When Otis Redding Sings Soul appeared last night my heart sank. I squeezed my eyes shut tight and let a tear or two fall. Why such a sudden and emotional reaction? Well, I’d recently finished granting a glowing review on Low-Life after a few days of similarly high ratings, and I knew with high certainty that this, too, would be a fantastic 5-star listen.
Why would the prognosis that I get to have a sweet and fulfilling album-listening experience render me so crestfallen? Well, I know that any string of 5-star ratings I bestow leads directly to a bolstering of the characterization of my personality by the algorithm, supported by other various sundry observers, as a “cheerleader”. The intent of this characterization is not to flatter, but to imply that granting a bunch of higher-than-average ratings is an attempt by me to artificially build the confidence of undeserving artists, and that perhaps my ratings are biased toward that purpose. The only reason such a criticism may hurt me is that there is a kernel of the same question buried deep inside my own questionable psyche.
In order to address this issue in my own mind it is necessary to briefly discuss the nature and purpose of statistical generalities and reductions. Normalizing data sets and manipulating data to bucket out outliers and use such assessments to develop a better understanding of patterns and trends in the universe is a time-honoured and valuable use of data. We can use this kind of work to predict the likelihood of a particular outcome in a general sense, but the reliability of these kinds of prediction's goes down the closer they come to trying to predict a specific outcome. For example, sockfire has an unusually high likelihood of awarding Korn- Follow the Leader 5-stars. It is the same likelihood of sockfire awarding any album 5-stars. When Korn was awarded 1-star the predictive power of the trend was nil. For this reason, the data can only ever be used as a general guideline informing reasonableness. Any prediction of a specific outcome must be informed by situation-specific conditions to be relied upon, and there must be an assurance that the source of the data is objective. The source of the data cannot be influenced by any objective to generate a specific curve shape. Data reduction or “grading on a curve” in order to develop general observations can only be fairly accomplished once all the data is in and quality assurance is established.
It is important that we, the project users, maintain a high degree of integrity. We are not simply observing this data, but we are also the SOURCE of the data. We cannot allow a pre-characterization of our data set to bias our ratings, lest any characterization of the trend becomes suspect. This reality both invalidates the algorithm’s premature characterization of my ratings as “cheerleading” but also clearly requires that I discount any internal pressure to or to not rank a certain way on any case-by-case basis.
With this discussion in mind, reducing Otis Redding Sings Soul to a single data point becomes simplicity itself.