DESCRIPTION: select reliable submissions


Cinemetrics lab is the latest addition to our site, and is still work in progress. When finished, the lab is envisaged to offer students of film history a range of analytical tools that will help them dissect, visualize and compare film-related data. We started with a large-scale comparative chart which looks a little like a star map. It is a scattergraph each dot on which represents a film available on our database. Select areas by dragging a rectangle to zoom in and see better different areas of film history. Once you have found your film on this map you will see how it relates to thousands of other films on the x-axis of time (past 111 years of film history) and on the y-axis of average shot lengths. What we intend to do is to add more tools to the lab in order to augment its statistics apparatus and enhance its means of data visualizations.

Year/ASL chart. Note the Log10 scale on the y-axis. It allows us to see all films in the very widely spread high ASL range.

WARNING: This page is not Internet Explorer friendly. IE might display this graph erroneously or very slowly. Please use a faster and less buggy browser (Firefox, Chrome, Opera, or Safari.)
This lab's films are shown in yellow.

List of films in this lab:

Golden Supper, The: (7) ASL 15.7
Girl and Her Trust, The: (7) ASL 6.6
House of Darkness, The: (7) ASL 11.6
Oath and the Man, The: (7) ASL 10.2
Rose O'Salem Town: (7) ASL 16.6
For His Son: (7) ASL 12.2
1776 or The Hessian Renegade: (7) ASL 18.5
Country Cupid, A: (7) ASL 12.9
His Trust Fulfilled: (7) ASL 25.8
Country Doctor, The (2nd try): (7) ASL 19
House with the Closed Shutters, The: (7) ASL 15
Swords and Hearts: (7) ASL 8.5
An Unseen Enemy (2nd Attempt): (7) ASL 6.9
What Shall We Do With Our Old?: (7) ASL 25.5
Burglar's Dilemma, The (2nd try): (7) ASL 10.6
Painted Lady, The: (7) ASL 9.6
His Daughter: (7) ASL 10.1
His Trust (2nd try): (7) ASL 14.5
Nursing a Viper: (7) ASL 17.1
Winning Back His Love: (7) ASL 18.3
Fighting Blood: (7) ASL 6.4
An Arcadian Maid: (7) ASL 15.9
Strange Meeting, A: (7) ASL 28.2
Billy's Stratagem: (7) ASL 7.4
Ramona: (7) ASL 14.3
Mender of Nets: (7) ASL 6.9
As It Is in Life: (7) ASL 16.2
Lesser Evil, The: (7) ASL 7.4
One Is Business, the Other Crime: (7) ASL 11.7
Sealed Room, The (2nd try): (7) ASL 20.4
Romance of a Jewess: (7) ASL 49.7
Last Drop of Water, The (2nd try): (7) ASL 10.8
New York Hat, The (2nd Attempt): (7) ASL 12.7
Renunciation, The: (7) ASL 29.6
Corner in Wheat, A (2nd try): (7) ASL 25.7
Light that Came, The: (7) ASL 23.4
Battle of Elderbush Gulch, The: (7) ASL 7.3
Two Brothers: (7) ASL 22.5
Usurer, The (2nd try): (7) ASL 17.4
Female of the Species, The : (7) ASL 11.3
Sunbeam, The (2nd try): (7) ASL 10.6
Redman's View, The: (7) ASL 34.5
Lesson, The: (7) ASL 13.1
Death's Marathon: (7) ASL 7.2
Flash of Light, A: (7) ASL 19.1
Musketeers of Pig Alley, The (2nd try): (7) ASL 11.2
In the Border States: (7) ASL 13.1
Miser's Heart, The (2nd try): (7) ASL 11.7
Enoch Arden (2nd try): (7) ASL 16.7
Friends (2nd try): (7) ASL 8.1
Indian Brothers, The: (7) ASL 15
Son's Return, A: (7) ASL 19.7
Lines of White on a Sullen Sea: (7) ASL 24.3
Adventures of Billy, The: (7) ASL 8.3
Battle, The: (7) ASL 8.6
Unchanging Sea, The (2nd try): (7) ASL 18.3
Voice of the Violin, The (2nd try): (7) ASL 43.3
Fugitive, The: (7) ASL 16.7
Mothering Heart, The (2nd Attempt): (7) ASL 8.2
Heart Beats of Long Ago: (7) ASL 11.7
The Gibson Godess: (7) ASL 23.6


The following notes will log, step by step, a series of experiments that Keith Brisson and I (Yuri Tsivian) embarked on starting January 25, 2011.

Our goal here is to find out if a statistically salient pattern can be detected as to how average shot legths fluctuate within the duration of a number of films regardless of the actual length of each.

In his topic recently posted on the Discussion Board Keith Brisson explains in detail how his new method works. Here is a brief non-technical description of it. We split each film into N equal-length partitions. For each of these partitions we count the number of shots that occurred in each. If a shot had a fraction in the partition we add that fraction. For each partition, dividing the partition's length by its shot count yields the ASL for that partition.  We can thus compare movies using data from this method by using the same N value for each film.  For instance, if N=100, then the first partition of each represents the first 1/100th of each film, and we can calculate and compare corresponding ASLs. This data can be averaged across films.  That is, for some N, for each partition we can average the ASL across all films.  So, if we choose N=100, and look at the 50th partition of each film, we can calculate the average shot length at the midpoint of each.  This average can then be averaged.  Doing this for each partition, we end up with something that resembles a curve (which is not smooth; to increase precision and alter smoothness we alter N) representing ASL at each point in the "average film".

This thus is the first, tentative, unchecked and untested method; so we decided to use this lab to test it. We are not yet publishing the result, but we agreed to make the process of trials and errors open to anyone's suggestions and criticisms (please use our Discussion Board for your comments).

1) January 25, 2011: The first step to be made was to apply the Brisson process to all the measurements then found on the Cinemetrics database. Predictably, the resulting diagram did not yield anything too well-structured: the database submissions vary from movies to TV programs to presidential speeches to sports events to the bursts of laughter of different audiences as they watch Chaplin's films. The graph below pushed us towards selecting a sample of more homogenous data.

2) February 9, 2011. As our next step Yuri Tsivian proposes to compare all D.W. Griffith's films submitted so far to the Cinemetrics database. The first task is to select the likeliest reliable measurements of 175 submissions by 14 contributors, one measurement per title. This resulted in a sample of 92 films Griffith made between 1908 and 1931 (to be called "Griffith-92 sample"). Roughly, this number is close to 1/4 of the entire corpus of films made by Griffith. The sample includes films submitted at different times by Charles O'Brien, Cid Vasconcelos, Torey Liepa, Kira Vorobiyova,  Charley Leary and Yuri Tsivian.

3) February 28, 2011. Keith Brisson applies his "average by partitions" procedure to the "Griffith-92 sample." This results in a graph reproduced below (1=start of the film, 250=end of film). This is what Keith writes: "The standard deviations of ASL for each partition are also provided. Note how high they are. This means there's a wide variation across films in any given partition. Nevertheless, there is a clear curve here; the value of average ASL for one partition is clearly correlated with the value of the next many average ASLs, which suggests a real curve does exist for Griffith."



4) February 29, 2011. Keith splits the “Griffith-92” sample into 3 sub-samples isolating a) films made between 1908 and 1913; b) films that are under 20 minutes long; c) films that are longer than 20 minutes.


As we can see, the longer (and later) films (the light-blue line) are cut faster than the shorts: their ASL oscillates between 5 and 25 seconds while the curves for the shorter films (green and purple) are placed between ASL of 20 and 50 seconds.
Another aspect worthy of note is that the green and purple curves are more pronounced than their light-blue counterpart which may point to a more uniform cutting pattern early (and shorter) films by Griffith lean towards. Their cutting rate appears to accelerate up to an ASL of 25 seconds past the middle (some 67% into the film), then to slow down to 45 seconds towards the end (note also a short tail wiggling up from 45 to 40 some brief 5% of time before the end).
A curious (and unexpected) thing about the graph for Griffith’s shorts is their fast beginning: the cutting is pretty fast (an ALS of around 20 seconds) within the span of the first 2% percent of the film’s length; then, around 6% into the film, the pace slows down. A possible reason for this may be Griffith’s use of expository titles (in early films, titles are as a rule shorter than shots).

5) To establish whether or not this is the reason for the faster cutting at the beginning the next thing to do is to try out what Barry Salt has dubbed “experimental film study.” To do this, Yuri Tsivian has selected 61 “advanced” measurements of Griffith’s “Biograph” shorts whose data can be presented either with or without intertitles. As our next step, Keith Brisson with generate two graphs, one with, the other without intertitles. We will then see. Meanwhile, keep your fingers crossed.

 6) Some time (almost a year) has elapsed since I last updated this research diary. Today is 2/3/2013. Some months ago as we planned (see point 5)) Keith Brisson did a new "partitions method" comparison 61 Biograph films measured in "the advanced mode" with title vs image distinction. The red curve represents the "partitions ASL" of these films with all intertitles factored out; the blue one with intertitled counted in. As we can see, most of the time intertitles shorten the ASL; indeed, in 1908-1913 these are mostly brief explanatory titles. As we predicted, the coriously long "beak" on the left (which signal faster cutting at the outset of the film) must be partly due to short expository titles which set up the scene; see comments on this below the graph.


As many screen-writing (photoplay) manuals instructed in 1912-1913, the exposition must be brief, which makes sense in the teeth of narrative economy one-reelers needed to observe. Knowing (from the said manuals) that the rule was 'one foot per word' in title editing, and that one foot runs for one second at 16 frames per second, here is a random collection of the expository titles (usually sandwiched between the opening credits which cinemetrics normally omits and the first shot):