God Knows I'm (Feeling) Good

To be honest I don’t know what prompted me to look into this one. I think I just came across a website that listed Bowie’s longest songs … that was it, no rhyme or reason, just that some of Bowie’s songs are quite long. Go figure.

But I came to wonder whether Bowie does something different with his long songs, considers different issues. I mean, long songs could just be recorded longer because of a preference for a particular arrangement – “Station to Station” has an interesting (and very long) intro, “Width of a Circle” has an extended guitar solo (especially the live version on Ziggy Stardust: The Motion Picture). Or it could just be that Bowie likes a particular chorus or lyric so much he wants to sing it a lot (like the la la las in “Time”). Or maybe, just maybe, long songs are longer than short songs because Bowie is exploring something. And if he’s exploring something, maybe he’s referring to different psychological constructs. And if he’s referring to different psychological constructs this should be reflected in the words he uses in the lyric.

So that’s it. No theory to test. And no particular expectation that there’ll even be anything interesting to report (although the very fact that I’ve posted this Chapter should lead you to suspect that something interesting did come out of the analysis).

Firstly, of course, we would expect that the longer the song, the more words there are.

And that’s what we find. The longer the song, the more words it uses [Footnote 1].

[Yawn]

Moving on, what we also find is that the longer the song, the fewer unique words there are [Footnote 2]. In other words, there is more repetition. This is something I wouldn’t have gone out of my way to predict but it’s perfectly understandable. In order to write a longer song, Bowie doesn’t simply write more and more verses, he repeats more choruses.

Sounds obvious really.

So that’s it for the basic counting. Everything from here on is based on percentages.

And I emphasise this because some people reading this might think, “of course there are more words about X, Y or Z … it’s a longer song … there are more words full stop”.

But that’s a ridiculous thing to say because the LIWC bases its measure of word categories on the percentage of word use. If a song has 100 words and 20 of them are about social processes then that’s 20%. If another song has twice as many words in total (i.e. 200 words) and twice as many words about social processes (i.e. 40 words) then the percentage of words relating to social processes is still the same, 20%. What would be interesting is if a song with twice as many words in total has three times as many words relating to social processes (60 out of 200 is 30%). That would mean there is something different about long songs versus shorter songs.

So what do we find?

Actually it’s quite interesting. Longer songs have more words relating to:

- First person plural (i.e. “we”, “us”, “our”)
- Positive feelings
- Anxiety
- Religion
- Swearing

And fewer words relating to:

- Money

[See Footnote 3 for details]

Okay, so the sizes of the correlations aren’t huge but they seem fairly robust. When I was first looking into this I entered only the song lengths of the albums I had to hand. I found some interesting correlations. So I dug out a few more albums. Most of the correlations were still there. I had albums with no track times so I searched online for the last few and still most of the same correlations were still there (the ones listed above). So there seems to be quite a bit of consistency across Bowie songs – as more and more are added, we keep getting the same result.

But as with some of the other analyses I’ve presented on this site, there is more we can do. Rather than just see what relates to what, we can determine which of these variables (word types) is actually predictive of the length of each song. To do this we need to use a regression analysis which tells us which variables are required to predict song duration. The idea is that some of the word-categories that correlate with song length might also correlate with each other and so it may be that not all of these word-types are pure predictors of the songs’ durations. [Footnote 4]

Regression analysis

Of the six variables that correlate (either positively or negatively) with song length, only three were independently predictive of song duration in a regression analysis. These were religion, 1st person plural and positive feelings – the higher the percentage of words from these word-categories, the longer the song [Footnote 5].

Now, between them, these three variables only account for about 8% of the variance in song length … this is a technical term, don’t worry if you don’t know what it means … simply notice that there is another 92% to find somewhere and that most of what predicts (or explains) the length of a song is not the word category. It’s an indicator of how good the model is at predicting song length … it’s “statistically significant” but it’s clearly not sufficient. [Footnote 6]

However, just to explore this analysis further, we can look at the five shortest songs and compare them with the five longest songs.

The five shortest songs last between 28 seconds and 1 minute 24 seconds and contain between 22 to 119 words. These songs are “Nathan Adler #1”, “Nathan Adler #2” and “Leon Takes Us Outside”, all from 1.Outside, as well as “Conversation Piece” from Space Oddity and “Future Legend” from Diamond Dogs.

The five longest songs last between 7 minutes 37 seconds and 10 minutes 11 seconds and contain between 204 to 641 words. These songs are “Let’s Dance” from Let’s Dance (obviously), “Bring Me the Disco King” from Reality, “The Width of a Circle” from The Man Who Sold The World, “Cygnet Committee” from Space Oddity and “Station to Station” from Station to Station (again, obviously).

The differences in use of religion, “we” and positive feeling songs between the five longest and the five shortest songs are shown in the figure below.

You’ve probably noticed that the percentages for these word types are very small, even for the five longest songs. However, for religion words, 0.7% is infinitely more than 0%; for “we” words”, 3.3% is about 20 times more than 0.16%; and, for positive feeling words, 1.4% is about 4 times more than 0.36%. So the magnitude of difference is pretty high across the three word categories.

So to summarise, in his longer songs, Bowie is more socially inclusive (as indicated by the use of “we” words) and expresses more positive and religious themes whereas, in his shorter songs, he is more individualistic, less positive and not at all religious.

Next question.

Why?

Well, this analysis can’t answer this. It can only identify the association. So in that sense the answer to the question “are there any differences between Bowie’s longer and shorter songs?” raises even more questions than answers. But above we ruled out the mere act that longer songs give more space to explore certain themes (because the prevalence of word types is based on percentages, i.e. number of words in a category as a percentage of the total number of words in the song, rather than a simple count).

But what about the other way round?

For themes that are particularly important, either because an individual is obsessed with something or where there is some difficulty dealing with a theme, the individual is likely to spend more time thinking or reflecting on that theme. So perhaps the themes of religion and social inclusion are important for Bowie and, as indicated by the use of positive feeling words, important in a good way.

There is certainly a very real sense that Bowie is wrestling with something deeply personal in at least 4 of the 5 longest songs. In what follows I should emphasise that these are not my interpretations but a summary of the offerings made by fans on a website about song meanings (http://www.songmeanings.net): “Width of a Circle” seems to reflect a discussion or conflict between the light and dark sides of Bowie or even a shift in the way he perceives (or believes in) God; “Cygnet Committee” takes a number of perspectives on the viability of the peace/love/hippy movement as a force for change; “Station to Station” is thought of by some in terms using spirituality in the transition from pain to love although, as others point out, it is replete with references to Kabala and Buddhism and to the “stations of the cross”; “Bring Me the Disco King” is discussed as a retrospective of Bowie’s life, thinking of the highs and lows and the lifestyle and excesses. Some consider Bowie to be the Disco King, other suggest that the Disco King is death itself (or himself). But even for “Let’s Dance”, Bowie has said in interview that the song refers to the dual nature of the symbol of red shoes, commercialisation and commodification (red leather shoes as a luxury item) and freedom and rebelliousness (“in blues songs it was always ‘put on your red shoes, baby’”).

Anyway, there is certainly not the same sense of personal involvement, discussion or conflict in the five shortest songs. Some are monologues rather than songs per se while “Future Legend” is post-apocalyptic and “Conversation Piece” is about being a loner and a thinker (rather than a talker). So while both of these are negative, they are unambiguously so – there seems to be little conflict or discussion.

What we find in this analysis is that discussing conflicting views takes a lot of extra lyric. And the conflicts that seem to be worth discussing (for Bowie) are those relating to social inclusion and religion/spirituality.

Footnotes

Footnote 1: rho = .37, p < .001

Footnote 2: rho = -.30, p <.001

Footnote 3:

* p < .05, *** p < .001

Footnote 4: Another way of putting this is to say that if A correlates with B and B correlates with C we might that A also correlates with C. Not because it has a true relationship with C, only because it correlates with B. It would be a spurious relationship that we can strip out in a regression analysis.

Footnote 5: I carried out a “stepwise” regression analysis and regressed song duration onto 1st person plural, positive feelings, anxiety, religion, swearing and money.

Religion was entered on the first step (ΔR-squared = .036, β = .19, t[1,264] = 3.27, p < .001), 1st person plural on the second step (ΔR-squared = .024, β = .16, t[1,263] = 2.77, p < .01) and positive feelings on the third step (ΔR-squared = .018, β = .13, t[1,262] = 2.26, p < .05). Total R-squared = .078.

Footnote 6: Some people get a bit hung up on the amount of variance that a variable (or group of variables) can predict in some outcome. A lot of people dismiss evidence when variables are found only to predict a small percentage of the variance. However, they are missing the point. Yes it’s true that most of what is needed to explain song length is not word-types. But that does not mean that word-type is unimportant. For example, smoking causes lung cancer. We know that. And yet, apparently, only 5% of the variance in the occurrence of lung cancer can be accounted for by smoking. To give another example, the vast majority of a cup of black coffee (either by weight or by volume) is water and only a small percentage is actually coffee granules. But if you dismiss that small percentage of coffee granules as unimportant and use only hot water, all you get is hot water, not a black coffee. And if all you got was hot water, Nero (other coffee shops are available) couldn’t charge you a fiver. That small percentage that is coffee granules is vitally important. (Both in terms of getting a coffee and in terms of charging over the odds).