Comments on: Some (brief) notes on cinemetrics II

By: Barry Salt

Barry Salt — Sun, 12 Jul 2009 14:04:09 +0000

Got the title of my note on accuracy in Cinemetrics wrong in the previous. It’s called “Requiring Split-second Timing”, though it is easy to find on the Cinemtrics site anyway.

By: Barry Salt

Barry Salt — Tue, 30 Jun 2009 23:34:41 +0000

I don’t think the sampling error would change the shape of a distribution that much, but it might stop it qualifying as Lognormal under your very restrictive criterion. John C.’s recording of “Verboten” was the least accurate of the three I checked in my piece on “The Question of Accuracy” on the Cinemetrics website. Some of the entries in the mass student class submissions are very inaccurate, I judge from the ASLs quoted for films I have previously counted.
Incidentally, I have just noticed that I had already said on page 219 of “Film Style and Technology” that “In the case of Lang, there is a clear transition from his European films.. etc.”, and of course this is obvious from the histograms in that chapter, as I already remarked.

By: Nick Redfern

Nick Redfern — Mon, 29 Jun 2009 20:41:08 +0000

The version of Verboten! referenced here is the one submitted by John C. I have found that a good recorder for cinemetrics can provide an reasonably accurate count (I added a comment about Chaplin to this effect on the Cinemetrics website under the discussion on making cinemetrics automatic). Unless the submission is widely inaccurate, there is no reason to assume that sampling error would be quite so long as to change the shape of the distribution.

Interestingly, if you divide this version of Verboten! into rough 30 min sections you find that for 1 second-1768.2s (218 shots) the median shot length is 2.9s; for 1768.2s-3837.8s (94 shots), median shot length = 4.2s; and 3837.8s-5172.0s (190 shots), median shot length = 2.9s.

By: Barry Salt

Barry Salt — Mon, 29 Jun 2009 12:44:54 +0000

Oh, and a minor point. Was the version of shot length record for “Verboten!” mine, which is farme-accurate, or that of John C?
There is a whole question about whether the approximations in the records obtained using the Cinemetrics programme affect the testing for Lognormality, and indeed other things.

By: Barry Salt

Barry Salt — Sun, 28 Jun 2009 18:48:54 +0000

Thanks. AS I read your results, there are 22 films out of 40 that meet your criterion for lognormality, and several others that come very close indeed. This seems a good reason to me to assume that the distribution of shots lengths is Lognormal for ASLs under 20 seconds, when investigating them. Also, there is a theory of how the Lognormal distribution arises, which gives a lead for investigating detailed causal factors, as I have mentioned in my books.
Scale of Shot distributions are very various indeed, particularly in the silent period, which may be the reasons that they resisted your approach. I like to “see” the actual shape of distributions, for what it might suggest to me.
In the case of Hitchcock, this aspect of his style also changed when he went to the US, as can be readily seen from the histograms in “Film Style & Technology”. And of course Lang reverted to his German Scale of Shot for “Der Tiger von Eschnapur” at the end of his life.
Auteur style analysis covers more than just Scale of Shot, and I discuss examples in more details in my books.