Broadcast TV is a well-known example of a two-sided platform where cross network effects on the viewer and advertising sides interact. Like many platforms, it is advertiser-financed. While the literature shows that viewers dislike advertising, we explore a unique data set and distinguish between paid and non-paid (informative) ads. Cross-network effects’ estimates show that the latter carry a positive network effect on viewership. We also explore a significant change in public interest for more information in TV content in Russia in 2022 to estimate structural changes to cross-network effects. The results indicate that negative paid ads’ cross-network effects on viewership demand become stronger while positive non-paid (information) ads cross-network effects­ become weaker, even conditional on TV programing changes. Symmetrically, on the other side of the platform, advertisers value viewership less after the preference change.

broadcast TV, advertising, cross-network effects, two-sided platforms, demand estimation

A central part of the business model of the largest corporations in the world, digital platforms, exploits cross-network effects. While the importance of such effects cannot be understated today, it has been explored in different industries for more than a century, by newspapers, radio and broadcast or free-to-air television. Television is a mass medium characterized by accessibility, low selectivity and high impact on the general public. In 2023, the number of TV viewers worldwide is estimated at more than 5.2 bln.¹ and is expected to grow in coming years.² Despite the growing popularity of digital media, boosted by the COVID pandemic (UNCTAD, 2020), television is still an important source of news and entertainment for a majority of the population and an important way for advertisers to reach an audience. Lambrecht et al. (2024) show that TV ads boost the engagement on online platforms, being an important part of effective online platform marketing strategies.

Broadcast or free-to-air TV is a well-known example of a two-sided platform, where cross-network effects influence commercial decisions and industry organization (Evans and Schmalensee, 2016; Pontual Ribeiro and Golovanova, 2020). The cross-network effects, which characterize non-trivial interrelation between the size of the audience and the amount of advertising, allow broadcast TV to be economically viable using paid advertising, on the one side of the platform, and free content access, on the other. Nevertheless, the audience may dislike increasing advertisement duration in the programs. If advertising results in viewers switching to other programs, this reduces the attractiveness of the TV channel for advertisers and its revenue from airtime sales. Learning about the size of network effects and this trade-off is important to TV executives, advertisers and regulators, as often there are limits to the total amount of advertising on programs.

Assessment of the cross-network effects requires estimation of the viewers’ demand for TV content and advertisers’ demand for airtime. Most of the empirical papers on the topic follow the demand model with network effects adopted by Berry and Waldfogel (1999). In the literature, the positive effect of TV viewership on advertisers’ demand and the negative impact of ads on TV viewership are confirmed in Wilbur (2008) for the U.S., Ivaldi and Zhang (2022) and Zhang (2025) for France and Ferrer and Richter (2025) for Spain.

On the other hand, advertising may not put viewers off. The first positive estimate is from Rysman (2004), on “Yellow Pages” catalogues in the U.S., suggesting that the readers of these catalogues are interested in such content. In the print media (magazine) markets Song (2021) on TV magazines in Germany found positive effects of advertising on readership (circulation). Ivaldi and Muller‑Vibes (2018) explore the French magazine market and provide cross-magazine type evidence. The cross-network effects are positive only if the magazine­ is specialized. Specific type of advertising — previews — might be used to increase viewership of forthcoming TV programs (Bae and Kim, 2022). This may suggest that the attractiveness of advertising is due to viewers self‑selecting and being exposed to different advertising, rather than the advertising itself generating different effects.

It remains to be seen if within a media cross-network effects related to advertising of different types may change due to some exogenous shock. Our paper fills this gap. We take advantage of Russian TV viewership (proprietary) data that segregates aired ads in two types, paid and non-paid advertising. The latter is often information to viewers such as network-led public announcements and coming attractions. The data provides a first opportunity to investigate cross-network effects on a common pool of viewers.

The Russian case is interesting for other reasons. First, broadcast TV is an important media outlet. In 2021, 97% of the Russian population watched TV at least once a month, and 64% of the population watched TV daily, according to Mediascope.3

Second, using the 2021—2022 data on the Russian TV market, we can explore the effect of changes in audience interest in TV content, viewership and advertising. Broadcasters cater to audience preferences. However, it is hard to identify interest changes in the absence of significant external shocks. The Ukrainian conflict as a treatment for these preference changes is a unique “natural experiment” that makes it possible to estimate changes in the demand­ model structural parameters’ cross-network effects. With the conflict, the interest­ of viewers in ­information may have significantly shifted, making them less tolerant to ad breaks, but more interested in non-paid ads, even with programing changes.

The structure of the paper is as follows. In Section 1, we characterize the structure of the television market and the specifics of TV advertising in Russia. Section 2 provides information on the data used and the empirical approach. The key results of the empirical analysis are presented in Section 3. The last section concludes.

Broadcast TV is a well-known two-sided platform, with viewers on one side and advertisers on the other. Advertisers find the channel more attractive as viewership rises. Yet for viewers, more advertising may have positive or negative effects on the attractiveness of a channel. Some users may find advertising informative and are not put off by more or longer commercial breaks, while others may find ad breaks a nuisance and switch channels if there is more advertising.

The actual estimation of these cross-network effects has been a topic of interest in the economic literature. The literature is scarce, given the difficulties in accessing relevant data for the platform. Most of the studies concentrate on traditional media such as newspapers, radio, magazines or television.

A positive advertising effect on viewership or readership has been estimated for “Yellow Pages” catalogues (Rysman, 2004), specialized magazines (Song, 2021) and print media (Ivaldi and Muller-Vibes, 2018). Argentesi and Filistrucchi (2007) and Fan (2013) could not estimate precisely advertising effects on newspapers readers. In the case of TV broadcast channels viewership, Wilbur (2008), Zhang (2025) and Ivaldi and Zhang (2022) found negative effects of advertising time on program viewership.

Following Berry and Waldfogel (1999), the literature uses a (nested) logit demand functional form to model viewer preferences. The demand model differs from the usual demand function as there is no explicit price for broadcast TV viewers. The model is flexible to include cross-network effects in the utility function, with the inclusion of the amount of advertising, namely:

ln(s_jt) – ln(s_0t) = α^V q_jt + β^V x_jt + ε_jt, (1)

where (on the TV market example) s_jt is the market share of the channel j in time period t; s_0t is the market share of “outside good” (calculated as 1 – ∑_j s_jt); x_jt is the vector of observed characteristics of the TV program (type, time of day, day of the week, length) on TV channel j during time period t; q_jt — the amount of advertising on channel j during time period t (α^V is the indirect network effect of advertising); ε_jt — an error term. The market share for viewership is calculated as a fraction of the total population that could be watching a program at the time.

There are several approaches to estimating the advertisers’ demand. It should be immediate to maintain symmetry across sides and assume the same functional logit form for advertiser demand. For example, as in Argentesi and Filistrucchi (2007) the advertisers’ demand function perceived by a TV channel would follow:

ln(s^A_it) – ln(s^A_0t) = β^A x^A_it + α^A p^A_it + μ_i^A + ρ y^V_it + ξ_jt, (2)

where s^A_it is the share of paid advertising minutes of channel i at time t relative to the total potential market size; s^A_0t = 1 – ∑_i s^A_it is the market share of “outside good”; x^A_it collect channel i programing observed characteristics; p^A_it is the advertising price; μ_i^A represent unobserved, fixed, channel i characteristics; y^V_it is the channel viewership; and ξ_jt — an error term.

Demand model estimates may have endogeneity issues that require the use of instrumental variables for identification, as market data are the outcome of interaction between supply and demand. Common in the literature is the use of BLP-type instruments, i.e. product characteristics of competing firms (Berry et al., 1995; Argentesi and Filistrucchi, 2007; Wilbur, 2008; Ivaldi and Zhang, 2022; Zhang 2025) or the use of Hausman type instruments, i.e., prices or product characteristics in alternative geographical areas (Zhang, 2025). These instruments are valid under timing restrictions based on the institutional setting of the industry, i.e., program characteristics are set before prices and quantities, but influence prices and quantities themselves.

Table 1 provides a brief overview of studies that estimate cross-network effects on both sides of the platform, namely, the advertising effect on audience demand and the audience effect on advertising demand equations.

In the identification of a demand equation in a system, there may be endo­genous variables. This problem may be minimized with the use of controls and high-frequency data. On the one hand, TV advertising industry institutional information indicates that prices are set in advance and change only monthly. Prices can be seen as predetermined within the month. In addition, ratings in the advertising demand equation and advertising in the demand equation may be endogenous from the system of equations specification as the TV channel observes both sides’ demand. Again, the institutional setting of the industry across the world suggests that this endogeneity effect may be small. In France, Zhang (2025) argues that advertising is contracted in advance for a time slot based on expected cost and viewership. Over time, the advertising amount may change, but not in the time frame of our data, namely hour-channel-day. In ­addition, market information sources point that ratings are not known to the TV channel­ until after the end of the program, if not the following day.⁴ In both cases, advertising amount and viewership as explanatory variables may be considered pre-determined with respect to current unobserved shocks to viewership and ad quantity when using high-frequency, hourly data.

To estimate the cross-network effects by advertising type and the influence of the 2022 shock on the Russian TV advertising market, we follow the literature and estimate two demand models: one for the audience side (TV viewers demand for content) and another one for the advertisers’ side (demand for airtime). In both equations, the unit of observation is hour-day-channel interval.

The institutional and technological setting of the industry, combined with the high frequency data show that the TV station cannot react contemporaneously (within the hour) to unobserved shocks to viewership and advertising demand estimated.

In order to estimate the geopolitical shock of February 2022’s impact on the indirect network effects, we provide a differences model within a structural model that predicts viewership and advertising demand for a channel, where we allow a break in the parameters from the start of the event.

4.1. Audience demand

Since the 19 TV channels considered in this paper are free, there is no price variable in the audience demand equation. In fact, the advertising duration acts as a ‘cost’ for the audience, as they pay attention to watching the advertisement in anticipation of the content they are interested in. Each viewer chooses to watch a TV channel or the “outside good,” maximizing his or her utility. The “outside good” option means to watch a channel that is not included in our sample or not to watch TV at all.

The audience demand equation can be presented by the discrete choice model (Berry et al., 1995; Berry and Waldfogel, 1999). Therefore, assessing this demand, we follow the methodology commonly used by researchers to estimate the indirect network effects (Rysman, 2004; Argentesi and Filistrucchi, 2007; Wilbur, 2008; Zhang, 2017, 2024; Ferrer and Richter, 2025). The basic equation is:

ln(s_jt) – ln(s_0t) = α₁ Paid_share_jt + α₂ NonPaid_share_jt + ω after +

+ α₃ after × Paid_share_jt + α₄ after × NonPaid_share_jt +

+ β_i x_jt + ε_jt, (3)

where s_jt is the share of TV channel j at time t, measured as a ratio of the number of viewers of the TV program broadcasted by the TV channel j at time t to the total market size, which is taken equal to the total population of Russia (144 million); s_0t is the share of “outside good” at time t, s_0t = 1 – ∑_i s_jt; Paid_share_jt = Paid_jt /3600 is the share of paid advertising on the air on channel j at time t, measured as the ratio­ of the duration (in seconds) of paid advertising on channel j during the time t (Paid_jt) to the total number of seconds in an hour; NonPaid_share_jt = NonPaid_jt /3600 is the share of non-paid advertising (previews) on the air on channel j at time t, measured as the ratio of the duration (in seconds) of non-paid advertising on channel j during time t (NonPaid_jt) to the total number of seconds in an hour; x_jt presents programs’ characteristics such as genre and duration; ε_jt is an error term. The coefficients α₁ and α₂ reflect the indirect network effects of paid and non-paid advertising, respectively.

We distinguish these two types of advertising, assuming that they have different­ effects on audience demand. If the corresponding coefficient is less than zero, the effect is negative and otherwise. We expect that α₁ < 0 because paid advertising annoys the audience, and the α₂ > 0 because non-paid advertising is the part of the searched content, informing viewers either about future programing or public interest announcements.

The fixed effects of TV channels, day of week, program genre, hour interval and year–month are included in the model as control variables (x_jt). Fig. 4 shows that there is seasonality within day and between months in viewership.

Figure 4.

The dynamics of TV channels’ ratings by month (2021–2022) and day (2021). Source: Mediascope, conducted by the authors.

We also include two program characteristics: (i) whether the TV program duration is more than an hour. Longer broadcasts (such as movies) are characterized by higher average ratings: their average rating is 0,84–8,86 compared to 0,52–0,55 for programs with a duration less than one hour;¹⁰ (ii) a prime time (from 7 PM to 11 PM on weekdays, from 8 AM to 11 PM on weekends and holidays) indicator. The audience demand for a TV program’s content is measured by the rating of the particular TV program, in percentage of total potential viewership.¹¹ Ratings are collected at the level of the program that was on the air at the beginning of the hour interval. Appendix A provides a description of the variables and descriptive statistics.

To estimate the impact of the Ukrainian conflict, the variable “after” is included­ in the equation. This variable is equal to 1 for the dates starting from 24.02.2022. In addition, the variable is interacted with the advertising minutes so that we allow preferences themselves change with respect to advertising. The coefficients α₃ and α₄ reflect changes in the intensity of indirect network effect of paid and non-paid advertising, respectively. Their signs are also a point of interest in the study.

Table 3 presents the results of the estimation of the audience demand model. Columns (A), (C) and (E) use the total number of advertising seconds within the programing hour, in line with the previous literature. Columns (B), (D) and (F) break the total number of commercials into paid and non-paid ads. Columns (C) and (D) change the base specification (A) and (B) to include channel and prime time fixed effects’ interactions in addition to the main effects included in the base specification. Columns (E) and (F) are an additional robustness check with the exclusion of Rossiya K channel, which has zero paid advertising (it is fully sponsored). Column (E) should be compared with column (C) and column (F) with column (D).

In all specifications of the regression models, the coefficient of paid advertising duration is statistically significant and negative, which is reasonable and consistent with the previous research results. As for the non-paid advertising effect, it is statistically significant and positive. This aligns with the hypothesis that previews have an informative function and play a role of “desired” content, at least compared to paid commercials. The total advertising duration effect is negative, as the paid commercial effect is larger and there are more paid commercials than non-paid commercials in a programing hour in general.

Moreover, we obtain statistically significant coefficients for all variables related to the effect of the February 2022 shock. Our finding is that the audience became less tolerant to the paid advertising. The same conclusion can be drawn about the non-paid advertising, although the effect is still positive. For example, in column (B), an addition of a second of non-paid advertising in the hour would lead to an increase in ratings by 1.61 before the shock. After the shock, the effect becomes 1.01 (1.61–0.60), still positive but smaller.

As for the prime time and duration variables, the corresponding coefficients are statistically significant and positive. It means that the audience likes watching TV during prime time hours and gives preference to longer programs, depending on genres.

It should be noted that we use an extensive set of fixed effects to control for unobservable preference shifters, such as channel, month of the sample, genre, day of the week, and the time slot, given the descriptive analysis above. These fixed effects should predict expected ratings that influence advertising slots in each TV show, rendering unobserved shocks to viewership unrelated to advertising quantity in the viewership demand model.

Table 4 depicts the results of the advertisers’ demand estimation. Column (G) and (H) are simplified, reduced form, specification as the price is omitted. Column (H) uses log rating (tvr). Column (I) includes (log) price. The audience effect is statistically significant and positive. This finding is consistent with previous research results that point that wide audience attracts advertisers.

The estimation of the February 2022 shock shows the decrease in demand for advertisement placement, as seen in the descriptive statistics above. Such an effect appears to be not driven by audience changes or price changes, as the effect is negative conditional on those variables (columns H and I). A possible explanation for this result may be the withdrawal of many foreign brands from the Russian market, including advertised ones.

Cross-network effects became weaker after the break, meaning that an additional audience level was not as attractive to advertisers as before the break. The price elasticity of demand for advertising decreased as well. While an additional audience is less enticing for advertisers, reducing its demand, the advertising becomes less sensitive to price per rating increases. As we use time and channel fixed effects, the price elasticity coefficient is identified from within channel variation, particularly before and after the shock. While the fixed effects control for real price changes (inflation/exchange rate adjusted, common to all channels), the coefficients reflect the preference changes of advertisers.

To provide more credibility to the results, in Appendix B we estimated the models (3) and (4) for month-by-month to see if the estimated effects were affected by seasonality, i.e., if the change in February 2022 and later, was driven by possibly lower coefficients at the end of winter and spring that may re-occur annually. Figs B1–B3 in Appendix B, that summarize the paid and non-paid advertising effects from the audience demand equation (3) and the viewership effect on the advertising demand equation (4), suggest that there are seasonality effects within the year. Nevertheless, the main results in Tables 3 and 4 above hold. In February, March and other months of 2022, compared with the same months in 2021, in the viewership demand, (i) the paid advertising negative coefficient becomes more negative and (ii) the non-paid becomes closer to zero and, in the advertising demand, so does the audience effect. Note that the confidence intervals in the month-by-month regressions are larger than in Tables 3 and 4 as the sample size is much smaller.

We see several reasons for such a change in advertisers’ preferences. First, many foreign brands exited Russia and stopped advertising on Russian TV channels. Domestic producers and companies from countries that did not join sanctions got a chance to capture the vacant part of the market. Structural changes in the markets of goods and services resulted in changes in the list of brands advertised on television in favor of domestic brands. Thus, changes in advertisers­’ preferences are, at least partly, resulted from changes in the list of companies that buy airtime. It seems that local companies are less sensitive to programs’ viewership and price per rating point.

The second reason is structural changes on advertised products’ markets in Russia. The exit of many well-known brands from the country made the markets up for grabs for new local brands and brands from countries that did not join sanctions. TV advertising is an effective way to inform a large part of the population about non-familiar products. Given that the speed of entering the market does matter, the factors that influence the related costs (including CPP and TVR) are expected to affect less the decision on ad placement under intensive non-price competition for market share.

The paper is aimed at exploring a long-debated question in the literature, namely whether ad breaks are a bonus or a nuisance to audiences by examining commercial and non-paid advertising network effects on viewership demand. Institutional features of the Russian TV market, jointly with high frequency, ­detailed TV programing and audience data allow a cleaner identification of demand parameters. Also, events in the country generated the opportunity to estimate demand parameter changes when there is a cognizable short, important shift in the public’s interest. This shift was reflected in TV programing changes, so it remains to be seen if structural parameters were affected as well.

The results indicate that paid advertising duration in programs reduces viewer­ship, and non-paid advertising has a positive, albeit smaller, effect on the audience. The former was known from other TV market estimates in the literature. The latter is previously unknown. After the start of the Ukrainian conflict, Russian audiences became less tolerant to paid advertising and more willing to accept non-paid advertising. On the other side of the platform, we confirm a negatively sloped demand for advertising with positive viewership effects. After the change in audience interest, there is a reduction in the strength of the network effect (in symmetry with the viewership demand) combined with an attenuation of the ­demand elasticity. This empirically confirms that the strength of the network effects is not a fixed characteristic of a media market. It may change significantly under external shocks that affect viewers’ and advertisers’ preferences.

It would be interesting to know if the preference changes were observed in other countries, closer or further from the same changes observed in late February 2022. This is the direction for further studies.

The article was prepared within the framework of the Basic Research Program at HSE University. Eduardo Pontual Ribeiro acknowledges partial funding from CNPq and FAPERJ. The authors thank two anonymous reviewers and the managing editor for their constructive feedback on the manuscript.