The Green Mile

Film Information

The Green Mile is a 1999 American fantasy drama film written and directed by Frank Darabont and based on Stephen King’s 1996 novel of the same name. It stars Tom Hanks as a death row prison guard during the Great Depression who witnesses supernatural events following the arrival of an enigmatic convict (Michael Clarke Duncan) at his facility. David Morse, Bonnie Hunt, Sam Rockwell, and James Cromwell appear in supporting roles.

All information in this section came from Wikipedia.

Clip Information

John Coffey, a prisoner on death row, is executed via electric chair. The prison guards watching have come to know John and know that he did not commit the crimes he is being executed for.

Abbrev	Film	Clip Start	Clip Stop	Duration
GreenMile	The Green Mile (1999)	02:47:40.700	02:51:39.700	239

Characteristic	Value
Format	MPEG-4
File Size	50.5 MiB
Duration	239.031
Frame Rate	23.976
Video Width	1920
Video Height	1072
Video BitRate	1.6 MB/s
Audio Channels	2
Audio SamplingRate	48000
Audio BitRate	129.2 kB/s

Subtitles

The following wordcloud shows the words used in this clip, scaled by number of occurrences and colored by sentiment (orange = negative, green = positive, grey = neutral or unsure). Note that the words have been stemmed and lemmatized and stopwords have been removed.

The table below shows all subtitles in this clip with the start and stop time of each subtitle’s appearance in seconds.

Start	End	Subtitle
5.773	9.275	MARJORIE: Does it hurt yet? I hope it does.
10.110	12.362	I hope it hurts like hell.
14.281	18.368	John Coffey, you've been condemned to die in the chair by a jury of your peers.
18.619	22.163	Sentence imposed by a judge in good standing in this state.
22.456	25.708	Do you have anything to say before sentence is carried out?
27.461	29.962	I'm sorry for what I am.
41.975	43.726	Please, boss...
45.145	47.814	...don't put that thing over my face.
48.982	51.317	Don't put me in the dark.
52.152	54.487	I's afraid of the dark.
64.164	65.832	All right, John.
86.019	89.021	[SOFTLY] Heaven, I'm in heaven
89.231	92.024	Heaven, heaven
92.276	94.110	I'm in heaven
94.695	97.280	Heaven, heaven
97.531	99.240	I'm in heaven
120.137	121.762	John Coffey...
122.764	125.892	...electricity shall pass through your body until you are dead...
126.059	128.269	...in accordance with state law.
130.647	134.400	May God have mercy on your soul.
168.727	169.936	Paul.
173.106	175.024	You have to say it.
176.693	178.945	You have to give the order.
196.964	199.298	COFFEY'S VOICE: He killed them with they love.
199.466	204.345	That's how it is every day all over the world.
218.151	219.569	Roll on two.
221.154	223.739	[ELECTRICITY HUMMING]

Holistic Ratings

A total of 76 participants watched this film clip and then provided holistic ratings on how the entire clip made them feel. These holistic ratings were completed using five Positive Affect items (i.e., alert, determined, enthusiastic, excited, inspired) and five Negative Affect items (i.e., afraid, distressed, nervous, scared, upset), each rated on an ordinal scale from 0 to 4. The plot below shows the

Dynamic Ratings

A total of 75 participants watched this film clip and used the CARMA software to provide continuous (i.e., second-by-second) ratings of how it made them feel. These continuous ratings were made on a single emotional valence scale ranging from -4 (very negative) to 4 (very positive).

Time Series

We can plot the distribution of all valence ratings per second of the film clip to get a sense of how its emotional tone changes over time. The solid black line represents the mean of all ratings and the yellow, green, and purple ribbons represent the central 50%, 70%, and 90% of the ratings, respectively.

Inter-Rater Reliability

A Bayesian generalizability study was used to decompose the variance in ratings of this video clip into the following components: timepoint variance (in average ratings of each second, across raters), rater variance (in average ratings from each rater, across seconds), and residual variance (including second-by-rater interactions and measurement error). The lower and upper columns in the table below represent the boundaries of the 95% equal-tail credible interval. Note that we dropped the first 10 seconds of each clip (as rater “warmup” time).

Component	Term	Estimate	Lower	Upper	Percent
Rater	Variance	0.974	0.742	1.422	0.455
Timepoint	Variance	0.236	0.200	0.295	0.110
Residual	Variance	0.930	0.910	0.950	0.435

From these variance components, we can estimate inter-rater reliability of the ratings. There are many formulations of the two-way intraclass correlation (ICC), but the most relevant to our purposes here is the balanced average-measures consistency formulation or ICC(C,k).

Term	Estimate	Lower	Upper	Raters	Error
ICC(C,k)	0.951	0.942	0.96	75	Relative

Below, we can also visualize the posterior distributions of each of these parameters. Values with higher posterior density are more probable.