Elias Aboujaoude (b. 1971) published Virtually You: The Dangerous Powers of the E-Personality in 2011, when blogging still mattered, Twitter still imposed a 140-character limit, and the psychological consequences of living online were new enough to seem strange. The architecture of online life, he argued, encourages a distinctive e-personality: more assertive, less restrained, more impulsive, more certain of what it knows. The danger he emphasized was permeability. Habits learned online migrate into the rest of the personality.
He was right about several of the routes. He worried that blogging frees a writer from the patient construction of an argument, and that Twitter compresses serious matters into a form hostile to sophisticated exchange. He tied impulsivity to a decline in what he called reflection, meaning the collection and evaluation of information before reaching a conclusion.
The model has a gap. If the border between the online and offline self is porous, traffic should run in both directions. Habits acquired through professional scholarship may migrate down into the e-personality while impulsivity, certainty and aggression migrate up out of it.
Steven L. Taylor and James Joyner (b. 1965) offer a test of that proposition. Their group blog, Outside the Beltway, began in January 2003. Joyner published thirteen posts on the first day. He later described early OTB as links, excerpts and often snarky commentary. Taylor, who had been blogging at PoliBlog, joined the OTB front page in November 2006. The site covers most of the lifespan of the political blogosphere, written by two men who built substantial professional identities outside it. Joyner became a professor of security studies at Marine Corps University’s Command and Staff College, working on national security, military affairs and civil-military relations. Taylor became a professor of political science and then dean at Troy University, publishing on parties, elections, electoral systems and comparative democratic institutions. He retired in July 2024. Joyner’s retrospective OTB at 20 supplies much of the site’s own account of its history.
That combination permits a harder question than whether two men got angrier over twenty years. I coded claims and tested four competing explanations against each other.
The first is Aboujaoude’s medium effect. Lower-friction, more interactive forms should produce more rhetorical heat and weaker epistemic regulation. The second is compensatory regulation. Experienced writers may keep making heated claims while attaching evidence, qualification, counterargument, concession or correction to them. The third is selection adaptation. Rather than learning to write every instantaneous reaction more carefully, an experienced blogger may learn not to publish many of those reactions at all. The fourth is arousal override. Political events of unusual emotional intensity may temporarily overwhelm whatever regulatory habits a writer has acquired.
The four do not receive equal support, and the two authors do not behave the same way.
For Joyner, the evidence favors a medium effect and gives partial support to compensatory regulation. For Taylor, the developmental story fails at the start. His blog writing already shows strong epistemic regulation in the earliest comparable material, especially when he writes inside political science. His long-term adaptation is selection. He publishes less of the old rapid-fire blogging and more developed analysis. Taylor also supplies the best evidence for arousal override, in the hours and days after Donald Trump’s (b. 1946) victory in November 2024.
The theoretical result may outlast the empirical one. Aboujaoude got the permeability right and the direction incomplete. At Outside the Beltway, the professional personality keeps invading the e-personality.
The best evidence in the study comes from one night in 2012.
During the Republican National Convention on August 29, Joyner watched Paul Ryan’s (b. 1970) acceptance speech and commented live on Twitter. He then embedded the stream in an OTB post called RNC Night 2: This Is More Like It. The opening line works as a manipulation check: he promises more coherent thoughts in the morning.
The Twitter stream is fast, funny and evaluative. It delivers instant judgments of speakers, jokes and sarcastic reactions. When Ryan blamed Barack Obama (b. 1961) for the closing of the General Motors plant in Janesville, Wisconsin, Joyner’s immediate response amounted to incredulity. The medium rewarded the reaction ahead of the research.
The next morning the same writer returned to the same claim about the same speech in Paul Ryan’s GM Plant: Fact Checking the Fact Checks. The structure changes. Joyner walks through the plant’s chronology, Obama’s statement and the competing fact checks. He separates the technically defensible portion of Ryan’s sentence from the implication Ryan wanted the audience to draw. He concludes that Obama cannot be blamed for the closure, and he considers stronger economic criticisms Republicans might have made instead.
Almost everything that ordinarily worries a longitudinal comparison is held close to constant here. Same writer. Same subject. Same politics. The available historical information barely changed overnight. We are not comparing young Joyner with middle-aged Joyner, Republican Joyner with post-Trump Joyner, or amateur blogger Joyner with professor Joyner. What changes is medium and about twelve hours.
The substantive conclusion survives the transition nearly intact. What changes is its epistemic architecture.
Had the Aboujaoude effect meant that Twitter changes a man’s political opinions, this pair would be weak evidence. His deeper concern ran to reflection, inhibition and the compression of complicated matters into quick reactions. The Joyner pair displays that compression and then its reversal. Twitter gets the conclusion first. The blog post gets the evidence, the distinctions and the rival interpretations.
Joyner has described the process. In a 2021 essay for War on the Rocks, Twitter Doesn’t Have to be the End of the Conversation, he contrasts the snappy, instantaneous exchange encouraged by social media with longer venues that require substantive analysis and editorial guidance. He grants that Twitter rewards sarcasm, cynicism and putdowns, and he argues that its rapid feedback can improve later thought. A year later, writing in Defense One, he made the admission personal. The platform creates incentives to perform for an online audience, and he says he fell for the trap.
One caution belongs here. Joyner knew the tweets would be archived on his own site the same night. That awareness cuts against the disinhibition reading, and it makes the observed gap between the two versions a conservative estimate rather than an inflated one. The line about more coherent thoughts in the morning also shows that the writer understood the limits of the medium while using it. Recognition of the affordance preceded the compensation.
This is support for the medium hypothesis, and it stops well short of medium determinism. Joyner’s 2021 argument explains why. Low-friction media generate useful criticism and expose an idea to people who improve it. His War on the Rocks essay began as a Twitter conversation. What shifts is the probability distribution. Faster media make an unregulated reaction easier to publish. Slower and edited forms create more chances for a brake.
What counts as a brake? I treated a claim as a heat event when it carried a strong evaluative judgment, negative person-focus, pejorative characterization, ridicule or argumentative sarcasm. Categorical certainty was coded separately. A confident empirical proposition is not by definition heated, and coding it that way would classify ordinary factual assertion as rhetorical disinhibition.
I then asked whether the heat event carried a local brake. Evidence counted. So did a qualification narrowing the claim, a serious counterargument, a concession to the other side, or a correction or update. Proximity governed. The brake had to appear in the claim, the same sentence or bullet, or the immediately adjacent sentence in a way tied to the claim.
That rule determines how to read Taylor’s November 6, 2024 post The List.
Taylor introduces nineteen predictions as his concerns and says the list came off the top of his head. Those are real epistemic signals. They tell a reader not to mistake what follows for a forecast with calibrated probabilities.
They are global signals. They cannot be sprinkled backward across nineteen separate predictions so that each one becomes locally qualified. If Taylor says a Trump administration will do X, the preface supplies no evidence for X. If he calls a possible appointee utterly unqualified, the introductory modesty does not regulate the evaluation. A prediction that contains its own “may,” its own evidentiary reference or its own competing possibility is a different case, and I coded it accordingly.
The distinction protects the measure. Without it, long-form writers get a free advantage: one disclaimer at the top of a thousand-word post converts every categorical assertion below it into a qualified claim, and the coding then measures politeness of framing rather than regulation of reasoning.
A second problem cannot be solved by any coding rule, and readers should hold it against the results throughout. A brake is observable behavior. Its motive is not. A qualification may express real uncertainty, or it may be reputational armor bolted onto a conclusion the writer never doubted. A political scientist who has spent thirty years being asked “how do you know that” acquires the reflex of anticipating the question. That reflex looks identical on the page whether it regulates the thought or merely dresses it.
Some brakes cost the author something, and those carry more weight. A concession that weakens the writer’s own conclusion, a correction that admits an earlier error, an update that abandons a preferred interpretation: these are expensive in a way that a hedging adverb is not. Where the argument below turns on costly brakes rather than cheap ones, I say so.
Two trajectories
The frozen claim-level corpus produced 71 fully segmented claims from nine posts whose source text could be recovered in full. Fifty belong to Joyner, twenty-one to Taylor. I coded those claims from scratch under the rules above. One coder produced every number in this essay, so read the percentages as descriptions of a small sample rather than as measurements.
Joyner’s numbers move in one direction. In the 2007 material I coded eight heat events, six of them locally braked. In 2013, seven heat events and six brakes. In 2020, six heat events and six brakes. In the strict 2025 coding, two and two. The series runs 75, 86, 100 and 100 percent.
The tidiness should raise suspicion, and the final cell carries a warning. Joyner’s 2025 post on Joe Biden’s (b. 1942) cancer diagnosis ends with the judgment that the public is entitled to such information about a president. Code that ordinary normative assertion as a strong evaluative heat event and the 2025 result becomes two of three, or 67 percent. One decision moves the cell by 33 points. See the post and judge for yourself.
The larger problem is arithmetic. Twenty heat events spread across four periods cannot sustain a trend test. Comparing Joyner’s earliest cell, six of eight, against his two most recent cells combined, eight of eight, falls far short of conventional significance in any test appropriate to numbers that small. Anyone who wants to reject the compensatory-regulation finding on the numbers alone can do so, and I would not argue. The percentages are a summary of what I read. The textual pattern behind them carries the argument.
That pattern is consistent. Consider Joyner’s 2013 post on Apple’s tax arrangements. The rhetoric is lively. He calls the arrangement brilliant and clever. He then asks why Congress left the loophole open for decades, notes that multinational corporations have exploited such differences for a long time, concedes that a lower corporate rate makes intuitive sense, questions whether it would solve the problem, and labels his own consumption-tax proposal his decidedly non-expert sense of things. The post runs hot without pretending that the heat constitutes expertise.
The 2020 material is stronger. In WSJ Chastises Trump’s Slander, Joyner uses nonacademic language about Trump’s behavior, and the pejorative conclusion sits on top of a long evidentiary structure: the accusation under criticism, the factual basis for rejecting it, the Wall Street Journal editorial’s reasoning, and the institutional problem of applying Twitter’s ordinary rules to a sitting president. High temperature, high evidentiary density.
Another 2020 claim shows the expensive kind of brake. Having first read Justice Department decisions on congressional insider-trading investigations through the politicization of the department under Trump and William Barr (b. 1950), Joyner reverses himself on reflection. He gives his reasons and separates the still-open investigation of Richard Burr (b. 1955) from the closed ones. That is visible updating, and it costs him the more dramatic version of his own story.
The mature Joyner is sometimes harsher than his younger self. The change is that the harsh assertion increasingly travels with its own audit trail.
His professional baseline predicts that habit. Joyner’s 1996 International Studies Quarterly article with John Oneal and Brad Lian, Are the American People “Pretty Prudent”?, tests a prominent theory of public support for the use of force against 38 major American uses of force in foreign-policy crises from 1950 through 1988. The authors do not find the periodization they went looking for. They report the failure and identify a different relationship in the data.
His civil-military work behaves the same way. In Trump’s Generals: A Natural Experiment in Civil-Military Relations, published in Strategic Studies Quarterly, he begins with two alarming and opposed predictions about retired generals serving in the Trump administration and concludes that neither worked well. The generals produced no militarization of policy and no reliable restraint of the president. The paper ends on an unresolved question about long-term damage instead of converting uncertainty into a dramatic finding.
The edited professional essays follow suit. This is Not a Civil-Military Crisis, written for Defense One, opens by granting the core concern of the argument he opposes, that retired senior officers have acquired an outsized political role, before disputing the stronger characterization. His essay on Kenneth Waltz (1924-2013), first published in The National Interest, separates admiration for Waltz’s theoretical achievement from acceptance of the policy conclusions structural realists drew from it.
The possibility raised by all this runs opposite to the obvious one. Blogging did not train Joyner to write like an academic. Over twenty years, habits recognizable from academic and edited professional work became more visible inside the blog voice.
Taylor keeps the study honest by refusing to fit.
His early OTB writing is already too regulated to leave room for progressive improvement.
The clearest early case in the frozen sample is his May 2007 writing on Hugo Chávez (1954-2013) and the Venezuelan opposition broadcaster RCTV. Taylor makes strong judgments. He reads the episode as part of an authoritarian consolidation and connects control of television to the maintenance of political power. Institutional reasoning accompanies each judgment: RCTV’s position in Venezuelan broadcasting, the station’s conduct during the 2002 coup crisis, and an argument about why that history fails to justify the 2007 action.
In my coding, Taylor produced five heat events in the 2007 material, all five locally braked. In the complete 2020 sample I coded five heat events and four brakes. That is 100 percent followed by 80 percent. Those cells are far too small to establish a decline, and one recode would erase the difference. They are enough to reject the attractive version of the compensatory hypothesis, in which Taylor gradually learns to regulate hot claims. No low starting point exists from which such an improvement could occur.
His scholarship explains why. A Different Democracy, written with Matthew Shugart, Arend Lijphart (b. 1936) and Bernard Grofman (b. 1944), compares American political institutions with those of thirty other democracies instead of treating familiar American arrangements as the natural baseline. His Oxford Handbook chapter on the American electoral system explains the durability of the two-party system through plurality elections, single-seat districts, primaries and the Electoral College. His 2025 Protect Democracy paper, Trapped in a Two-Party System, extends the same institutional analysis and proposes open-list proportional representation as one route out. The argument carries a normative implication. Its explanatory machinery is structural.
That machinery keeps showing up on the blog. His 2020 discussion of the Oregon Republican Senate primary could have been another culture-war post about a QAnon-linked candidate. Taylor moves outward from the candidate to the incentives of a noncompetitive statewide race, weak candidate recruitment, party labels as information shortcuts, the American primary system and the effects of electoral-system design. The person is the news. The institutions are the explanation. The site’s own summary describes the post as more about structural conditions than about the Republican Party.
Taylor shows that professional regulation can walk in the door on the first day.
The second long-term change is easier to see. A writer can adapt to an impulsive environment in two ways. He can improve the impulse after expressing it. Or he can stop converting every impulse into a publication.
Early OTB was prolific by current standards. Joyner published thirteen items on January 31, 2003. Reviewing that archive ten years later in Joyner Unfair to Krugman, he wrote that the vast majority of his February 2003 output consisted of InstaPundit-style short posts with little or no analysis, unlike the longer pieces normal by 2013.
May 2007 still looks like that environment. The archive shows Taylor publishing several times on some days and Joyner producing a run of items on unrelated subjects within a few hours. Collecting fifteen eligible substantive posts, working backward from the end of the month, took six calendar days for Taylor and seven for Joyner.
By 2013 the same fixed target required 23 days for Taylor and 21 for Joyner. In 2020 it required 21 days for Taylor and eight for Joyner. In 2025 it required eleven for Taylor and eighteen for Joyner.
No smooth secular decline appears. Composition changed more than frequency. The site changed with the writers. Joyner says OTB’s May 2010 redesign moved it from a conventional reverse-chronological weblog toward a magazine format. A February 2018 redesign made the site mobile friendly and removed the Quick Picks section. After Quick Picks disappeared, Joyner says he all but stopped writing the InstaPundit-style micro-post and seldom published unless he had at least a few paragraphs to say. The 2020 rebound in his output has a separate cause. Doug Mataconis had stopped writing for the site, and Joyner says he and Taylor picked up the pace to compensate.
This is selection adaptation. The adapting unit is partly the writer and partly the institution the writers built.
The distinction improves Aboujaoude’s model. An experienced Internet user can alter the architecture around publication. He can abandon the format most likely to turn a passing reaction into a permanent item. A website can delete the box that invited one-line links. A writer can adopt a personal rule that a thought earns a post only when it supports several paragraphs.
The Internet creates affordances for impulsivity. Experienced users remove affordances.
The same behavior extends to the comment section. On November 7, 2024, the day after “The List,” Taylor published Don’t be a Troll, announcing that he was tired of commenters who insult rather than argue and setting out three rules: no personal attacks on post authors, actual engagement with the post, and show your work if you want a debate. The next day Joyner seconded him in And Another Thing!, asking commenters to drop the nicknames for the president-elect on the grounds that they signal unseriousness and lower the tenor of the conversation without advancing any argument.
Two men who had just written some of their hottest prose in twenty years turned within forty-eight hours to rebuilding the rules of the space they write in. Aboujaoude’s model has a place for the disinhibited reaction and a place for its migration into ordinary life. It has no obvious place for the moderator who responds to his own disinhibition by rewriting the house rules.
The claim about rising post length has to stay modest. The frozen corpus did not preserve reliable author-only word counts separating Joyner’s and Taylor’s prose from the often extensive material they quote. Reporting per-1,000-word heat rates or a median-length series on that basis would be false precision. The evidence establishes a categorical shift away from micro-posting far more securely than any continuous increase in author-written length.
In 2003, by Joyner’s account, most posts were short weblog entries. By 2023 he seldom bothered to post without a few paragraphs to contribute. The disappearance of a unit of production is a behavioral change.
If Taylor is the hard case for increasing regulation, he is a good case for arousal override.
“The List” appeared the day after Trump’s victory. Its nineteen predictions range across the civil service, the Justice Department, Ukraine, Israel, tariffs, appointments, NATO, public health, the Federal Reserve, immigration enforcement, the press and the international monetary order. Taylor is writing after an outcome he considers dangerous.
The post shows why global and local qualification have to be separated. “These are my concerns” brakes the document. “Off the top of my head” brakes the document. Many of the predictions then arrive as bare propositions. The format discourages development. Little evidence attaches to individual items, competing routes to the same outcome go untreated, and confidence varies little across nineteen predictions of unequal plausibility.
Some of that is defensible on Taylor’s own terms, since he says the purpose is to put down a marker for later discussion. Putting down a marker and locally regulating each prediction are different acts.
The comment threads of those days run hotter than the posts. Taylor reports deleting a comment and says he is tired of one commenter’s sneering. Minutes after, weighing another commenter’s severe geopolitical forecast, he calls the possibility frighteningly real and reaches for the ending of Planet of the Apes.
This is the strongest Taylor material for Aboujaoude. Immediate, interactive, emotionally charged. The political scientist’s habit of decomposing a proposition into institutional parts has not vanished, and for a few exchanges it recedes.
One confounder gets its own sentence. Taylor retired from Troy University in July 2024, four months before he wrote “The List.” The daily scaffolding of professional life, teaching, colleagues, administration, ended shortly before the episode this study reads as arousal override. Nothing in the data separates the two explanations. Anyone testing the reciprocal-migration model on a larger sample should treat retirement, career change and institutional exit as predictors rather than biography.
Even here the result refuses to simplify. The next day Taylor published What Explains a Trump Win?. He opens by correcting the emerging landslide narrative with the popular-vote margin, says a firm explanation must wait for more complete data, describes his account as initial thoughts, and works through party identification, incumbency, inflation and retrospective voting. He remains morally alarmed. The alarm does not prevent analytical qualification.
Four days after that, in “One Weird Trick” Election Takes, he is arguing against monocausal explanations of Kamala Harris’s (b. 1964) defeat. He goes after the pundit’s temptation to assume that some favored messaging adjustment would have flipped the result, and he insists on multifactor explanations that no campaign fully controls. Professional training reasserts itself inside the same high-arousal week.
Arousal override survives in a narrow form. High arousal can weaken local regulation for a period measured in hours or days. It does not erase the habit from the writer.
The January 2026 reassessment of “The List” strengthens that reading. Taylor returns to the predictions and audits them one by one. Some were borne out. Some require qualification. On Schedule F he records that Trump did not implement the proposal he predicted, then separates that failed prediction from other civil-service actions he considers damaging. The rhetoric stays severe. The updating is the expensive kind.
The comments show the other face of interactivity. A reader catches Taylor writing FEC where he means FCC. Taylor’s reply is immediate: correct, and thanks for the note. The section that amplified heat in November 2024 works in January 2026 as miniature post-publication review.
Interactivity can disinhibit the author or correct him. Which one depends on the week.
The medium effect performs best. Joyner’s August 2012 convention pair comes close to a within-person experiment. The live Twitter version is faster, more sarcastic, less developed. The next-morning blog version is sourced, qualified and argumentative. His later reflections identify the same affordances from the inside. No twenty-year change in ideology or personality explains a difference that appears overnight.
Compensatory regulation is supported for Joyner and fails as a general developmental law. The proportion of his heat events receiving local brakes rises from 75 percent in 2007 to 86 percent in 2013 to 100 percent in 2020, with 2025 at either 100 or 67 percent depending on one borderline call. Those cells are too small to carry the claim by themselves, and I have said so. Taylor does not reproduce the pattern. He begins at 100 percent and sits near 80 percent in 2020.
Selection adaptation receives substantial support in one form and none in another. Both writers slow sharply from the 2007 burst rate by 2013, then rebound irregularly, and Joyner’s 2020 rebound has a staffing explanation. The threshold for publication is what changed decisively. The 2010 move toward a magazine format and the 2018 removal of Quick Picks accompany the disappearance of the tiny link-and-comment post, and Joyner says he now rarely publishes without several paragraphs to contribute.
Arousal override receives the narrowest support. Taylor on November 6, 2024 is the case. Global qualification survives while local regulation weakens and comment interaction raises the temperature. The effect is temporary and uneven; his November 7 and November 11 posts are heavily regulated in the same emotional aftermath. Joyner’s convention pair points the same way. Change the medium from live Twitter to next-day blogging and regulation returns while the event holds still.
One prediction failed, and the failure carries more weight than the confirmations. The professional-habits account implies that each man should be best regulated inside his field and worst outside it. The first half holds. Taylor’s institutional writing is where academic habit is easiest to see in blog prose, and Joyner’s national-security work is most regulated when it passes through Defense One, War on the Rocks or Strategic Studies Quarterly, which is partly a genre effect, since an edited essay supplies external brakes as well as internal ones.
Joyner’s Apple tax post sits far outside his field, and his response is to surround his judgment with reporting and expert discussion and then label his own proposal nonexpert. His 2025 Biden piece leans on the medical expertise of Ezekiel Emanuel (b. 1957), gives reasons for treating Emanuel as credible, and marks his own inferences about what Biden knew with “may well” and “more likely.” Taylor’s 2025 post on fabricated citations in the MAHA report is hot and sarcastic on a subject far from electoral systems, and the factual case comes almost entirely through NOTUS reporting; when he speculates that artificial intelligence generated the fake references, he calls it his first assumption rather than a finding, and he later concedes a point before turning it against its proponent.
Distance from expertise may trigger compensation among experienced writers: more conspicuous source dependence, stronger self-labeling of inference. The matched claims are too few to estimate that interaction. They are enough to make the simple alternative, that people get sloppier whenever they leave their field, untenable here.
Combining the two men into an OTB average would bury the central result. Joyner fits a developmental account, entering from the first-wave blogosphere and acquiring, over two decades, a combination of rhetorical freedom and epistemic restraint, along with an explicit consciousness of what the medium does to him. Taylor arrives with the professional habits already installed, which creates the ceiling, and adapts instead by changing what he publishes, with episodic breakdown under high arousal.
The pilot intuition survives only after being split in two. Joyner supplies the compensatory-regulation evidence. Taylor supplies the early-carryover, selection and arousal evidence. Both supply medium-effect evidence. None of it supports the proposition that either man became more extreme because he spent two decades online.
The simple version of Virtually You predicts contamination. We enter a low-friction environment, become more impulsive, grandiose, hostile and certain, and those traits colonize the offline person. Aboujaoude describes the border as porous and worries about online behavior bleeding into ordinary life.
Nothing here disproves that. The Joyner Twitter comparison is good evidence for it. Taylor’s election-week comments are another instance. The old OTB archive preserves a medium in which a professional scholar could move from an observation to public prose in minutes.
OTB also shows a reciprocal process the theory needs.
People do not arrive online empty. Taylor brings comparative politics with him. The e-personality complains about Trump; the political scientist keeps asking which electoral structures, party incentives and institutional rules produced the event. Joyner brings international relations and security studies. The blogger renders a sharp judgment; the researcher arrives a paragraph later asking what the source says, whether an alternative interpretation survives, and whether the first reaction needs revision.
Professional institutions leave residue. Peer review teaches that claims need support. Editors teach that someone else will test the sentence. Scholarship teaches that contrary evidence is an object of analysis. Long exposure to those practices builds regulation that operates when no editor is present.
Joyner’s 2021 War on the Rocks essay nearly describes the loop. A Twitter conversation generates an idea. The idea becomes a blog post or a longer essay. Editors and reviewers strengthen the later version. That version returns to social media, where criticism starts again. He points to a hierarchy of forms. Information travels up and down it.
Taylor supplies a parallel case. His 2025 Protect Democracy paper is a polished professional version of arguments OTB readers had been reading for years. The day after publication he promoted it on OTB, and his recoverable Bluesky post does little more than point readers to the paper. The fast medium serves as the distribution network for the slow one.
The recoverable Bluesky evidence is too thin to support a symmetrical platform comparison. Joyner’s profile identifies the professional subjects one would expect; Taylor’s identifies comparative democratic institutions and OTB. Individual Joyner posts were not recoverable in sufficient number to build a responsible longitudinal series. Absence from search indexing is not absence of behavior, and treating it as evidence would repeat the error this study is designed to measure.
So the e-personality can infect professional life with speed, certainty, performance and aggression. Professional life can infect the e-personality with evidence, qualification, counterargument, revision and the learned suspicion that a first reaction may be wrong. Sometimes the decisive act of self-regulation happens before either personality writes a word, when the writer decides the thought does not yet deserve a post.
They learned the Internet. The Internet learned them.
The longitudinal frame used four comparable OTB periods, anchored in May 2007, May 2013, May 2020 and May 2025. I did not manufacture a 2003-06 Taylor series. Taylor joined the front page in late 2006 and the old PoliBlog record is incomplete. Joyner’s 2003-06 material establishes the early-blogging ecology and cannot support a symmetrical two-author estimate.
The frozen sampling frame held fifteen eligible substantive public-affairs posts per author in each anchor year, 120 in all. Eligibility required substantive author-written analysis rather than a bare link, a housekeeping item or an open thread. Working backward from May 31 gave a reproducible selection rule and prevented later outcome coding from determining which posts entered the sample.
Claims were the substantive unit. A claim is the smallest proposition that can independently be true, false, plausible, implausible, justified or unjustified. One sentence often contains two. Quoted material was not coded as the author’s claim unless he endorsed it, and rhetorical questions were coded only when the intended proposition was unambiguous. I coded categorical certainty, negative person-focus, pejorative characterization, ridicule or argumentative sarcasm, strong evaluative judgment, evidence, qualification, counterargument, concession, and correction or updating. Heat comprised the four evaluative and person-directed features. A local brake required evidence, qualification, counterargument, concession or correction inside the claim or in immediately adjacent material governing it.
“The List” served as the calibration case. Its introductory language was coded as global epistemic framing and was not propagated into the nineteen predictions. The rule makes the test harder on long-form writing and prevents a disclaimer from laundering unsupported claims into regulated ones. A secondary context-adjusted analysis, in which a governing global qualification regulates claims within its stated scope, remains available to anyone who wants to run the numbers the other way; it raises Taylor’s 2024 figures and does not change the comparison between authors.
Nine of twelve frozen segmentation-check posts were fully recoverable. One was partially recoverable and two were blocked. I did not reconstruct missing claims from titles, snippets or expectation. The numerical audit therefore rests on the 71 claims in the nine recovered posts. Supplemental cases, including the 2012 Twitter and blog pair and the 2024 Taylor sequence, were used for within-person leverage and hypothesis testing and were kept out of the frozen percentages.
The original plan called for rhetorical-feature rates per 1,000 author-written words, which remains the right denominator for comparing raw frequencies across posts of radically different length. Reliable author-only word counts were not available across the frozen corpus, in large part because OTB posts carry long quoted blocks. I have not manufactured those rates. The principal statistic, local brakes divided by heat events, sidesteps the mechanical problem that longer posts contain more opportunities for both heat and regulation. Longitudinal estimates of median author-written post length remain unfinished measurement rather than a result.
The selection test rests on three measures of unequal strength. The calendar span needed to accumulate fifteen eligible posts gives a reproducible frequency indicator. The archive documents the disappearance of the micro-post. Joyner’s retrospective supplies direct author testimony that the threshold changed. What cannot be claimed is a continuous estimate such as a percentage rise in median length.
Expertise distance was classified as core, adjacent or distant, assigned separately from rhetorical coding so the same judgment did not drive both variables. Taylor’s core covers parties, elections, representation, comparative politics, Latin American politics and democratic institutions. Joyner’s covers national security, international relations, military affairs, civil-military relations and professional military education.
The arousal test stays modest. The preregistered definition treats claims within seven days of a federal general election as event-proximate, which keeps the predictor from being defined by the emotional language it is supposed to predict. The 2024 Taylor material is a revealing qualitative case, and the complete claim corpus is too small to estimate a stable author-by-arousal coefficient. I treat arousal override as supported by episodes rather than as a population parameter.
Several falsification tests carry more weight than the raw percentages. The medium hypothesis would weaken if Joyner’s Twitter and next-day OTB versions were equally regulated; they are not. A universal compensatory-regulation hypothesis predicts a Taylor rise from early to late periods; the early ceiling contradicts it. A simple selection hypothesis predicts monotonic decline in publication rate; the rebounds contradict it. A deterministic arousal hypothesis predicts poorly regulated Taylor writing across the whole post-election period; November 7 and November 11 contradict it.
The academic baseline is not a generic category of offline writing. For claims about elections and democratic institutions, Taylor’s A Different Democracy, his Oxford Handbook chapter on the American electoral system and his Protect Democracy paper supply the relevant professional habits. For national security and military questions, Joyner’s International Studies Quarterly article with Oneal and Lian, his Strategic Studies Quarterly piece on Trump’s generals and his edited civil-military essays supply the corresponding baseline. Matching subject matter this way separates a loss of professional regulation from an ordinary difference in topic.
The largest limitation is reliability. I am one coder. The blind-coder packet is built to permit independent replication, and no independent human has performed it, so no inter-coder reliability statistic exists to report. A real test requires another person, blind to these conclusions, to segment and code the designated material and then compute agreement and a statistic such as Krippendorff’s alpha. The sensitivity of Joyner’s 2025 cell to one borderline strong-evaluation decision shows the cost of its absence. A second coder is the single improvement that would do most for this study.
A related limitation is conceptual. The coding sees behavior and infers nothing about motive. A brake that signals professional caution to an audience and a brake that regulates the writer’s own reasoning are indistinguishable on the page. Costly brakes, meaning concessions that weaken the author’s conclusion and corrections that admit error, are better evidence than hedging, and both authors produce them. That is the most I can claim.
The social-media record is asymmetric. Joyner’s 2012 Twitter stream survives because OTB captured it. Taylor’s recoverable Bluesky material runs to professional self-promotion and links to longer work. Older material is inconsistently indexed or inaccessible. I use social media where the underlying content is recoverable and do not read missing search results as zero heat, zero posting or zero regulation.
Finally, OTB is not a neutral container around two independently evolving authors. It became a magazine-style publication in 2010 and was redesigned again in 2018. Those changes are confounders if the question is whether aging alone changed the two men. They are the answer if the question is how experienced Internet writers adapt. Taylor and Joyner changed their site in part because their conception of what deserved publication had changed. The environment acted on them. They also redesigned the environment their own e-personalities work in.
That is where the study modifies Virtually You. Aboujaoude was right that the online self does not stay sealed behind the screen, and he describes online and offline life as engaged in constant back-and-forth exchange. The evidence from Taylor and Joyner suggests taking the phrase more literally than the book does.
