It is often thought, especially by elite figures in politics, the media, and academia, that high levels of prior political experience are important for a candidate for US President. This is a reasonable theory. It would make sense if experience in a candidate for public office were valued much as experience in a candidate for being a doctor or an engineer were valued: one would generally not prefer to receive brain surgery from an enthusiastic amateur promising to sweep into the hospital and drain the swamp of medical professionals.

In the 2016 election, supporters of Hillary Clinton proclaimed that she was a candidate with unusually high levels of experience as part of their claims that she was a good candidate who should be supported, that is, linking experience to candidate quality. Outgoing President Barack Obama even claimed, In fact, I don't think there's ever been someone so qualified to hold this office. Ellen Fitzpatrick, a Professor of History at the University of New Hampshire, defended H Clinton in the NYT by citing Mrs. Clinton’s career in public service, unrivaled by any female presidential candidate — and almost any male president.

As was widely pointed out at the time, these claims are obviously false: H Clinton's eight years as a Senator and four years as a Cabinet member are historically not especially notable. Within the preceding few decades alone, the Democrats had nominated candidates as experienced as John Kerry, a twenty-year Senator and two-year Lieutenant Governor, and Bill Clinton, a twelve-year Governor; and the Republicans had nominated John McCain, a twenty-two year Senator and four-year Congressman, and Bob Dole, a twenty-eight year Senator and eight-year Congressman. Even her main primary opponent, Bernie Sanders, was a ten-year Senator, sixteen-year Congressman, and eight-year Mayor. H Clinton's experience was only really notable in contrast to her eventual opponent, Donald Trump, who had no experience of public office at all.

But the truth or otherwise of the claims is less interesting than the fact that they were made, in such prominent interventions and by such significant advocates, as if they were positive reasons to back H Clinton. These claims are not unique to Democrats either: when it was Republicans' turn to nominate a much more experienced candidate, McCain vs Obama, they trumpeted his experience and scorned Obama's inexperience. Part of this, in both cases, was political opportunism, yes, but political opportunists only make arguments that they believe will have resonance.

The question is also important for what it implies about other, wider questions. In the narrow H Clinton case, if she lost despite her experience advantage working in her favour, that should prompt Bayesians to adjust our priors in favour of the feminist hypothesis that female candidates are the victims of patriarchal oppression. Conversely, if experience is not generally a factor, or is even a hindrance, we should adjust our priors in favour of the realist hypothesis that there is no patriarchal oppression affecting female candidates. Other wider questions include the relationship between American voters and the establishment, and how parties should choose their candidates in future to maximise their chances of winning.

So is candidate experience a good predictor of success in US Presidential elections? My original analysis says otherwise. As a rule, the winning candidate will be either the candidate who has already been President (or Vice-President), or the candidate with the least experience, not the most. The effect is not merely neutral: it favours the less-experienced candidate.

The data points

There have been fifty-eight Presidential elections in the United States since its founding, fifty-four of which were contested by more than one major party.¹ Thirty of those fifty-four elections featured a former (usually incumbent²) President, and a further six featured a former Vice-President³ running against a non-incumbent. In the remaining eighteen elections, neither candidate had been either President or Vice-President.⁴

The experience data is inevitably more subjective and less clear-cut. I have included data for ten distinct categories of political office, and weighted them by duration of office, recency of last holding that office, and a weight for relevance to the position of President of the United States. It is important to note that these are only my subjective choices, and other weights, especially for the relevance score, would produce different results. Anyone who wishes to perform their own dissenting analysis should feel at liberty to do so.

The ten distinct categories of political office included are: President⁵; Vice-President; Cabinet member (including Ambassador); Governor; Lieutenant Governor; US Senator; US Congressman; State Senator; State Congressman; and Mayor. I have not included experience in the military, business, or policy work.⁶ Any analysis has to have some limit.

The scores for these groups, I decided on by setting the Presidency itself to 100 and a US Congressman to 1, and pegging all other scores to that. All scores are my own intuitive sense, informed by a small number of considerations about how the role relates to the Presidency. First of all, the existence of a role is something in itself, as everything can be spun into a tale of experience, so minor roles are weighted up from where they would otherwise be. Second, and this accounts for the largest variance, is the number of people who hold the office at any given time. For example, there are roughly four times more US Congressmen than US Senators, so US Senators have four times the weight of US Congressmen. The third consideration is the tension between the wider stage vs executive power. Finally, junior offices have been counted at roughly a fifth of the senior office.

So, the scores are as follows:

President	100
Vice-President	20
Cabinet	2
Governor	6
Lieutenant Governor	1
US Senator	4
US Congressman	1
State Senator	0.5
State Congressman	0.1
Mayor	0.5

All of these are included for completeness. However, as we will soon see, most of them don't make much difference to the results in practice.

Duration of service is included as a direct input.⁷ Recency of service is slightly modified: √(((r/4)+1)^-1). Recency itself has to be made positive as it is a denominator and incumbents would otherwise have a value of 1/0, so there is a +1 there. The division by 4 relates it to the length of Presidential terms, and mostly has the effect of reducing the denominator, making the eventual score more readable. It also smoothes out the difference between incumbents and people who recently left an office. The score is finally square rooted to reduce its importance relative to duration. This is another subjective decision that you should feel free to alter if you do your own analysis. The reason is that time elapsed since leaving office seems to me to only have a weak effect on narratives about candidates' experience.

The experience score for each candidate is then: Σ[ w * d * √(((r/4)+1)^-1) ] across the ten categories. As an example, upon standing for the Presidency, Calvin Coolidge had been a 2-yr State Congressman 16 years prior [0.1], a 2-yr Mayor 13 years prior [0.5], a 4-yr State Senator 9 years prior [1.1], a 3-yr Lieutenant Governor 6 years prior [1.9], a 2-yr Governor 4 years prior [8.5], a 2.5-yr Vice President 1.5 years prior [42.6], and a 1-yr incumbent President [100], giving Coolidge a score of 154.7, or 155 to the nearest integer (and I will refer to integers going forward).

You will notice that despite Coolidge's storied career, only his experience as a President, Vice President, and Governor had a large effect on his score. This is generally true across the board, and makes the decisions about weighting minor roles and exactly how to temporally discount them somewhat academic⁸. So, although you are welcome to have strong views about how those elements work, in practice they will mostly not make much difference to the analysis.

What counts as a high score? This depends on whether you include Presidential and Vice Presidential experience or not, as they naturally dominate so heavily over all other experience, since they are highly weighted and are always recent, or even current. The highest score with President-level experience included is for George Washington in 1792, who had 1,251 points, 1,250 points of it from his 12.5 years as leader of the nation. The highest score for a contested election is for Franklin Roosevelt in 1944, 1,162 points, 1,150 of them from his 11.5 years as President. The highest scores excluding President-level experience are for Kerry [81 in 2004], Lewis Cass [73 in 1848], and W Clinton [72 in 1992]⁹ on the Democratic side¹⁰, and for Dole [115 in 1996], McCain [90 in 2008], and Rufus King [70 in 1816]. H Clinton's score was 24, in line with the post-Washington average of 22.¹¹

A final bit of housekeeping: there are two ways of deciding who "won" a Presidential election: who received the most votes; and who won in the Electoral College. The latter is what ultimately decides victory, but accounting for the former can be useful when modelling Presidential victories. I have two versions of the results variable, one that only tracks who won the election, -1 vs 1, and one that has a third 0 code for elections where the Electoral College victor lost the popular vote, which is what I will refer to, sometimes as a "draw", unless specified otherwise.

The analysis

My approach to the analysis took two parallel tracks. Firstly, what is the effect of President-level experience? Secondly, what is the effect of non-President-level experience? They are separate because they do not go in the same direction, which is what had made me examine the data in the first place. So we can divide our data into major groups based on these splits.¹²

President-level vs President-level with less non-Presidential experience: W1 D0 L3
President-level vs non-President-level with less non-Presidential experience: W8 D0 L2
Non-President-level vs President-level with less non-Presidential experience: W6 D2 L14
Non-President-level vs non-President-level with less non-Presidential experience: W6 D2 L10

These results are very striking. Unless one is a President-level candidate facing a non-President-level candidate, non-Presidential experience is not only not helpful, it is an outright liability. Candidates in the other three categories had results against their less-experienced opponents of W13 D4 L27, a worse than 1:2 win ratio. Yet the experienced President-level candidates won easily against non-President-level candidates, at a 4:1 ratio.

So the rule for winning Presidential elections is two-fold: either already have President-level experience, or have less non-Presidential experience.

Extensions

This could have been done with a multiple regression to determine what weights should apply to the job categories. Not only was that outside the scope of what I wanted to do, I think it runs the risk of over-fitting. Nonetheless, if anyone creates such a model, let me know! There are also other extensions that could be done even within the framework of the existing model, and I will likewise be interested to hear about those if anyone tries them out.

These results are essentially correlations (though I have chosen not to do a direct correlation as there are too few data points and the model is not sufficiently precise), and do not tell us about causation. It would be interesting to investigate why Americans tend to vote for candidates with less political experience. It would also be interesting to look into the effects of this phenomenon. Do outsider Presidents really behave differently in office to insiders?

We could use the model to predict the 2020 election. It suggests that Trump is the favourite to beat Joe Biden, as the former has vastly less non-Presidential experience and they are both President-level candidates. However, although a 1:2 win ratio is striking, it's hardly enough to make predictions about individual results, especially out of series. The model is also not enough to say for sure which Democratic candidates would have been best placed to beat Trump. There are lots of factors that go into the non-selection of a prospective candidate that cannot be covered only by examining selected candidates. We cannot say that a lunatic like Marianne Richardson would have been a winning candidate because she had no relevant experience! (Although maybe…) We can at most hypothesise that Biden was an inferior choice to a hypothetical similarly popular candidate with less Washington experience. But perhaps no such candidate existed. Nonetheless, although it would be foolish to treat the model as holy writ, it would also be foolish for parties to continue to treat large amounts of experience as an unalloyed good, and instead give more consideration to acceptable outsiders.

sum θoətiz abaʊt prezidenʃəl inekspiəriiyəns

hii bii oftən θinkid, espeʃəlii baiy iliit figəriz in politiks, ðə miidiiyər, and akədiimiiyə, ðat hai levəliz ov praiyə pəlitikəl ekspiəriiyəns biiy impʊətənt foər a kandideit foər y.s. prezidənt. ðis biiy a riizənəbəl θiərii. hii wʊd meik sens if ekspiəriiyəns in a kandideit foə publik ofis biiyid valyʊʊwəð muc az ekspiəriiyəns in a kandideit foə biiyin a doktər oər an enjiniə biiyid valyʊʊwəð: um wʊd jenərəlii not prifuə tə risiiv brein suəjərii from an enθyʊʊziiyastik aməcə promisiŋ tə swiip intə ðə hospitəl and drein ðə swomp ov medikəl prəfeʃənəliz.

in ðə tʊʊ-θaʊn-dein-soh elekʃən, supʊətəriz ov hilərii klintən prəkleimid ðat hii biiy a kandideit wið unyʊʊʒʊʊwəlii hai levəliz ov ekspiəriiyəns az paət ov diis kleimiz ðat hii biiyid a gʊd kandideit hʊʊ ʃʊd bii supʊətəð, ðat bii, linkiŋ ekspiəriiyəns tə kandideit kwolitiiy. aʊt-goʊwiŋ prezidənt barak oʊbaəmər iivən kleimid, in fakt, mii θink not ðeər evə biiyiv sumum soʊ kwolifaiyəð tə hoʊld ðis ofis. elən fitzpatrik, a prəfesər ov histəriiy at ðə yʊʊnivuəsitiiy ov nyʊʊ hampʃə, difendid h. klintən in ðə n.y.t. bai saitin misiz klintən-iis kəriər in publik suəvis, unraivələð baiy enii fiimeil prezidenʃəl kandideit — and oəlmoʊst enii meil prezidənt.

az biiyid waidlii pointəð-aʊt at ðə taim, ðiiz kleimiz bii obviiyəslii fols: h. klintən-iis nuə yiəriz az a senətər and fʊ yiəriz az a kabinət membə bii historikəlii not espeʃəlii noʊtəbəl. wiðin ðə prisiidiŋ fyʊʊ dekeidiz aloʊn, ðə deməkratiz nomineitivid kandideitiz az ekspiəriiyənsəð az jon keriiy, a tʊʊ-dein-yiə senətər and tʊʊ-yiə lʊʊtenənt guvənər, and bil klintən, a dein-tʊʊ-yiə guvənər; and ðə ripublikəniz nomineitivid jon məkein, a tʊʊ-dein-tʊʊ yiə senətər and fʊ-yiə kongresmən, and bob doʊl, a tʊʊ-dein-nuə-yiə senətər and nuə-yiə kongresmən. iivən hiis mein praiməriiy əpoʊnənt, buənii sandərz, biiyid a dein-yiə senətə, dein-so-yiə kongresmən, and nuə-yiə meə. h. klintən-iis ekspiəriiyəns biiyid oʊnlii riəlii noʊtəbəl in kontrast tə hiis evencʊʊwəl əpoʊnənt, donəld trump, hʊʊ havid noʊw ekspiəriiyəns ov publik ofis at oəl.

but ðə trʊʊθ oər uðəwaiz ov ðə kleimiz bii les intərestiŋ ðan ðə fakt ðat dii biiyid meikəð, in suc prominənt intəvenʃəniz and bai suc signifikənt advəkeitiz, az if dii biiyid pozitiv riizəniz tə bak h. klintən. ðiiz kleimiz bii not yʊʊniik tə deməkratiz aiðə: wen hii biiyid Republicans' turn to nominate a much more ekspiəriiyənsəð candidate, McCain vs Obama, they trumpeted his ekspiəriiyəns and scorned Obama's inekspiəriiyəns. Part of this, in both cases, was political opportunism, yes, but political opportunists only make arguments that they believe will have resonance.

The question is also important for what it implies about other, wider questions. In the narrow H Clinton case, if she lost despite her ekspiəriiyəns advantage working in her favour, that should prompt Bayesians to adjust our priors in favour of the feminist hypothesis that female candidates are the victims of patriarchal oppression. Conversely, if ekspiəriiyənsis not generally a factor, or is even a hindrance, we should adjust our priors in favour of the realist hypothesis that there is no patriarchal oppression affecting female candidates. Other wider questions include the relationship between American voters and the establishment, and how parties should choose their candidates in future to maximise their chances of winning.

So is candidate ekspiəriiyəns a good predictor of success in US Presidential elections? My original analysis says otherwise. As a rule, the winning candidate will be either the candidate who has already been President (or Vice-President), or the candidate with the least ekspiəriiyəns, not the most. The effect is not merely neutral: it favours the less-ekspiəriiyənsəð candidate.

The data points

The ekspiəriiyəns data is inevitably more subjective and less clear-cut. I have included data for ten distinct categories of political office, and weighted them by duration of office, recency of last holding that office, and a weight for relevance to the position of President of the United States. It is important to note that these are only my subjective choices, and other weights, especially for the relevance score, would produce different results. Anyone who wishes to perform their own dissenting analysis should feel at liberty to do so.

The ten distinct categories of political office included are: President⁵; Vice-President; Cabinet member (including Ambassador); Governor; Lieutenant Governor; US Senator; US Congressman; State Senator; State Congressman; and Mayor. I have not included ekspiəriiyəns in the military, business, or policy work.⁶ Any analysis has to have some limit.

The scores for these groups, I decided on by setting the Presidency itself to 100 and a US Congressman to 1, and pegging all other scores to that. All scores are my own intuitive sense, informed by a small number of considerations about how the role relates to the Presidency. First of all, the existence of a role is something in itself, as everything can be spun into a tale of ekspiəriiyəns, so minor roles are weighted up from where they would otherwise be. Second, and this accounts for the largest variance, is the number of people who hold the office at any given time. For example, there are roughly four times more US Congressmen than US Senators, so US Senators have four times the weight of US Congressmen. The third consideration is the tension between the wider stage vs executive power. Finally, junior offices have been counted at roughly a fifth of the senior office.

So, the scores are as follows:

President	100
Vice-President	20
Cabinet	2
Governor	6
Lieutenant Governor	1
US Senator	4
US Congressman	1
State Senator	0.5
State Congressman	0.1
Mayor	0.5

All of these are included for completeness. However, as we will soon see, most of them don't make much difference to the results in practice.

The ekspiəriiyəns score for each candidate is then: Σ[ w * d * √(((r/4)+1)^-1) ] across the ten categories. As an example, upon standing for the Presidency, Calvin Coolidge had been a 2-yr State Congressman 16 years prior [0.1], a 2-yr Mayor 13 years prior [0.5], a 4-yr State Senator 9 years prior [1.1], a 3-yr Lieutenant Governor 6 years prior [1.9], a 2-yr Governor 4 years prior [8.5], a 2.5-yr Vice President 1.5 years prior [42.6], and a 1-yr incumbent President [100], giving Coolidge a score of 154.7, or 155 to the nearest integer (and I will refer to integers going forward).

You will notice that despite Coolidge's storied career, only his ekspiəriiyəns as a President, Vice President, and Governor had a large effect on his score. This is generally true across the board, and makes the decisions about weighting minor roles and exactly how to temporally discount them somewhat academic⁸. So, although you are welcome to have strong views about how those elements work, in practice they will mostly not make much difference to the analysis.

What counts as a high score? This depends on whether you include Presidential and Vice Presidential ekspiəriiyəns or not, as they naturally dominate so heavily over all other ekspiəriiyəns, since they are highly weighted and are always recent, or even current. The highest score with President-level ekspiəriiyəns included is for George Washington in 1792, who had 1,251 points, 1,250 points of it from his 12.5 years as leader of the nation. The highest score for a contested election is for Franklin Roosevelt in 1944, 1,162 points, 1,150 of them from his 11.5 years as President. The highest scores excluding President-level ekspiəriiyəns are for Kerry [81 in 2004], Lewis Cass [73 in 1848], and W Clinton [72 in 1992]⁹ on the Democratic side¹⁰, and for Dole [115 in 1996], McCain [90 in 2008], and Rufus King [70 in 1816]. H Clinton's score was 24, in line with the post-Washington average of 22.¹¹

The analysis

My approach to the analysis took two parallel tracks. Firstly, what is the effect of President-level ekspiəriiyəns? Secondly, what is the effect of non-President-level ekspiəriiyəns? They are separate because they do not go in the same direction, which is what had made me examine the data in the first place. So we can divide our data into major groups based on these splits.¹²

President-level vs President-level with less non-Presidential ekspiəriiyəns: W1 D0 L3
President-level vs non-President-level with less non-Presidential ekspiəriiyəns: W8 D0 L2
Non-President-level vs President-level with less non-Presidential ekspiəriiyəns: W6 D2 L14
Non-President-level vs non-President-level with less non-Presidential ekspiəriiyəns: W6 D2 L10

These results are very striking. Unless one is a President-level candidate facing a non-President-level candidate, non-Presidential ekspiəriiyəns is not only not helpful, it is an outright liability. Candidates in the other three categories had results against their less-ekspiəriiyənsəð opponents of W13 D4 L27, a worse than 1:2 win ratio. Yet the ekspiəriiyənsəð President-level candidates won easily against non-President-level candidates, at a 4:1 ratio.

So the rule for winning Presidential elections is two-fold: either already have President-level ekspiəriiyəns, or have less non-Presidential ekspiəriiyəns.

Extensions

These results are essentially correlations (though I have chosen not to do a direct correlation as there are too few data points and the model is not sufficiently precise), and do not tell us about causation. It would be interesting to investigate why Americans tend to vote for candidates with less political ekspiəriiyəns. It would also be interesting to look into the effects of this phenomenon. Do outsider Presidents really behave differently in office to insiders?

We could use the model to predict the 2020 election. It suggests that Trump is the favourite to beat Joe Biden, as the former has vastly less non-Presidential ekspiəriiyəns and they are both President-level candidates. However, although a 1:2 win ratio is striking, it's hardly enough to make predictions about individual results, especially out of series. The model is also not enough to say for sure which Democratic candidates would have been best placed to beat Trump. There are lots of factors that go into the non-selection of a prospective candidate that cannot be covered only by examining selected candidates. We cannot say that a lunatic like Marianne Richardson would have been a winning candidate because she had no relevant ekspiəriiyəns! (Although maybe…) We can at most hypothesise that Biden was an inferior choice to a hypothetical similarly popular candidate with less Washington ekspiəriiyəns. But perhaps no such candidate existed. Nonetheless, although it would be foolish to treat the model as holy writ, it would also be foolish for parties to continue to treat large amounts of ekspiəriiyəns as an unalloyed good, and instead give more consideration to acceptable outsiders.