Washington Capitals line combinations

The NHL publishes a shift chart for every game: for each player, the moment each of his shifts began and the moment it ended. It does not publish lines. No feed says which three forwards a coach considers his second unit, and a site that shows you one has either licensed it from a vendor who watches warmups or typed it in after reading a beat reporter. What the shift chart does support is a measurement — which three forwards were on the ice together, for how long, and in how many games — and that measurement is what this page is.

A measurement and a depth chart are different objects, and the difference shows up exactly when they disagree. A coach who names a line in the morning and splits it up after the first period has one depth chart and two measurements, and this page shows the second thing. It cannot tell you what he intended. It tells you what he did, with the evidence attached.

The naive version of this measurement — rank every pair of teammates by the ice time they share — produces a table in which a club's top power-play unit outranks its actual fourth line, and nothing in the output admits it. Five skilled players sent out together for two minutes at a time will always beat three checkers rolling a dozen short shifts a night, and a reader looking at the finished table has no way to tell which row is which.

So only five-on-five play is counted here. Shift data carries no strength state of its own, and it does not need one: counting how many non-goalie players each club had on the ice at each instant is the strength state, measured continuously rather than sampled whenever a play happened to occur. Intervals with five a side survive. Power plays, penalty kills, four-on-four, three-on-three overtime and the scramble after a goaltender is pulled do not, because none of them has five a side. The distribution that filter produced is printed on this page, so it can be checked against what a hockey game is known to look like rather than taken on trust.

Filtering that hard costs coverage, and it is the right trade. A meaningful slice of every game is played at some other strength and every second of it is discarded here. What is left is the state in which a line is actually a line, and a unit that only ever appears on the man advantage does not appear at all.

Coaches juggle, and they juggle hardest when losing. A winger can finish a single game having taken shifts with three different pairs of forwards, and across half a season the number of distinct combinations he appears in runs well into three figures — most of them seconds long, the residue of a change that went wrong rather than a decision anyone made.

That is why every unit here arrives with its evidence rather than only its total: the time together, the number of games it appeared in, and the number of separate shifts it was sent out for. A trio seen twice and a trio seen forty times must not look alike, and a table sorted by minutes alone would let them. Time per game is shown beside every total for the same reason — a unit with a large total and a small per-game figure is one that has been together often and briefly, which is a different animal from a line.

It is also why the chart on this page exists. It shows, for each skater, how much of his own five-on-five ice time went to his single most-used combination. A defenceman at four fifths has a partner. A centre at a third has a rotation, and naming any one of his trios his line would be a confident answer to a question that has no single answer. The count beside it — how many combinations it takes to cover three quarters of his ice time — is the same fact stated the other way round.

It cannot see intent, injury or the morning skate. A line that appears for two games and vanishes may have been broken up deliberately or may have lost its centre to a groin strain, and the shift chart records the same thing either way. It does not know about a trade: a unit's games are counted where they were played, so a club's list can hold a combination that no longer exists and will not say so.

Positions come from the league's own listing, which is looser than it looks — plenty of players listed at centre spend the season on a wing — so a trio of three listed centres here is usually an ordinary line rather than a fault. Ordering is by measured time together and by nothing else. Nothing in shift data says which trio is a first line, and presenting a ranking as a depth chart would be inventing one. The rows are not numbered for that reason.

The sample is whatever has been ingested, and the window it covers is printed on this page. It is not always a full season. Two clubs measured over different numbers of games are not comparable on totals at all, which is why time per game sits beside every total, and a club whose shift charts have not been ingested is absent rather than shown as a club with no lines.