DWTS Season 28 - Historical Placements

This (belated) Season 28 analysis is meant to assess the relative strength of Dancing with the Stars contestants based on historical comparisons.It includes an additional factor to account for the new elimination procedure. Note that these are not meant to be predictive. Age and height are based on best available Internet data.

Job-age comparison groups for each star are listed below. "Job" categories derive from ABC's now defunct "Cast DWTS" website. Age ranges are typically the star's age +/- 5 years (with exceptions noted). Past seasons' results are scaled from 1-12 (to represent 1st through 12th place) so that they are comparable regardless of field size. Withdrawals, All-Star, and Athletes seasons are mostly excluded.

Comparison groups are listed from strongest to weakest average.  (Scroll to view all groups):


Notes: Age range was broadened for Mary, and age-job category broadened for Karamo and Sean, because of insufficient comparison..

Here's a summary of the age-job group averages, in clusters of strongest to weakest historical average placement. Red print indicates averages based on limited or adjusted comparison groups.

Here are the pro averages, from strongest to weakest. These averages include results from the Athletes season. Red print indicates averages based on limited pro histories. For new pro Pasha, the listed average is that of four male pro debuts within the last 10 seasons (Artem, Keo, Alan, Brandon).

Next are the height group averages. For each star, the historical average for same gender height +/- 1 inch is shown from highest to lowest average placement. Red text indicates guesstimated height for Sean, and +/- 3 inches to compensate insufficient comparisons for Lamar.

Next are the weighted overall averages*, arranged in clusters from strongest to weakest. The "Place" column shows rounded averages to better represent places 1-12.  Please note that this chart displays automatic clustering by average round placement. (A fuller consideration of automatic clustering can be found in this retrospective of Season 27 results.)

*Technical Note: Averages are weighted as 50% Age-Job average, 33% Pro average, 17% Height average, based on correlation analysis of weighted historical averages versus actual historical results.

Finally, because season 28 introduced a new elimination process, I attempted to adjust these historical averages accordingly. To factor forthcoming judging deliberations into the overall averages, scores from the first two weeks are employed as proxies for judges' relative preferences. (Total scores were scaled 1.0 to 12.0 for 1st to 12th place.) The weighted averages are shown below, with clusters listed from highest to lowest average placement.

Kate and Mary (highlighted in blue) are considered wildcards because of their pros' limited histories.

*Technical Note: Averages are weighted as 31% Age-Job average, 25% Pro average, 25% scaled Scores, 19% Height average.

Comments:

First, a reminder that these are not predictions. Inevitably, some stars will beat their historical averages and some will fall short. The general expectation of cluster analysis is that stars are most likely to finish within their relative clusters, less likely to finish +/- 1 cluster away, and least likely to finish more than 1 cluster away. With just three clusters, there should be relatively few surprises. As of this writing (heading into week 6 of the competition), the historical top cluster (historical top 4) has already avoided any possibility of finishing in the actual bottom cluster (actual bottom 3). Also, two of the historical bottom cluster have already been eliminated. At this point, the only historical deviation would be the remaining bottom cluster star placing in the top 4.

Retrospective: DWTS Season 27 - Historical Comparisons

The results of Season 27 on Dancing with the Stars were unexpected in many ways, but not without historical precedence. In a preseason 27 analysis, I included the following note:

"Please note that this chart displays automatic clustering by average round placement."

I included that note because the clustering method was different than previously posted analyses. Previously, I had clustered groups based on spacing between overall averages, while avoiding single-entry clusters. Had the season 27 analysis been clustered as previously, there would have been three clusters, with a middle cluster of six contestants (Mary Lou Retton through Bobby Bones).

Three clusters are generally less interesting from an interpretive standpoint than four or five clusters. Large middle clusters are less interesting still. In this example, six of 13 contestants would have finished within one historical cluster no matter where they placed. The only "interesting" result would have been Nancy McKeon being eliminated much earlier than history would suggest.

Automatic clustering by rounding to the nearest whole placement produced a more interesting visual chart, but obscured the close spacing among the middle averages. In particular, Bobby Bones' historical average was within 1.0 placement of seven other contestants. The historical overreach of his (and Sharna's) final placement was primarily with respect to the top cluster of DeMarcus, Juan Pablo, and Nancy (and their respective partners).

The upside of the automatic clustering was that the apparent display of deviations from historical patterns matched public perceptions. The outcry over the season 27 results indicated that those results at least appeared unexpected and interesting, even if historically they were arguably less so.