Presidential Grading: Method and Findings

I have graded all forty-four presidents, and I want to show the work.

What follows is the whole method, written out so that someone else can reproduce it, attack it, and see exactly where my judgment enters. The short version: the letter grades are mine, and the underlying scores are not. Every number below comes from C-SPAN’s 2021 Historians Survey, and the transcription was checked three independent ways before I used any of it.

The reason to write it down is that a ranking presented as a bare list is unfalsifiable. You can disagree with it; you cannot check it. A ranking presented with its rubric lets you find the exact cut point where you would have decided differently, which is the only useful thing to learn from a disagreement of this kind.

One more thing before the method. The point of this exercise is not to produce a better number than C-SPAN’s. It is to produce a number that can be argued with in public, and to be honest about which parts of it are arithmetic and which parts are me.

1. Provenance

C-SPAN scores each president on ten characteristics and publishes both the per-category scores and an overall figure. The overall figure is the equal-weighted mean of the ten categories, which is what made the transcription checkable (§3).

This is not a primary-source assessment. It is a re-reading and re-grading of expert judgments. Every factual claim underneath belongs to the 142 respondents; only the grades, the weighting argument, and the analysis are the author’s.

2. The ten characteristics

In C-SPAN’s published order, which is the order used throughout this document and in every profile string. Position matters; the string is read positionally.

3. Transcription validation

The per-category scores were transcribed from C-SPAN’s published tables. Before any analysis, three independent checks were run against C-SPAN’s own published numbers.

Check 1, arithmetic. C-SPAN publishes each president’s category scores and an overall score equal to their mean. If the transcription is right, the ten values must sum to that published overall figure.

43 of 44 presidents: sum of ten categories == published overall score
 1 exception: Monroe, 643.6 transcribed vs 643 published
              (0.09%, inside C-SPAN's own rounding; does not cross a grade band
               at any cut point, so Monroe's grades are unaffected)

Check 2, order. Independently of the arithmetic, the ranking implied by the transcribed scores was compared against C-SPAN’s published overall ranking.

44 of 44 rank-order agreement

Check 3, independent re-fetch. A later copy of C-SPAN’s live results page was fetched and parsed from scratch, with no reuse of the original transcription. The ten category scores were then compared score by score.

42 of 44 presidents: all ten category scores reproduced exactly
 2 exceptions: name-format differences in the page's markup, not score differences
0 profile strings disagreed

Three independent checks passing is the basis for treating the dataset as faithful. Every figure in this paper is reproducible from the C-SPAN source tables.

4. The grading scale

Each of a president’s ten published scores is converted to one of four numbers:

The cut points are a choice. They are not in C-SPAN and not in any standard. They are stated here so they can be argued with. Sensitivity was tested: moving the cuts to 90/75/60 or 80/65/50 leaves the ordering correlated at 0.99 and 0.98 respectively, and at no cut setting tested does the membership of the top four change.

The bands are structurally unequal. Bands 4, 3, and 2 are 15 points wide; band 1 spans 55 points. Consequence: 26.7 points of real C-SPAN difference collapse into a single grade at the bottom, while 15 points separate a 3 from a 4. This is the single largest artifact introduced by the grading scheme, and §7 discusses it.

5. The profile string

Each president is represented as a ten-character string of digits 1–4, in the category order of §2.

Kennedy    3322222333
Reagan     4222312313
Eisenhower 3333333223
Obama      3223221332
Lincoln    4434343444

Read positionally: the first digit is Public Persuasion, the second Crisis Leadership, and so on. 4222312313 says Reagan graded 4 on persuasion, 2 on crisis leadership, and 1 on moral authority and equal justice.

The string is not unique. Of 44 presidents there are 29 distinct strings:

Fourteen presidents receive all ten grades identical. That is a property of the scale, not an error: 26.7 points of spread fit inside band 1.

The string is not a hash of the man. It is a hash of a man under a stated rubric. Two profiles are comparable only if the same cut points and the same category order produced them. Uniqueness was tested as a goal and rejected: increasing resolution to 9 decile bands does yield 44/44 uniqueness, but under ±1.5 points of plausible scoring noise that uniqueness fails in 25% of trials. Uniqueness for a 44-item census is a property of the sample, not of the encoding. The encoding is kept coarse, readable, and tied to a declared rubric.

6. Sorting

Presidents are sorted by the mean of the ten grades, which is the arithmetic mean of ten integers in 1–4, so its values are always multiples of 0.1.

This is the only use of averaging in the method. It is retained for sorting and for the compact summary of the ten grades. It is not treated as a measure of a president; §7 tabulates the profiles that the sort collapses.

Sorted order (first ten):

The mean grade reproduces the C-SPAN ordering at Spearman 0.94 while using only 17 distinct values across 44 presidents.

7. What the sorting destroys, and why no order should be quoted alone

A mean hides which categories earned the grades. Four examples, in each case identical by the number and different by the record:

A mean rewards flatness and punishes distinction. The clearest cases of the grading shifting a president’s position relative to C-SPAN’s own ordering:

A correction to an earlier draft of this analysis: it was claimed that the scale systematically punishes spikiness. Tested, the correlation between a president’s within-ten spread and his rank shift from grading is only −0.275, too weak to support the general claim. The mechanism is real in the individual cases above, but it is not a systematic bias of the scale, and any statement to that effect would have been an over-claim. The flatness reward is better demonstrated by the specific pairs in the first table of this section than asserted as a rule.

8. The rubric problem, and the weighting argument

C-SPAN weights its ten characteristics equally, each contributing 1/10 to the overall figure. The grading inherits that weighting. It was never argued for. It means “Pursued Equal Justice for All” counts exactly as much as “Economic Management,” which is a substantive position nobody has defended.

The equal weighting also produces a baseline worth naming, because the labels imply otherwise:

mean grade across all 440 ratings (44 presidents x 10) = 1.72
distribution:  4: 5.0%   3: 14.5%   2: 27.5%   1: 53.0%

The average president grades below “adequate.” The labels (“superior / satisfactory / adequate / unsatisfactory”) imply a passing midpoint, but a 1–4 scale with no neutral point has its midpoint at 2.5. The labels and the arithmetic disagree, and the labels are the ones in the published output. Either band 2 should be relabelled, or the labels should be treated as ordinal only.

Weighting is where authorial judgment belongs. Because the weights are explicit, they can be varied and the effect measured. Two findings, both reproducible:

Finding 1: the weights do not move the top. Across 2,000 random weightings drawn from a Dirichlet distribution:

exact top four = {Lincoln, Washington, F. D. Roosevelt, T. Roosevelt}  1,926 / 2,000
presidents ever entering the top four, across all weightings          9

Lincoln is unassailable. Washington and T. Roosevelt can be pushed down, but only by weightings that zero whole categories, which is no longer a ranking. For any defensible rubric the top four is fixed.

Finding 2: the weights decide the middle of the table. This is where authorial judgment has force, and it is the finding that matters for anyone tempted to argue with the order:

The top is stable, the bottom is volatile in a way that mostly reflects how little the data distinguishes those presidents, and the middle is where a considered judgment changes the answer.

9. Charisma: the empirical result

Public Persuasion is the highest-variance characteristic after grading: SD 1.065, the largest of the ten, 12.1% of total grade variance.

Two measurements:

• Excluding charisma entirely leaves 16 of 44 ranks identical, measured against the graded order. (Against C-SPAN’s own order it is 13/44; the two baselines differ, so the baseline is named here rather than left implicit.) That makes it the most consequential single category to include or omit among the mid-range categories. Economic Management is 34/44 against the same baseline, Moral Authority 24/44, Equal Justice 27/44.
• Tripling its weight changes no rank by more than six places. Reagan moves +3, Jackson +6. The top five never move.

Both facts hold at once, and together they are the finding: charisma is consequential but not directional. It matters more than any other single category, and it moves the order least, because at the extremes it agrees with everything else: the presidents who scored 4 on persuasion also scored well on vision, crisis leadership, and the rest.

Where it does bite is the middle: Jackson (persuasion 78.6, relations with Congress 51.9) and Reagan (89.1 and 68.4) are the men whose standing depends on whether mass persuasion counts as leadership. Charisma is a tiebreaker in the middle of the table, not a driver anywhere.

A trait the single score conflates. Public Persuasion (mass) and Relations with Congress (retail) correlate at only r = 0.832, so they are related but not the same ability. Men who had one and not the other:

A single “charisma” score double-counts the men who had both (FDR: 94.8 and 80.5) and hides the specialists. LBJ is graded 2 on persuasion and 3 on Congress; he was one of the most effective legislative operators in the office’s history.

10. Summary of the method

1. Take C-SPAN’s 2021 per-category scores for all 44 presidents (10 each, 0–100).
2. Validate the transcription twice, against C-SPAN’s own published numbers (arithmetic and ordering). Both must pass before anything downstream is used.
3. Grade every score on a four-point scale at stated cut points (85/70/55).
4. Emit a ten-character profile string in C-SPAN’s category order.
5. Sort by the mean of the ten grades, for ordering and compact summary only.
6. Report the profile alongside any mean, never the mean alone.
7. State all weights; treat the top four as rubric-invariant and the middle as judgment-dependent.

11. Stated limitations

• Secondhand. This re-grades 142 historians’ judgments. It is not a reading of the primary record, and it inherits every blind spot of that survey, including the ones its own participants named. Edna Greene Medford, on the 2021 results: “we still have slaveholding presidents at or near the top of the list.”
• Time-bounded. The data predates Biden’s presidency and Trump’s second term.
• Recency effects are documented in the source itself. Richard Norton Smith described a “boomerang effect” in which presidents are rated at a nadir just after leaving office and rise later; Ulysses S. Grant is the survey’s largest mover, +13 places between 2000 and 2021.
• No single-number score should be quoted without its profile. §7 documents the specific instances where the mean states the opposite of the record.
• The cut points are unargued. They were chosen to be round. Sensitivity testing shows the ordering is robust to them, but a different author could pick others and is entitled to.

Sources

All figures in this paper were transcribed from C-SPAN’s published tables and then verified twice: once against C-SPAN’s own published totals (§3), and once against a freshly fetched copy of the live results page, which reproduced all ten category scores for 42 of 44 presidents exactly. The two remaining presidents differ only in the name format used by the page, not in their scores.

• C-SPAN. 2021 Historians Survey of Presidential Leadership: Total Scores and Overall Rankings. https://www.c-span.org/presidentsurvey2021/ This is the primary citation. The live page carries the overall scores and all ten per-category tables (440 values), which is the dataset this paper grades.
• C-SPAN. 2021 Press Release: C-SPAN Releases Fourth Historians Survey of Presidential Leadership, 2021-06-30. Participant count, advisory panel, and the survey’s own commentary on category movement. https://static.c-span.org/assets/documents/presidentSurvey/2021%20Press%20Release.pdf

Note on the PDF mirrors. C-SPAN also publishes the survey as standalone PDFs (2021-Survey-Results-Overall.pdf and 2021-Survey-Results-Category-Rankings.pdf under static.cspanvideo.org). Those URLs serve a GDPR consent interstitial to automated requests rather than the file, so they are not cited as the verification route: a reader following them may land on a consent gate instead of the data. The HTML results page above is reachable and carries the same figures.

PRH | huffmanwrites.org | © Philip Huffman