Security Crises and Executive Approval
Chapter 5 of the book manuscript Rally or Reprimand? APSA Annual Meeting · September 2026
Ryan Carlin · Timothy Hellwig · Will Horne · Gregory Love · Cecilia Martínez-Gallardo · Matthew Singer
Three Security Crises
September 11, 2001 Rally.
Madrid, March 2004 Reprimand.
Türkiye, 1991 & 1995 Reprimand.
What separates them?
Photos: Ground Zero, Sept. 14, 2001; Atocha station, March 2004.
9/11: the largest rally on record, and unusually durable — elevated for thirteen months (Hetherington & Nelson 2003).
Madrid: the government’s handling of the attacks fueled the opposition in the election held three days later.
Türkiye: PKK casualties were blamed on the incumbent and produced gains for right-wing opposition parties (Kibris 2011).
Same class of event; rally, reprimand, or nothing. That variation is the chapter’s target.
What we know: three schools, no complete framework
Mueller (1970): rallies follow events that are “specific, dramatic, and sharply focused.”
Affective Threat, patriotism, out-group hostility
Why salient attacks end in reprimand
Rationalist Performance judgment; casualties
Casualties weighed inconsistently
Opinion leadership Opposition restraint → no criticism in the news (Brody 1991)
Direct measures of elite conflict
Most incidents produce modest bumps or none; the second generation of research turned to explaining the variation in rally size.
Affective: Huddy et al. 2005; Merolla & Zechmeister 2009 — explains why rallies concentrate in wars and attacks, not why some salient attacks end in reprimand.
Rationalist: Gartner & Segura 1998; Gelpi, Feaver & Reifler 2006 — supplies a reprimand mechanism, but the public weighs casualties inconsistently, and approval often moves before results can be judged.
Opinion leadership: Brody 1991 — the right actors, but so far without direct measures of elite conflict spanning countries and crisis types. That measurement gap is what we fill.
The salience–conflict framework (Chapter 2)
1. Salience is a necessary condition.
2. Given salience, elite conflict decides the direction. rally .unchanged or reprimand .
3. Conflict also influences duration.
Conflict is measured two-sided : government ↔︎ opposition.
Severity and salience correlate only imperfectly: in our account it is salience that matters.
Direction: when government and opposition speak in one voice, the balance of messages reaching the public favors the executive and approval rises. When elites clash over the response or a war’s motivation, the public divides.
Duration: approval reverts as credible criticism returns to the news (Brody 1991; Groeling & Baum 2008). Reprimands work symmetrically.
Two-sided: criticism of the executive by political actors AND criticism of the opposition by the executive — not opposition criticism alone.
Which crises suppress elite conflict?
Blame turns on plausible exogeneity (Hellwig et al. forthcoming).
Attack directed from abroad
Wars, incl. wars of choice
Domestic, recurring violence
Restraint; rally most likely
Debated from day one
Blame easy; conflict baseline high
Expect: rallies after international attacks and interstate wars; reprimand, or nothing, where violence is homegrown.
Terrorism is coded by the origin of its perpetrators, not the location of the attack.
Perfect exogeneity is rarely achieved; what matters is where an event falls on a spectrum of plausible exogeneity.
A surprise attack from abroad is the paradigm case — hard to blame on the incumbent, at least at first.
Wars sit in the middle: even when governments argue necessity (the US in Iraq), taking a country to war invites debate over motivation and conduct from the first day.
Where violence recurs — cartel violence in Mexico, the Basque conflict, the PKK — the public expects prevention, blame comes easily, and a new incident produces a small row rehashing old positions.
Do discrete security events move approval?
EAD: monthly approval, 35 countries · CED: validated security events (20+ deaths; GTD, ICB, UCDP)
959
discrete security events
~3%
of country-months see an onset
15%
of country-months under an ongoing crisis
The Crisis Events Database validates every event against news or archival sources; it distinguishes interstate conflicts from (largely domestic) terrorism.
Onsets are rare and concentrated: fourteen of thirty-five countries record none; Mexico, the US, Russia, the UK, Colombia, and Turkiye account for over four-fifths.
Ongoing: Turkiye spent 71% of its sampled months under crisis (mostly terrorism), the US 69% (almost entirely war).
Event results: onsets rally — but only interstate ones
Onset lifts approval: +0.45*
All of it interstate: +1.18*** — terrorism: n.s.
Severity (fatalities): little , interstate only
Rarity: bigger rallies after long calm
Ongoing crises: no average effect
Model details: CRE GLS with lagged approval, growth, inflation, and election-cycle controls; SEs clustered by country (Table 5.3, Models 1-2; severity and frequency in Table 5.5).
Severity: fatalities recoded to in-sample quartiles; interstate severity +0.38, terrorism n.s. Frequency: log months since the last crisis; interstate +0.37 *.
The punchline for the transition: a crisis dummy observes neither salience nor elite conflict — a war that has faded from the front pages enters these models exactly like one commanding every headline. That is why we measure the coverage itself.
Measure what the public receives
News-based measures, by crisis type:
Salience = crisis articles mentioning the executive ÷ all articlesElite conflict = share of coverage paragraphs with government–opposition conflict (months with 10+ paragraphs) Searched sample = 100+ archived articles that month; empty searches are true zeros (24 countries) War = estimated within war coverage months
Salience: relevant articles divided by the total the country’s papers published that month.
The 10-paragraph screen: a conflict share computed from a handful of paragraphs is mostly sampling noise; results are identical for any screen between 10 and 30 (backup slide).
Searched sample: the archives at times hold only partial collections, so we require at least one hundred archived articles; a searched month with no coverage is a genuine zero, not a gap.
War is episodic, not a standing threat — a month in which France is not at war is not a month of low war salience. Estimating within war coverage months follows the rally literature (Mueller, Brody, Baker & Oneal). The searched-sample war estimate is a backup slide.
Median conflict runs 11-15% of coverage paragraphs across the three types.
War: salience rallies while elites hold their fire
Elite conflict
−1.86*
+2.75*
−1.59*
Salience
+9.25*
+105.1* +9.60
Conflict × Salience
−24.1
−756.2* −17.7
N (countries)
4,961 (24)
643 (4)
4,961 (24)
Table 5.6. CRE GLS, lagged DV, clustered SEs. * p ≤ .05. War: AUS, FRA, UK, US.
The rally dissapears at ~15% conflict
Three results: (1) security crises move approval only when substantially covered; (2) elite conflict is associated with lower approval in the pooled models, even at low salience; (3) for wars, salient coverage helps only where conflict in that coverage is low.
The figure: marginal effect of war salience across elite conflict. The histogram shows the distribution’s mass sits almost exactly where the effect turns negative — around 14% of paragraphs. Chapter 6 puts the health threshold at 28%, so the war rally window is about half as wide.
Controls: lagged approval, growth, inflation, election cycle. The war column rests on the four countries with both a war episode and media data — worth saying out loud.
Terrorism: Exogeniety Matters
Elite conflict
−2.11* −0.92
Salience
+7.84
+103.3*
Conflict × Salience
+13.9
−635.7*
N (countries)
4,961 (24)
4,961 (24)
Table 5.7. Same specification, searched sample. * p ≤ .05.
International = war’s pattern. Domestic = reprimand only.
International estimates lean heavily on the US.
Pooled terrorism showed no salience effect; splitting by perpetrator origin is where the theory bites.
International terrorism reproduces the full war pattern: salient coverage lifts approval where elites hold their fire and hurts it when conflict is high.
Domestic terrorism: salience does nothing, directly or conditionally; elite conflict is simply associated with lower approval. Publics do not rally around executives managing homegrown violence.
Caveat to name out loud: the international estimates lean heavily on the United States, where the largest episodes concentrate.
Case Study: the United States
Salience
thin
2.1% — highest in our data
Elite conflict
8% 4%
Approval
no move
47.6 → 63.5
Conflict returns — 9% by December — and the rally decays.
August 1998: after al-Qaeda bombed the embassies in Kenya and Tanzania, coverage was thin and unusually conflictual — congressional accusations that Clinton’s retaliatory strikes were timed to distract from scandal. Approval did not move.
September 2001 inverted both conditions: the highest international-terrorism salience in our data, with conflict at 4% — the consensus of a stunned political class. Approval rose from 47.6 in August to 63.5 by November.
Then conflict climbed month by month, reaching 9% of coverage by December; salience receded; the long decline in the figure began.
High salience with low conflict produced the rally; the return of conflict began its end. Media series in the figure cover measured months only (Figure 5.11).
Conclusions
Salience is necessary; elite conflict decides the direction. Events: interstate onsets rally; severity adds little; rarity matters .
The gradient holds: international terrorism and war rally (conditionally); domestic violence only reprimands.
The US around 9/11: highlights the mechanism, month by month.
Security crises can rally, reprimand, or leave approval unchanged — and the outcome tracks salience and elite conflict.
Discrete events lift approval on average, but the lift is all interstate; terrorist events produce no general rally.
The media measures show why: war coverage rallies only below roughly 14% conflict; international terrorism reproduces the pattern; domestic terrorism, recurrent and entangled with standing policy disputes, produces no rally at all.
This chapter in the book
Ch. 2 theory → Ch. 3 experimental results → Ch. 4 measures → Ch. 5 security → Ch. 6 health → Ch. 7 natural disaster → Ch. 8 economic
Security: where rally conditions are strongest — outside threat, informational advantage, wary opposition.
Health (Ch. 6): same logic, rally window twice as wide (28% vs. 14%).
One framework, one set of measures, applied across crisis domains.
Chapter 2 develops the salience-conflict framework; Chapter 4 builds the news classifiers used here.
This chapter tests the framework where rallies were first documented and where the conditions for one are most often met: threats that arrive from outside, an executive with an informational advantage, an opposition wary of politicizing the dead.
Chapter 6 finds the same conditional logic in public health crises with a rally window twice as wide. The remaining chapters ask how the logic fares where those conditions hold less firmly.
Backup — The conflict screen doesn’t drive the war result
Conflict × Salience
−508.5
−550.0
−756.2* −766.8*
−791.7*
−878.3*
(SE)
(356.7)
(350.1)
(321.8)
(333.3)
(328.7)
(316.9)
N
651
650
643
637
634
627
Appendix Table 5A.3: the featured war model with conflict screened at each column’s minimum paragraph count. A share computed from a handful of paragraphs is mostly sampling noise; the estimate is stable for any screen from ten paragraphs up and strengthens as reliability rises — the errors-in-variables signature.
Backup — War on the searched sample
Elite conflict
−0.71*
+0.43
Salience
+123.7* +47.5
Conflict × Salience
−804.7
−428.0
N (countries)
4,985 (24)
4,961 (24)
Appendix Table 5A.2: war re-estimated on the 24-country searched sample, months without war coverage as zeros. Salient war coverage still lifts approval contemporaneously; the interaction attenuates, as expected when wartime months are pooled with months carrying no measurable war signal.
