Rating Prediction: How to forecast the success of new TV shows

Veröffentlicht: 8. Oktober 2013 in Communication, Forecasting

Widespread approaches to estimate success of new TV shows

Much is at stake when it comes to new TV shows. The development costs are usually pretty high and so are the risks of ratings far below expectations – and subsequent the risk of burning money. Everyone involved in the development of the show is working hard to make it a hit instead of a flop. They do so by permanently estimating which option might have which impact on the ratings. The approaches, however, are widespread:

Writers, Directors, Actors: All of them have gained a lot of experience from former projects what does work and what does not. They build assumptions and quite a few really perform well only relying on their gut feelings. The main issue they look at when making their assumptions about the chance of success is the actual product, the show.

Market researchers: Their approach is much different. They focus on the audience, the demand: How big is the target group? What are their needs and will the show meet them? How does the audience react to the program? How do test viewers rate the show?

Media planners: Next to TV producers also media planners have an interest in predicting the ratings of new shows. They need to place their customer’s media spendings right and make use of elaborate computer models therefore: They predict which show an individual will tune in given the program broadcasted before and at the same time of the new show based on the past viewing pattern. By combining the individual predictions they can make estimations about the show’s overall likelihood of success. The main thing they look at is the environment the new show is broadcasted in.

So which approach is most beneficial when it comes to predict the success of a new TV show?

I worked on this question for quite a while as it was the subject of my master thesis. And this is what I found out: I made different prediction models. One focused on variables that are product related (e.g. genre, duration, theme, …).  The model was able to explain a bit of variance in TV ratings out of sample but all in all the performance was rather poor. The next model put variables of the environment (e.g. which other shows are broadcasted at the same time, which program runs in advance,…) into focus. The model was able to explain some very weak ratings in advance but all in all did not perform that well. The third model was filled with the results of a survey I had conducted in advance. For every show I had collected data on how my survey participants rated their interest in the show based on the press release (demand approach). The model did ok but some peaks could not be explained by it. You might already have an idea on what comes next… Of course I combined all the three models and crunched it into one model. The out-of-sample prediction success of that model was far better than any of the models‘ focusing on variables deduced from one approach only. In 9 out of 10 times I was able to predict whether the ratings of the show would beat the average or not. The model explained 77% of the variance in the ratings. When you try to predict ratings it is essential to look at all relevant aspects: The product, the audience and the environment. If you read that here it sounds pretty obvious. However, in praxis the different approaches are still hardly combined and I think that is one of the reason why still many predictions fail.

How to put this knowledge into praxis

First: Break down the borders between departments

As the study shows, when it comes to prediction accuracy it is essential to study the object from a lot of different angles. Therefore also parties which are normally not involved into the development of a new show (e.g. media planning) should be asked for their advice and give their prediction on the show’s success. Since James Surowiecki’s fantastic book “The wisdom of crowds” we know that the average of individual predictions about the number of Jelly Beans inside a glass will be very close to the actual number. The same will be true for ratings: Just ask members of different departments for their estimation and calculate the average. If the people you pick look at the show from manifold perspectives, judge the show independently from each other and don’t have a self-interest in the program’s success (which is likely to influence the estimation) the average prediction should be pretty close to the actual result.

Second: Standardize data and build up a strong database

Even more beneficial than making use of the “wisdom of crowds” it will be to build up a big database with all shows that were brought on air on your channel. For every show you should collect as many data as possible from the spheres “product” (e.g. genre, track record of director), “audience” (e.g. rating of show in test screenings) and “competitive environment” (e.g. shows broadcasted in parallel, time of the year). Except for the target variable “rating” the database should only contain data that are available before a show is broadcasted the first time – otherwise they are useless for forecasting. Once the data collection is finished it is time to build the prediction model. There are a lot of statistical methods available. In the case of my master thesis a simple regression model worked quite well. And how, when you make a new show type in the variables, apply the formula and you should get a prediction that you really can base decisions on.

Will it be possible to make 100% accurate predictions?

No. Prediction models are always limited by a couple of factors like variable characteristics that are unknown beforehand (take the weather for example) or characteristics of the show that can hardly be put into numbers (like story quality which can only be “measured” by audience reactions). The model will give you good advice on how to invest wisely. But it will not make you sleep better the night before the ratings are announced.

PS: If you are interested in reading my thesis just send me a message

Advertisements

Kommentar verfassen

Trage deine Daten unten ein oder klicke ein Icon um dich einzuloggen:

WordPress.com-Logo

Du kommentierst mit Deinem WordPress.com-Konto. Abmelden / Ändern )

Twitter-Bild

Du kommentierst mit Deinem Twitter-Konto. Abmelden / Ändern )

Facebook-Foto

Du kommentierst mit Deinem Facebook-Konto. Abmelden / Ändern )

Google+ Foto

Du kommentierst mit Deinem Google+-Konto. Abmelden / Ändern )

Verbinde mit %s