Measuring the Unmeasurable: Ranking One’s Favorite Music, Part II

In late June, I wrote the first in a series of essays outlining the evolution of the methods I use to rank my favorite tracks (a term I prefer to “songs”), beginning with my first-ever mix cassette tape in August 1981. In the interim, however, moving to a new apartment, the ongoing search for a publisher for my book, and life in general delayed a follow-up essay.

Now, though, we resume the journey – starting with a quick synopsis.

By January 1993, I had constructed 88 cassette and videocassette mixes comprising 1,163+ unique tracks – 92 if one counts a two-cassette favorite tracks of the 1980s mix and two mixes made for romantic interests; two lost cassettes were partially reconstructed from memory. Mixes were mostly “catch as catch can”: combinations of tracks recorded from the radio, cable music channels or other people’s collections; tracks I had just acquired; and single-artist mixes (February 1992 two-cassette Genesis mix, July 1992 Roxy Music mix). A handful of “pure” mixes recorded for a long drive or as a study aid followed no particular sequencing rules, other than grouping tracks by the same artist and keeping their original album sequencing.

In late August 1989, I recorded 136 tracks – only nine appearing for the first time – on a six-cassette mix for my upcoming drive from suburban Philadelphia to suburban Boston. The other 127 tracks were my favorite among the 919 unique tracks recorded over the preceding eight years onto 56 mixes.

In January 1993, I decided to rank my then-favorite tracks by assigning a numeric value to each, though I only entered data into my spreadsheet from the 36 mixes starting with “Boston Drive Vols. I-VI;” I had left all of the pre-Boston-Drive mixes in my mother’s apartment in Penn Valley. The 512 total tracks – 136 from August 1989 plus 244 first recorded after, and 132 first recorded before, August 1989 – occupied 749 “slots;” fifty tracks appeared three-five times. These multiple appearances and new ideas about sequencing led to a series of “ranking hypotheses”:

Ranking hypothesis #1: I like every track I recorded onto a mix more than any track I have never recorded onto a mix.

Ranking hypothesis #2: The number of mixes onto which I record a track is positively associated with how much I like that track.

Ranking Hypothesis #3: The tracks I like the most on a mix or set of mixes are the first tracks I record on a cassette/CD – especially the anchor track – and the final track I record.

Ranking Hypothesis #3a: RH3 is not true before August 1989.

Ranking Hypothesis #3b: RH3 is sometimes true between September 1989 and February 1992.

Ranking Hypothesis #3c: RH3 is always true after February 1992.

Ranking Hypothesis #4: I like tracks included on designated mixes more than those only included on non-designated mixes.

Ranking Hypothesis #4a: How much more I like tracks included on designated mixes varies by designated mix.

With these hypotheses, I essentially weighted number of appearances by the “quality” of the mix – giving particular weight to the Boston Drive and “Top 40 of the 80s” mixes – to calculate a score for each track. Those scores are long gone – but a handful of Top 100 tracks, Top 50 albums and Top 75-100 artists rankings from the 1990s remain.

**********

In early March 1993, I decided on a lark to fly to Clearwater, Florida to see four Philadelphia Phillies spring training games. Despite a severe case of sunburn – I sat shirtless in the afternoon sun for four hours during the first game I saw – this was one of the best trips I have ever taken. A few months later, I met the woman my wife Nell calls my “first wife,” and she and I returned in March 1994, and then each March from 1996 to 2000.

Entertaining me that first trip was Stuff and Such Vol XXXIV, the first mix I constructed after my first attempt to rank my favorite tracks using mix-appearance data. Only five of its 22 total tracks appeared for the first time – three by Roxy Music and one each by Devo and Yello. This mix was an early example of how I was rethinking track sequencing, beyond not having consecutive tracks by the same artist, and having a “rocking” Side A and a “mellow” Side B. I now wanted tracks to flow into each other as much as possible musically, so the tracks formed a continuous suite. This is very difficult to describe in detail because it is how I hear the tracks, as opposed to matching keys or instrumentation or vocals or something equally tangible. I just really liked, for example, how “Hit” by The Sugarcubes flowed into “It’s A Fine Day” by Opus III which then flowed into “Private Life” by Oingo Boingo…and so on. The payoff – a cohesive musical whole greater than a simple collection of tracks – was worth the extra time it took to “hear” both the end of a track and the beginning of a possible subsequent track, recalling that I had to play every track straight through in order to record it.

Over the rest of the decade, I perfected my mix-construction style:

I began by compiling, over time, a list of tracks I wanted to put on the mix: combining tracks recently recorded for the first time I had not yet “burned out” on, older recorded tracks I wanted to hear again, and new tracks either from my album/CD collection or which I acquired by prowling new and record stores, sometimes buying an entire album for a single track. A portable cassette/CD player I bought around 1992 allowed me to record from CDs, albeit without Dolby noise reduction. In late summer 1997, however, I purchased an Aiwa combination radio/CD player/cassette player/cassette recorder though which I could play (and record) my new Panasonic turntable; it had terrific Dolby noise reduction, and I thought it was state of the art.

The length of time it took to record even a single cassette made me extremely selective; I occasionally included only part of a track to make everything fit. I recorded an average 25 tracks onto a 90-minute cassette, which could take as long as two hours to complete. A three-cassette mix, like the one I recorded in March 1997, could take up to six hours to complete.

The sequencing rules were as follows:

The first track of the mix – the “anchor” track – was the recently-acquired track I liked best. It needed to be “rocking” as it often marked the start of a vacation or long drive.
The second track was one previously recorded, though not recently, which flowed musically from the first track.
As much as possible, I alternated new-repeat-new-repeat, keeping the musical flow.
The last track on a cassette was also an intriguing new track.
Sides alternated “rocking” and “mellow.”
With rare exceptions, there were no consecutive tracks by the same artist.

Between March 1993 and March 2003, just before I bought a CD burner, I recorded 62 individual cassettes (Table 1).[1] Of the 1,548 tracks recorded, 669 (43.2%) were recorded for the first time – 11 out of 25 tracks, on average, or just under half – including both newly-acquired, newly-discovered from my album and CD collection and first recorded prior to August 1989.

Excluding 1993 and 1994, when I was still perfecting the process, first-time percentage increases to 47.5%, or 12 out of 25 tracks, on average. And from 1998 onward, shortly after acquiring the Aiwa, the new track percentage increases to 51.6% (12.8 out of 24.9). An increase in disposal income, a new sound system and a determination to collect as many new tracks as possible drove increases in both first-time percentage and number of cassettes recorded per year, the latter rising from 4.3 (1993-96) to 8.8 (1997-2000), on average, before dropping to 3.3 (2001-03). Not coincidentally, the middle period overlaps with my best years with first-wife; I often recorded mixes for vacations.

Table 1: Summary of mix cassettes recorded between March 1993 and March 2003

Year	# Cassettes	Total Tracks	Tracks/ Cassette	Total New	% New	New/ Cassette
1993	6	142	23.7	23	16.2%	3.8
1994	4	93	23.3	22	23.7%	5.5
1995	2	51	25.5	22	43.1%	11.0
1996	5	128	25.6	56	43.8%	11.2
1997	9	238	26.4	84	35.3%	9.3
1998	10	238	23.8	122	51.3%	12.2
1999	6	144	24.0	77	53.5%	12.8
2000	10	267	26.7	123	46.1%	12.3
2001	4	99	24.8	55	55.6%	13.8
2002	2	50	25.0	31	62.0%	15.5
2003	4	98	24.5	54	55.1%	13.5
*Total*	62	*1,548*	*25.0*	*669*	*43.2%*	*10.8*

Incidentally, the dip in new track percentage and per-cassette average in 1997 results from a two-cassette update of the “I-92 memoriam” tape I first constructed in 1994. Only three of the 52 tracks I recorded appeared for the first time. Excluding those, first-time percentage increases to 43.5% and first-time per cassette average increases to 11.6 (out of 26.6), matching the preceding two years.

In March 2003, preparing for a return vacation trip to the Boston area, I constructed a four-cassette mix – my record to that point – called Stuff and Such Vol. LXXXVI to LXXXIX. These were the 148^th through 151^st mixes I had constructed since August 1981 (146 cassettes, five videocassettes).

These were also the last mixes I ever recorded onto any form of cassette.

**********

In September 2002, I was promoted to Research Project Manager at the Family Planning Council of Southeastern Pennsylvania. Not only did this come with a substantial raise and an office, I started to earn additional income through data analytic consulting. The increase in disposable income allowed me to move into a nice one-bedroom apartment in a complex in suburban King of Prussia the following February. I also significantly upgraded my technology: I had a CD player installed in my Buick Century (loading up to six CDs into an apparatus in the trunk), and along with the CD burner I bought a DVD player and a desktop computer with enough computing power to run statistical software packages. I could now create mix CDs from CDs, cassettes and vinyl albums. The computer contained a program called Musicmatch Jukebox which allowed me to extract CD tracks onto my computer and to burn CDs. I do not recall precisely when I realized I could buy tracks from the internet and store them in Musicmatch Jukebox.

The first three CD mixes I constructed were learning experiences: a collection of tracks primarily by Stan Ridgway/Wall of Voodoo and Renaissance (August 2003), a compilation of older tracks, and what amounted to digitization of an Icehouse album with some Miles Davis tracks (both October 2003). These CDs had 12-13 tracks, well below capacity for a blank 80-minute CD. I did learn that burning CDs took far less time than recording tracks onto a cassette – and that blank CDs were generally cheaper than blank cassettes. This is why, over the next 13 years, I burned multiple CDs for the same “mix,” with only two exceptions – August 2004 and November 2010.

In November 2003 I burned a three-CD mix to listen to on an epic Saturday drive (CD Stuff Vol. III-V); I was now up to 16 tracks per CD. In June 2004, I took a long, much-needed vacation into the Northampton, MA region, burning 99 tracks onto six CDs. This latter mix of CDs, which opens with “We Will Become Silhouettes” by The Postal Service, was a revelation.

Because these were mixes designed for a long, multi-day solo drive, I wanted to start with a bang. Why I chose The Postal Service tune, more chirpy than kick-ass, is a mystery. But the first CD was “rocking,” so the second CD, opening with a recording of “Rhapsody in Blue” recorded by Eugene Ormandy, conducting the Philadelphia Orchestra, was “mellow.” The other four CDs alternated rocking and mellow, as I had done with cassette sides, but each also had its own standalone “personality.” The 5^th CD, which opens with “Bigmouth Strikes Again” by The Smiths, became a particular favorite – especially as I began taking extravagant Friday night baths featuring music and a modicum of scotch (I later added darkness, candles and lavender “milk bath”).

And just as I had ended the third of the three November 2003 CDs with the epic “Rudy” by Supertramp, I saved “Racist Friend” by Special AKA to close out the sixth of the six June 2004 CDs, its gentle fadeout adding a soft period to the mix.

Table 2: Summary of mix CDs recorded between November 2003 and August 2016

Month	# CDs	Total Tracks	Tracks/CD	Total New	% New	New/CD
Nov-03	3	47	15.7	22	46.8%	7.3
Jun-04	6	99	16.5	67	67.7%	11.2
Aug-04	1	18	18.0	13	72.2%	13.0
Dec-04	4	69	17.3	38	55.1%	9.5
May-05	6	99	16.5	66	66.7%	11.0
Aug-05	6	93	15.5	54	58.1%	9.0
Dec-05	5	80	16.0	32	40.0%	6.4
Mar-06	3	57	19.0	28	49.1%	9.3
May-06	6	99	16.5	50	50.5%	8.3
Aug-06	6	116	19.3	60	51.7%	10.0
Nov-06	6	109	18.2	51	46.8%	8.5
Apr-07	5	99	19.8	55	55.6%	11.0
Jun-07	3	58	19.3	33	56.9%	11.0
Dec-07	5	98	19.6	62	63.3%	12.4
Apr-08	5	96	19.2	49	51.0%	9.8
Dec-08	5	95	19.0	54	56.8%	10.8
Aug-09	6	118	19.7	70	59.3%	11.7
Jul-10	11	214	19.5	126	58.9%	11.5
Nov-10	1	19	19.0	15	78.9%	15.0
Jun-12	6	115	19.2	56	48.7%	9.3
May-13	12	227	18.9	127	55.9%	10.6
Aug-13	12	242	20.2	121	50.0%	10.1
May-14	12	236	19.7	125	53.0%	10.4
Aug-15	12	221	18.4	84	38.0%	7.0
Aug-16	8	150	18.8	81	54.0%	10.1
*Total*	*155*	*2,874*	*18.5*	*1,539*	*53.5%*	*9.9*

As Table 2 shows, I created 25 CD mixes between November 2003 and August 2016, using six CDs on average – the maximum I could fit into my trunk apparatus[2] – on which I burned an average 18.5 tracks, 10 for the first time (53.5%). In fact, I did not burn more than six CDs for a mix until July 2010, when I burned 214 tracks – 126 for the first time (58.9%) – for a solo trip to the Philadelphia area. Through 2009, I constructed an average of three mixes per year, limiting the number of tracks available for any mix, and averaging 17.9 tracks per CD; this increased to 19.2 beginning in July 2010. The number of CDs constructed per mix also doubled from 4.8 to 9.3, albeit with only one large mix constructed per year, on average.

A number of changes account for these increases. I received my first iPod in 2007, slowly upgrading to my current 125-GB classic flywheel model, which has 9,694 tracks on it. When I bought a new computer around 2010, I switched from Musicmatch Jukebox to iTunes. The portability and capaciousness of the iPod combined with a central, synchronized location spurred the idea to consolidate ALL of my music onto iTunes, and from there to my iPod. Copying my large CD collection onto my computer was the first step. I then slowly accumulated every track I had recorded onto a cassette, videocassette or CD I only had on a mix cassette – upgrading the quality of the recording. And, starting with albums by The Beatles I still only had on vinyl, I began to digitize my substantial vinyl collection, painstakingly recording album tracks onto CDs and then onto my computer.

I do not recall which CD mix was the first I constructed entirely on my computer, building a series of playlists – likely in iTunes on a new computer – I then burned onto blank CDs. The first time I burned ANY music onto a CD from my computer was in April or May 2005, burning the first tracks I ever bought on MusicmatchJukebox onto a CD I then used in my CD burner to make a six-CD mix in May. I was also creating potential track lists in Excel, dividing tracks into “New” and “Old” then – starting in May 2006 – using color-coding to signify on which playlist (i.e., CD) they went. I would likely have done because I was constructing the mixes on my computer using Musicmatch Jukebox, suggesting this six-CD mix was the first so constructed. Sometime between getting my first iPod in the summer of 2007 and July 2010, when I constructed the first mix to appear on my current iPod, I switched permanently to iTunes – and computer-generated CDs. And, after acquiring an iPod-to-cassette converter (likely in early 2013), I stopped burning CDs altogether (though I kept the naming convention), simply playing a “meta” playlist through my cassette deck; not having to change the CDs in the trunk was nice benefit.

The upshot was that I had upgraded from the real-time recording of tracks onto cassettes to the somewhat faster CD burner to the significantly faster burning of CDs from my computer. I had also created a far more efficient process because I not only had thousands of tracks at my disposal – with millions more a simple transaction away – I could now play snippets of a track before recording it, improving the musical flow of each CD mix and leading me often to cry aloud, “I am such a genius!!” for a particularly good end-into-beginning. Moreover, I was acquiring dozens of new tracks in serendipitous ways: a stash of CDs left behind in our apartment building, a DJ’s collection of 1980s albums I was given to store, many albums and CDs rescued when we sold Nell’s mother’s house in Georgetown.

What this also meant, however, was that constructing mixes required less and less selectivity on my part. For the first 21 years of mix-construction, I was limited by the time and expense of recording tracks onto blank cassettes. Blank CDs were a bit cheaper and faster to burn, but I still had to be somewhat selective, recognizing my car’s six-CD limit. But iPod playlists are limited only by the capacity of the iPod – and I have not yet used 40% of my current model’s capacity. I was still limited by my sequencing rules, though, and I now had a new problem. By the time I switched to CDs, more than half of my “repeat” inclusions were tracks I had only recently recorded for the first time. But when recording between 121 and 127 tracks for the first time on a set of CDs, I then want to put at least 70 of them on the next CD mix – along with 80, 100, 125 first time tracks, leaving little room (theoretically) for “older” tracks I want to include again.

That all being said, by 2012, I was only making CD mixes for my annual drive to the Philadelphia. In the summer of 2017, Nell, our daughters and I made the trip together, so I did not make a mix. For some reason, I did not make CD mixes in 2018 or 2019, the last time I drove there, due to the COVID-19 pandemic. Thus, the eight-CD mix I recorded in August 2016 is likely the last such mix I will ever make. Its 150 tracks somewhat fittingly end with the gloriously bonkers “(Bad News At The Dynamite Ranch)/Beyond Tomorrow” by Stan Ridgway, the 3,381^st unique track – covering 6,335 slots – recorded on 309 individual cassettes, videocassettes and CDs over 35-years.

**********

Back in 2004, meanwhile, I finally rebuilt my entire mix-track database using the current version of Microsoft Excel. Over my long winter holiday, I consumed an untold number of pots of coffee – mostly decaffeinated – as I patiently started from the August 1981 My Stuff mix, entering the following data from the 168 individual media (cassettes, videocassettes, CDs) – 1,984 tracks covering 3,694 slots – I had constructed to that point:

Name of individual medium
Month and year of creation
Presence – “1” if a track appeared on a mix, blank otherwise.
Position – using a three-digit number for cassettes (1,2 for side; 01, 02, etc. for position)
Affect – an ordinal 1-5 scale for how I felt at the time of recording about each track, with 5 being the most excited
Time Adjustment – an algorithm based upon how many months had passed since track was recorded.

I recreated the track listing and ordering for two early cassette mixes – My Stuff and Stuff Vol. V (August 1982) – as best as I could from memory. As of now, I can only recall seven tracks from the latter cassette. “Affect” and “Time Adjustment,” meanwhile, reflect two new ranking hypotheses:

Ranking Hypothesis #5: At the time I constructed a mix, I liked some tracks selected for that mix more than other tracks selected for that mix.

Ranking Hypothesis #6: The more time that has passed since I included a track, the less I like it now.

I was now grappling with the passage of time and the difference between “all-time” and “right-now” favorites: ≤8 years of data was one thing, but ≥23 years of data was something else entirely. While “affect” was straightforward – albeit requiring me to recall emotional reactions dating back 23 years – “time adjustment” proved diffcult in practice. I began with the simple inverse of the number of months that had passed since a track was recorded (plus 1, to avoid denominators of zero), creating a “time” weight between 0 and 1. However, 280 months had elapsed since August 1981 and 1/280 = 0.004, making time adjustment more like time punishment. To increase this minimum value, I used the square root of the ratio, which still had a maximum of 1.000. This only increased minimum time weight to 0.060. I then changed the ratio numerator to 10, increasing the minimum time weight to a reasonable 0.189. Of course, to maintain a maximum of 1.000, I had to make sure the numerator was never larger than the denominator, which – to be fair – was only a problem for the most recent mixes.

As for those recent mixes, they were suddenly receiving too much weight, especially if a track appeared on a mix both two years and one year ago, giving them a time adjustment as much as 10 times the minimum. Older tracks I still liked but had not yet rerecorded were thus severely down-weighted relative to tracks with which I was infatuated NOW, but may not like as much in the future. Conversely, there were tracks I very much liked that I planned to record again, making those tracks’ “actual” value much higher than what the algorithm would suggest. And then there were the related questions of burnout – recording a track too many times over a relatively short period of time – and rerecording a track because I had a “cleaner” recording (e.g., CD vs. scratchy LP) or I was updating an earlier mix (e.g., 1994 vs. 1997 “I92” mixes).

This led to the final three ranking hypotheses:

Ranking Hypothesis #7: Tracks recorded on mixes in the current or previous calendar year should, all things being equal, have lower scores than tracks recorded two or more calendar years ago.

Ranking Hypothesis #7a: Tracks recorded in the current calendar year should, all things being equal, have lower scores than tracks recorded in the previous calendar year.

Ranking Hypothesis #7b: Tracks recorded for the first time in the current or previous year should, all things being equal, have lower scores than previously-recorded tracks recorded in the current or previous year.

Ranking Hypothesis #8: Any track recorded on a mix for reasons other than a desire to hear that track on that mix, or recorded on a mix shortly having been recorded on two mixes in a short period of time, should receive no credit for this recording.

Ranking Hypothesis #9: If a track has not been recorded on a mix for a substantial period of time and there are concrete plans to record that track on a future mix, the current score should anticipate that recording.

I will spare you the convoluted algorithm I ultimately devised to account for the latter five ranking hypotheses, with “future placement” becoming an easy – and highly subjective – way to increase or decrease a track’s final score if it lacked “face validity,” meaning it just didn’t feel right, particularly when ranking tracks overall, or within artist or year of release.

The point being that as much as I tried to use an objective algorithm to construct a valid score for each track, I was already making hard-to-defend adjustments to individual elements. My straightforward notion – adjusting presence on a mix by time-of-recording affect; passage of time; and various measures of burnout, future recording and recency – proved extremely difficult to execute. The individual measures were reasonable – the final score had reasonably high construct and content validity – but my “objective” algorithm had far more “subjectivity” than I had anticipated.

In essence, I found myself here:

Perhaps this is why, after quickly updating these scores so I could count down – on my 40^th birthday in September 2006 – my favorite song released each year of my life, I stopped regularly updating these scores.

It was also the case, as we will see in Part III, that I was creating new and complementary favorite track measures.

Until next time…please wear a mask as necessary to protect yourself and if you have not already done so, get vaccinated against COVID-19! And if you like what you read on this website, please consider making a donation. Thank you.

[1] I exclude a video mix I recorded in 2002.

[2] Which I had transferred to my new black 2005 Honda Accord when I bought it on September 6, 2005, just a few days after moving back to the Boston suburbs again.