On Shot Lengths and Film Acts: A Revised View

In a recent issue of Projections, Cutting, Brunick, and DeLong (2011b) reported that the structure of film acts—the four relatively equal-length stretches of film determined by the progression of the narrative—influenced shot lengths and shot transitions. We compared 143 films released from 1935 to 2005. To do so we first needed to normalize each film to the same time scale. That is, because films in our sample ranged from just over 200 to just over 3,000 shots, and their average shot length varied from just over 25 seconds to just under 2.5 seconds, statistical transformations were necessary. Using a resampling method we stretched or compressed the waveform of the shot pattern for each film to fit into 1,000 bins, what we called 1,000 " adjusted shot lengths. " Of our major reported findings, one was that shot lengths within an act were scalloped in shape. That is, each act began and ended with slightly longer shots than those toward the middle. We reported the peak-to-trough magnitude of this effect to be about 1.1 seconds for the films that we investigated , which had a mean shot length of 7.7 seconds. Barry Salt was intrigued by our finding and, to his credit, tried to replicate it in his data and in ours, which are available on the cinemetrics website (www.cinemetrics.lv). Salt found several patterns that matched ours, but others that did not—particularly the longer shots that we found at one-quarter, one-half, and three-quarters of the way through films. In our correspondence, Salt suggested a possible artifact created by the method we used. When dividing the stream of shots of each film into temporal quarters as we did, we created regions near the ends of these quarters (particularly near and around bins 250, 500, and 750) where the alignment of shot lengths coincided across films, whereas they did not coincide away from these boundaries. This alignment alone, he argued, would cause averages at those locations to be inflated. He suggested a simple way to test for this: divide each film into thirds and see if peaks occurred at these new boundaries, or at bins at and around 333 and 667. We reanalyzed our data with our original interpolation technique and with two others that were similar but computationally different. Indeed, Salt is correct. With all three methods, when films were divided into quarters there were internal peaks at and around bins of …