generalized Linear-Models, and Gauss-Newton Method.”

Biometrika 61, 3 (1974): 439–447.

Wong, J. E. H. “ ‘No le cambies a mi novela’: Dominación y negociación entre géneros en el acto de ver
televisión.” (Spanish: ‘Don’t tune out my telenovela’: Gender domination and negotiation in television viewing)
Comunicación y Sociedad 1 (2004): 163–182.

Yang, S., V. Narayan, and H. Assael. “Estimating the Interdependence of Television Program Viewership
between Spouses: A Bayesian Simultaneous Equation Model.” Marketing Science 25, 4 (2006): 336–349.

Yang, S., Y. Zhao, T. Erdem, and Y. Zhao. “Modeling the Intrahousehold Behavioral Interaction.” Journal of
Marketing Research 47, 3 (2010): 470–484.

APPENDIX 1

Tobit Model Specification

In estimating equation (1a) and (1b), the authors needed to relate the observed utilities y*h,i,p and y*w,i,p to
individuals’ program viewership. Program viewership, yr,i,p, where r represented either spouse, h or w in
household i , was defined as the average of the individual’s viewership over all 15-minute segments of program p
within the sample period. Viewership was observed directly from the people-meter data as the number of
minutes watched in a 15-minute segment of an episode of the program by each individual. As there were many
programs that an individual did not watch, program viewership at the program level would have many zeros.
Such left-censoring of the viewership variable required a Tobit model to relate viewership y to utility y*:

The simultaneous equation system (equation 1a and 1b) together with the foregoing Tobit relation was
estimated with the standard econometric method of two-stage estimation with instrumental variables.
Specifically, the current authors used the procedure developed by Smith and Blundell (1986) and extended by
Greene (2007) for two-stage maximum likelihood estimation of Tobit simultaneous equations.

APPENDIX 2

Endogeneity of Index of Shared Viewing

As program viewership was used in defining the dependent utility variable (Expression A1) and the index of
Shared Viewing (Expression 2), there was a potential endogeneity issue with the ISV variable. This issue was
handled by using a set of co-viewing predictors as the exogenous instrumental variables for the ISV variable.

Specifically, the authors first created a new variable: the predicted ISV based on the exogenous instrumental
variables. Then, as in the standard econometric technique of two-stage least-squares methods, they replaced
the original ISV variable in equation (1a) and (1b) with the predicted ISV, which would not be correlated with the
error terms in equation (1a) and (1b), εw,i,p and εh,i,p.

Note that ISVwh,i,p was bound between 0 and 1 but had potentially more dispersion than an indicator variable.
When creating the predicted ISV, a generalized linear model (GLM) was an appropriate way to relate ISVwh,i,p to
the set of instrumental variables, Zw,h,i,p:

To control for the potential over-dispersion in ISVwh,i,p, the researchers used quasi-likelihood (QL) estimation
(McCullagh and Nelder, 1989; Wedderburn, 1974). In the current sample, conventional maximum likelihood
estimation yielded the same coefficients, but with larger variances.

The instrumental variables, Zw,h,i,p included all exogenous variables in Xw,i,p and Xh,i,p, relevant to ISVwh,i,p.
Specifically, Zw,h,i,p contained program type/genres, type of family, age group, geography, and social economic
class. However, the state dependence variables in Zw,h,i,p differed from those used in the equations (1a) and
(1b). Instead of the two past-viewership variables, Zw,h,i,p contained two past behavioral variables of ISV:

the ISV of the couple for the 15-minute time interval preceding program p, regardless of program/channel,
averaged across the 11-week estimation sample; and
the ISV of the couple for program p, averaged across the 5-week initiation sample.
1Based on the 2013–2014 season scorecard of Metacritic.com

2 The people-meter data from Mexico was made available to one of the authors by IBOPE-AGB México.

3 The authors used the term preference as in the economics literature, where utilities of different choice
alternatives represent preference, and which is relatively stable (Varian, 1984, p. 111). In the current study, the
authors considered co-viewing one major source of instability in choice situations. Stable preference for a
television program does not necessarily mean consistent repeat viewing, as individuals have many more choice
options outside of television which disrupt regular viewing (Shachar and Emerson, 2000).

DOI: 10.2501/JAR-55-1-022-036

© Copyright Advertising Research Foundation 2015

Advertising Research Foundation
432 Park Avenue South, 6th Floor, New York, NY 10016
Tel: +1 (212) 751-5656, Fax: +1 (212) 319-5265

www.warc.com

All rights reserved including database rights. This electronic file is for the personal use of authorised users based at the subscribing
company's office location. It may not be reproduced, posted on intranets, extranets or the internet, e-mailed, archived or shared electronically
either within the purchaser's organisation or externally without express written permission from Warc.