Print Figure 2 of the Federal Reserve's Summary of Economic Projections from last Wednesday, September 16, and count the dots in the first column. Each dot is one policymaker's view of where the federal funds rate should sit at the end of 2026. There are eighteen of them: four at 4.375 percent, twelve at 4.125 and two at 3.875.2 Now turn to the Fed's own guide to the document, which says the projections are collected from each member of the Board of Governors and each Reserve Bank president.7 The Board's web page lists seven governors, with no vacancy, and there are twelve Reserve Banks.8 That makes nineteen. One place on the page is empty, and the document does not say whose it is.
The missing dot belongs to the chair, Kevin Warsh. He said so himself at the press conference that afternoon, as he had in June.45 And since June the Committee's statement has carried no forward guidance, no sentence about the likely direction of the next move.6 Those two changes alter what kind of document the dot plot is. It has always been a survey: individual answers from a known list of people, summarized by a median. For most of its history it was also read, with some justification, as a sketch of a plan: a path the Committee and its leader intended to follow. This year the parts that supported that second reading were removed.
My argument is that the table is now plainly a survey of eighteen colleagues drawn from a frame of nineteen, and that its median tells you where the Committee leans, not what it plans to do. This is not a claim that the dots are useless, that the missing dot biases the median, or that the Fed will or will not raise rates again. As it happens, in September the missing dot could hardly have moved the medians at all. The claim is about what the number can stand for.
The table records who answered, and what they answered.
Nothing on the page says who did not, or what anyone intends.
What happened last week
The news itself was expected. The Federal Open Market Committee raised its target range by a quarter point, to 3.75 to 4 percent, by a vote of 12 to 0.1 The day before, futures prices implied about a 92.5 percent chance of exactly that, according to the CME FedWatch tool as reported by Fox Business.12 Inflation had picked up over the summer: consumer prices rose 0.4 percent in August and 3.4 percent over the year, with gasoline doing much of the work.13 The statement was short. Inflation remains elevated, it said, and the Committee will deliver price stability. It said nothing about what comes next.1
The projections moved up. The median participant now puts the funds rate at 4.1 percent at the end of this year and next, and 3.9 percent at the end of 2028, against 3.8, 3.6 and 3.4 percent in June.2 It is tempting to read those medians the way they have been read for a decade, as the Fed's path. The rest of this piece is about why that reading claims more than the document does.
What the table is
The Fed did not set out to publish a plan. When Ben Bernanke announced expanded projections in November 2007, he explained why the Fed would publish individual views rather than one consensus forecast: a committee so large and so spread out across the country would find a detailed consensus forecast "quite difficult as a practical matter."9 The first summary appeared that month as an addendum to the minutes.11
The rate dots came later, in January 2012. When the Committee agreed to add them, the minutes recorded a warning: some participants feared the public could mistake the projections as the Committee's intention to follow "a specific policy path," rather than as individual views that depend on how the economy turns out.10 Medians were added in 2015.11
The document still says what it is, if you read the small print. Participants submit their projections of the most likely outcomes for growth, unemployment and inflation. The rate projections are different: the SEP says they are not forecasts of the likeliest outcomes for the funds rate, but each participant's "individual assessments of appropriate monetary policy."2 So the table answers two kinds of question at once. For inflation it asks each person what will probably happen. For the policy rate it asks what each person thinks should happen, given their own outlook. Neither question is "what will the Committee do."
The missing row
For years that distinction mattered less than it might have, for two reasons. The chair submitted a dot like everyone else, and the chair is the participant who, by long practice, leads the Committee to its decision. And the statement often carried guidance about the likely direction of policy. The April statement this year still had a sentence about "the extent and timing of additional adjustments."6 With both in place, the median dot was a survey that behaved roughly like a plan, because the person most likely to set the plan was inside the survey and the statement pointed the same way.
In June both supports went. The guidance sentence disappeared from the statement, and it is still absent.61 And Warsh, asked about that month's projections, said that half his colleagues wanted the rate at its current level or lower by year end and half wanted it higher. Then he added: "That 19th voter was me, and I didn't submit one."5
The June dots bear him out. Of eighteen year-end dots, nine sat at or below the then-current midpoint of 3.625 percent and nine above it.3 With an even number of answers, the median is the average of the middle two, so the published median of 3.8 was really 3.75, a value no participant wrote down. A nineteenth dot would have broken the tie and made the median either 3.625 or 3.875. The missing participant was, by his own description, the tiebreaker.
In September he kept to the same course. He presented the table, he said, as the views of his colleagues, without a projection of his own.4 The SEP's own note says only that eighteen participants submitted, one of whom skipped 2028 and 2029.2 In March, nineteen had.3 When a reporter pointed out that the new median did not return inflation to 2 percent until 2029, Warsh's answer was the cleanest description of the document I have read: "Those are the forecasts of my 18 colleagues."4
Here is the surprising part. I added a hypothetical nineteenth dot to each column of the September chart, once at the lowest value anyone wrote and once at the highest. For 2026, 2027 and 2029 the median does not move either way. Only 2028, where two answers are missing, changes, and only if the extra dot is low: 3.875 becomes 3.75. Figure 2 lets you try it. This September, the empty place changes what the table means, not what it shows.
An earlier essay here, The missing row still has a cause, argued that a database records only the people who passed through its collection process, and that missing rows can quietly bias an average. This case is different in a useful way. The missing row is one, it is known, it is deliberate and it was announced by the person missing. My check shows it barely moves the numbers. The issue is not bias. It is what the table can claim to represent.
Read it as a survey
A survey is a perfectly good thing to read, as long as you read it as one. The classic example is more than a century old. In 1907 Francis Galton collected 787 cards from a weight-judging competition at a fat stock show in Plymouth, where visitors paid sixpence to guess the dressed weight of an ox. He took the middle estimate, the median, rather than the average. It was 1,207 pounds. The ox weighed 1,198.1415 The FOMC median is Galton's middle estimate applied to policymakers.
But Galton's case worked because of conditions the dot plot does not meet. His voters all answered the same question about one fact, the weight of one animal. Each dot answers a different, personal question: what policy would be right if my own outlook comes true. And in Galton's crowd every card counted once. In the dot plot, the participant whose view usually carries the most weight did not hand one in. The middle of the crowd can be a fine summary of the crowd and still not be anyone's plan.
So read it with a survey's questions. Who answered? Eighteen of nineteen, and seventeen for the later years. What were they asked? Appropriate policy, not likeliest outcome. How large is the uncertainty? For the end of 2027 the dots run from 3.125 to 4.375 percent, a spread of 1.25 points, with eight of the eighteen at the top. The Fed's own 70 percent confidence band for that year, built from its past forecast errors, runs from 2.4 to 5.8 percent, a width of 3.4 points.2 The band of historical error is nearly three times as wide as the disagreement in the room. Even a complete set of nineteen dots would be a much less certain thing than a single median makes it look.
Objections, considered
Everyone knows the dots are not a plan. The Fed says so. Many readers do know. But the Fed's own minutes in 2011 foresaw the misreading, and in June a reporter told the chair that markets had taken the dots as forward guidance.5 The point here is not that the warning is missing. It is that this year removed the two things that used to make the misreading roughly right.
One dot out of nineteen hardly matters. In September that is mostly true, and my own check says so. It is the limit of the claim. But June shows it can matter: the median was a tie, and the missing voter was the tiebreaker. More to the point, the missing dot is not a random one. It belongs to the person who presents the table and usually leads the decision. Its absence changes what the median can stand for even when it does not change the number.
With no guidance, the dots are the best signal left, so calling them a survey is pedantic. A survey is a good signal, and nothing here says to ignore it. It says to read it with the tools that fit: the frame, the count, the question asked, the spread and the error band. The chair himself said last week that he is not in the forward guidance business.4 Taking the median as a plan reads into the document a commitment its own presenter declines to make. None of this is a view on where rates go next.
A median of colleagues tells you where the room leans; only someone who commits can tell you where it is going.
Before you quote the next median
- Count the dots. Compare the count in each column with the frame of nineteen, and note the columns where it drops.
- Find out who is missing. The SEP will not say. The press conference may.
- Read the distribution before the median. Two camps with a gap between them are not the same as a cluster, even if the medians match.
- Check whether the median is a value anyone wrote. With an even count it may be the average of two answers no one gave.
- Hold the spread against the Fed's own error band. The disagreement in the room is usually far narrower than the record of past misses.
- Note which question was asked. Rate dots are views of appropriate policy, not forecasts of what the Committee will do.
Back to the printed page. The gold column is a careful survey: eighteen answers, honestly counted, with a median in the middle. The blue lines are what would turn it into a plan, and this year every one of them is blank, starting with the nineteenth place. What this rate does to a project's arithmetic is Sunday's question.