Reasonable metric for "popularity" in movies

I'm trying to do a statistical comparison of the "most popular" movies by year. That is, I'd like to look at the top 20-50 or so movies for each year from maybe 1950-2010. I'm flexible on the exact details, but that should give an idea of the spirit of what I'm doing.

The question:

What's a semi-rigorous way to determine which movies were most popular? Since "most popular" is nebulous, I'm looking for an approximation.
I'd like to use a metric which actually has data available. (this is the crux of my issue; the different metrics I consider don't seem to have readily available data, which makes me wonder if I'm using poor metrics)

My lines of thought:

The most obvious choice seems to me to be the box office gross. I've found reasonably detailed lists of the top grossers overall or the top individual by year (1, 2, 3, etc). However, for the top X movies in a year (where X is not 1), the best I've found is on IMDb, although they only seem to have box office data back to about 1971. Also, I'm skeptical as to the reliability of this data, as IMDb has a wiki component and they assert that box office grossings are not always accurate anyways. (That said, I'm willing to settle for less than perfect accuracy)

I've also considered finding several awards that have been around for a long time, and using the awardees as my data points. For example, I could use the winners (and to get more data points, perhaps the nominees as well) of the Academy Awards (around since 1929) or, as I'm concerned primarily with sci-fi, the Saturn Award (unfortunately only around since 1972). My concern here is that awards don't necessarily reflect popularity, and I don't know enough about movie awards to assert that they're at least a rough approximation of popularity. Also, they're not going to provide as many data points.

Does there exist some state of the art on this? Is what I'm looking for fundamentally impossible?

EDIT: I changed the title and made some tweaks to be more clear that I'm not asking for lists. My goal is to determine a reasonable way to assert the popularity of movies at different points in history. The existence of lists to prove that a metric is usable would certainly be a bonus, but I'm not incapable of using a search engine.

EDIT2: Just for anyone in the future who is making the same google searches I was, regarding an authoritative source for film grossings: after speaking to several academics in media studies, there's general agreement that the trade magazine Variety is the most rigorous place to get that data. Every year in one of their first issues, they list the top 250 grossing films from the previous year. I haven't found a good open source of this data, but a standard "good" library will have this. (I'm currently going through tons of microfilms)

cinema-history

Best Answer

I think the topic is a tough, tough problem. Not impossible but very hard. And I don't think any single metric will do the job. I suspect that you will need to look at a range of different types of metrics and search for commonality among them.

Box office is imperfect but important. But don't use it without making some corrections for inflation. And there might be more subtle signals in some sources that might also help such as longevity (i.e. how long the movie lasted in theaters or how well long term sales on digital or DVD held up).

Some accessible rating scales are also valuable. IMDB ratings are more robust than most as discussed here Are IMDB ratings objectively credible? . However, more robust doesn't mean very robust. But worth a look not least because they cover many old movies which tend to get missed in many magazine "vote for your favorite movie" polls.

Aggregate scales that allow comparisons from many sources may also be worth looking at. Rotten Tomatoes aggregates views from many critics (which might give you some idea of how much they disagree, as well as what their average rating is). Users can all vote there so you have a ready reckoner for how much critics agree with movie lovers.

I expect there is no truly robust method by itself. But I would be interested to see whether there are any significant common signals, if several sources are compared. This might be a significant statistical problem, though, potentially worth it.