Intelligence for human security: measuring outcomes quantitatively

ABSTRACT This article examines whether increased intelligence capacity improves global security, a key assumption in intelligence theory. Using the partial least squares structural equation modelling method, the research statistically analyzes data from the U.S. International Intelligence Behaviour dataset and Global Terrorism Database. Grounded in intelligence studies and international relations theory, the study integrates a constructivist human security framework. Surprisingly, the results show a significant correlation between increased intelligence capacity and the degree of terrorism, suggesting intelligence may undermine rather than enhance human security. This finding challenges traditional assumptions, though it must be viewed cautiously due to potential endogeneity.


Introduction
In 2022, Amy Davidson Sorkin of The New Yorker asked, 'Has the CIA done more harm than good' since its inception in the late 1940s?Her approach to this question contrasts the narratives of Central Intelligence Agency (CIA) successes written by agency insiders with the scandalous past plaguing the agency, which 'has had a "defining failure" for every decade of its existence -sometimes more than one'. 1 She notes the difficulties investigators face in thoroughly studying the CIA, owing primarily to the veil of secrecy that has historically surrounded U.S. intelligence activities.Her conclusion, then, is a call for further inquiry into the CIA, suggesting that, 'if the C.I.A. keeps falling down all the time, something must be tragically amiss in the agency's structure or culture, or both'. 2 Security scholars have been equally skeptical of intelligence's direct value to American national security. 3Harry Howe Ransom (2000) considered this among other major questions for intelligence studies to investigate in the 21 st century.Indeed, we need look no further than the infamous U-2 flight of 27 October 1962, which was sent to collect measurements and signals intelligence (MASINT) in the arctic, for an illustration of how intelligence operations can lead to disaster.The flight which ultimately crossed into Soviet air space, helped precipitate the tensions of 'Black Saturday', when the United States and the Soviet Union came to the brink of nuclear war. 4 Khrushchev himself told Kennedy that the U-2 could have been easily mistaken for a nuclear bomber. 5Recognizing the stakes of intelligence operations, and building on prior skepticism, the present study asks, How does intelligence capacity affect security outcomes?In other words, does more intelligence mean more security?
An optimistic reading of the literature leads to a central assumption of intelligence theory -that more intelligence does indeed lead to more security.Nevertheless, answering this question has been complicated for intelligence scholars, given the dearth of available data on intelligence agencies and their operations.Data available for study has traditionally come from four sources, (1) memoirs, (2) whistle-blowers, (3) scholarly analyses, and (4) Government reports. 6Each of these sources has CONTACT Steven Stottlemyre stottlemyre@southwales.ac.uk limitations that branch throughout the intelligence studies literature, resulting in a very thin scientific foundation. 7To help thicken it, this study statistically analyses the impact of intelligence capacity on global levels of terrorism using a panel dataset including data from the U.S. International Intelligence Behaviour (USIIB) and Glob Terrorism Database (GTD) datasets (see Table C1 for descriptives)..The partial least squares structural equation modelling (PLS-SEM) method is chosen to examine the net impact of increased intelligence capacity on the degree of terrorism worldwide.The literature suggests that U.S. intelligence cooperation during the examined period (2000-2006) which is captured by USIIB -was increasingly focused on counterterrorism, thus its effect should be most pronounced in the field of counterterrorism.At the same time, GTD data shows a relative decline in U.S. terrorism fatalities abroad.The last five years of the 20 th century show a persistent terrorist threat against Americans overseas, with a marked decline in the years following 9/11-when Afghanistan and Iraq are removed from the data (See Figure 1). 8he literature does not suggest that a positive relationship between cooperation and outcomes exists, and this study does not examine the implications of cooperation as a unique function separate from other intelligence operations.Rather, this paper views cooperative events as captured in USIIB as indicative of a collection of overall increases in capacity (e.g., collection, analysis, action, etc.), rather than a simple variable, 'intelligence cooperation'.Given the structure of the model presented here, which relies on latent variables, each single-factor latent variable (including intelligence capacity) is left for future alteration, expansion, and analysis.USIIB presents the only publicly available information that are interpretable with quantitative methods as measuring net increases in intelligence capacity, but the model presented here does not presuppose that USIIB is, or should be, the only measurement of a state's intelligence capacity.It is simply the only such quantitative data available.Although it could be theoretically combined with other measures, such as intelligence community funding or personnel distribution, no other compatible datasets exist.
Situated within the interdisciplinary field of intelligence studies, this study draws on the principles of international relations theory to scrutinize the role of intelligence in shaping global security outcomes.By adopting a constructivist, human security framework, it emphasizes the significance of broader social contexts in which intelligence operates, and the criticality of protecting individuals from a wide range of threats.To do so, this study examines a core assumption of intelligence studies (i.e., that intelligence is necessary to security), while exploring the auxiliary hypotheses that protect them. 9revious studies leveraging USIIB indicate a positive relationship between six latent variables and the likelihood that the United States will engage in intelligence cooperation with a given state: terrorism, military cooperation, regime type, cultural factors, economic factors, and international ties. 10To support this study, similar data are transformed to include a lagged factor for investigating the lagged effects of U.S. intelligence intervention, which are assumed to have a net additive effect on local intelligence capabilities.Given the likelihood of reverse causality between the independent and dependent variables, and the issue of multicollinearity among examined factors, PLS-SEM is chosen as the primary analysis method.PLS-SEM is specifically designed to deal with these issues, in addition to issues of data distribution. 11To enhance the generalizability of the analysis, regime type is examined as a moderating variable.The comparative intelligence literature strongly suggests the presence of a correlation between degree of democracy, intelligence activities, and security outcomes.
Unexpectedly, three of the four examined models indicate that an increase in intelligence capacity is significantly correlated with the degree of terrorism, supporting the perspective that intelligence may lessen, rather than enhance, security in the international context.Additionally, regime type has a negligible moderating effect between intelligence capacity and the degree of terrorism.Overall, these results suggest the additive value of intelligence cooperation to security is negligible at best, and that such cooperation has similar effects for security, regardless of the cooperating state's regime type.These results should be interpreted with caution, given the possibility of endogeneity in the dataset, despite efforts to diminish its effects in the model.Nevertheless, they have significant implications for the development of intelligence studies, which has oftentimes assumed intelligence a critical component of security.

The methodological gap in intelligence studies
This study stands at the intersection of a nascent trend in intelligence studies emphasizing the implementation of quantitative research methodologies.Although the discipline has predominantly relied on qualitative approaches, leaving a conspicuous gap in quantitative research, the tide has been slowly turning.Traditional sources in intelligence studies, largely grounded in historical, legal, and political science domains, have not typically facilitated quantitative analysis.This trend was evidenced by a 2013 survey conducted by Loch Johnson and Allison Shelton, which stressed the paucity of quantitative data, and highlighted mixed opinions on the introduction of quantitative methods among leading intelligence studies scholars.
However, it is crucial to recognize that this effort is not unprecedented, as there has been a recent expansion in the number of quantitative intelligence studies.One example is the work of Levin  (2023), which quantitatively assesses the impact of exposure on the efficacy of covert operations, focusing on the 2020 U.S. elections as a case study.Similarly, Roberts (2023) employs quantitative methods to explore the dynamics between state preferences, alternative strategies, and the use of covert action by the United States during a significant portion of the Cold War era.While the research questions these studies address diverge from the one at hand, their approach demonstrates the potential for quantitative research in the field of intelligence studies.The incorporation of statistical analysis in these studies signals a burgeoning shift towards quantitative methods across the discipline.The present study aims to join and further this emerging trend by applying a scientific, systematic approach to test a core tenet of intelligence studies -that intelligence is integral to national security.Leveraging existing datasets (USIIB and GTD) in an innovative way, this study statistically probes that assumption, thereby contributing to the growth of quantitative intelligence studies.
Studies on the success and failure of intelligence (and thus intelligence capacity) suffer from the problem of selection on the dependent variable, which 'entails a high probability of getting the wrong answer'. 12Robert Jervis has spoken directly to this problem: Improvement is not likely without learning.But the IC has not had a robust program of postmortems, and those that have been done have generally been of cases of dramatic failure.[…]Causal inferences drawn from failures select on the dependent variable and may detect procedures and ways of thinking that characterize accurate as well as inaccurate estimates.Even if we found that certain factors were present in all the cases of failure, we would not be on firm ground in providing explanations and prescriptions unless we could also establish that those factors were absent in cases of intelligence success.Oxygen is not a cause of intelligence failure despite being present in all such cases.Doing this broader evaluation is not easy and may not yield useful recommendations, and the lessons may be hard to act on.But without careful and sustained self-study, few improvements are likely. 13r Richard K. Betts (2007), measuring intelligence is a simple cost-benefit analysis: 'If the blood and treasure saved because setbacks were averted by good intelligence exceed the blood and treasure spent to get the intelligence, the system pays for itself'. 14The persistent growth of intelligence spending and intelligence studies suggests an overwhelming confidence that the United States can answer in the affirmative.Yet others have argued that policy makers largely ignore this growing apparatus in making national-level policy decisions.

Intelligence and Human security
The central argument of this article is rooted in two schools of thought: social constructivism, and human security.The social constructivist perspective, often associated with scholars like Alexander Wendt, stresses the importance of ideational factors like norms, ideas, and identities in shaping political behaviors and outcomes.This perspective contrasts with realist and liberal theories of international relations, which generally prioritize material capabilities and institutions respectively.An emphasis on the social constructions of reality provides a lens through which we can understand how ideas and beliefs impact state behavior, leading to a varied and complex international political landscape.By casting international security as a construct, the human security paradigm (a concept developed by the United Nations Development Programme) urges the placing of individual security at the center of state security concerns, emphasizing the protection of people from a broad range of threats, including poverty, disease, environmental degradation, and armed conflict: The concept of security has for too long been interpreted narrowly: as security of territory from external aggression, or as protection of national interests in foreign policy or as global security from the threat of a nuclear holocaust.It has been related more to nation-states than to people.[…]Forgotten were the legitimate concerns of ordinary people.[…]For many of them, security symbolized protection from the threat of disease, hunger, unemployment, crime, social conflict, political repression, and environmental hazard. 15 consolidating human security as a school of thought, Newman (2001) defines it as a 'normative, ethical movement' aimed at moving the referent of international security from the sovereignty and territorial integrity of states toward issues of more immediate concern to the wellbeing of individuals and communities. 16He argues that 'the emergence of the [human security] … reflects the impact of values and norms on international relations'. 17hus, the values that govern individual behavior and feelings (e.g., empathy) can impact upon international relations, in contradiction to structural and natural theories.
There is disagreement among human security theorists on the types of issues that should be considered relevant to individual security. 18Some 19 see the concept in more general terms (e.g., 'freedom from want'), aligned with the UNDP perspective, while others 20 focus specifically on 'freedom from fear'.Krause (2008) argues that narrowing the term to simply refer to individual physical security concerns is theoretically consistent with traditional views of the liberal state.In essence, articulating security policy in terms of upholding the 'freedom from fear' allows for the easy employment of existing state security mechanisms to support human security.

Theoretical framework
Intelligence supports security, but there are competing narratives in the community that find individual public officers as serving either the state, or the public.In security organizations, personnel are centrally concerned with providing security -not just for the ruling regime, but more generally.That intelligence officers serve the state is simply one narrative expression of how intelligence behaves.Certainly, many scholars -most of those cited here -see intelligence as a type of policy support; in the United States, the entire intelligence apparatus can be understood as collectively providing a service to the President through their individual missions.This state servant narrative, however, competes with a separate, agent-focused narrative where, at least in the immediate term, intelligence officers prioritize public security -protecting individuals from fear -regardless of political context.In the U.S. intelligence community, the conflict between these narratives plays out between source handlers and their managers over when and how to cut sources loose, between analysts and policymakers over when to change security posture, and between numerous other groups working to improve security.Based on the latter narrative, an important metric for determining intelligence effectiveness is the extent to which its operations result in less death and destruction -less fear.This study, then, asks whether more intelligence means more security, and uses the impact of security incidents on individuals as its dependent variable.Since intelligence officers are immediately focused on public security, this framework predicts that any increase in their capacity will result in a decrease of 'fear' among local individuals.

Independent Variable: Intelligence Capacity
The study of intelligence lacks a cogent structure, despite its more than eight decades of existence in the United States.The interdisciplinary field, which is presented in this context as a subfield of international relations, remains in the early stages of theoretical development.Several attempts to develop and collect theory have surfaced, 21 but none has drawn and significant group of followers.The failure of major theories to infiltrate the field, in my view, results from two related factors in intelligence studies, (1) that it is dominated by practitioners, and (2) the persistent debate over how to even define 'intelligence'.Regarding the former, practitioners (including the author) tend to rely on the conception of intelligence they learned through their profession, which is (of course) inclusive of their personal intelligence vocation.This has led most intelligence studies publications to include some mention of the definitional debate, alongside a purpose-built definition that includes one or a combination of typical 'intelligence' professions, including collection, analysis, production, counterintelligence, covert action, and policy.This study instead seeks to build upon the frame developed by the late Dr. Sherman Kent, often described as the 'father' of American intelligence analysis.Kent (1949) devised a tripartite definition allowing for intelligence to be described in terms of knowledge, activity (i.e., collection and analysis), or organization. 22The latent independent variable (intelligence capacity) is conceived, then, as a combination of how these attributes combine to contribute to security.
Very little literature exists on the quantitative measurement of intelligence.Such discussions in established intelligence studies journals focus on measuring the accuracy of intelligence assessments, 23 rather than the impact of intelligence on security outcomes.The graduate thesis of Christian Hippner (2009), submitted to Mercyhurst University, creates a method for measuring the size of worldwide intelligence spending.However, his data on the intelligence spending of individual states is largely calculated based on a formula he developed from the statistical comparison of 45 openly available intelligence budgets and associated GDPs; when compared to other, since revealed budgets, his estimates appear quite inaccurate.Musa Tuzuner's doctoral dissertation ( 2009), submitted to Kent State University, used machine coding to compile all observations of U.S. international intelligence cooperation by the Agence France-Presse (AFP) newswire for the period 2000-2006 into the USIIB dataset.He further used the data to examine the antecedent variables to U.S. intelligence cooperation behaviour, such as Terrorism, Military Cooperation, Regime Type, Cultural Characteristics, Economic Characteristics, and Ties to the International Community, finding that each had a somewhat predictive effect on whether the United States would engage in such cooperation. 24

Cooperation as capacity during the global war on terror
Formal American international intelligence cooperation (or liaison) began in 1948 with the UKUSA agreement that divided responsibility for signals intelligence (SIGINT) collection between the United States and four partners, Australia, Canada, New Zealand, and the United Kingdom -collectively referred to as 'Five Eyes'. 25Similar, but less expansive agreements have followed under the auspices of international organizations, like NATO, or through bilateral agreements.According to Jennifer Sims, the U.S. State Department's first Coordinator for Intelligence Resources and Planning, liaison relationships like these can (1) increase access to sources (2) lower intelligence costs and risks (3) enhance the timeliness of intelligence, and ( 4) foster joint operations.Liaison during 'the war on terror [was] broadly viewed as essential to protecting the U.S. homeland and the allied states who share western values that make them attractive targets for al Qa'ida, and the so-called Islamic State'. 26As of 2022, the U.S. Intelligence Community conducts international cooperation under the 'information sharing environment' established by the Intelligence Reform and Terrorism Prevention Act of 2004 (IRTPA). 27IRTPA provided the newly created Director of National Intelligence with the power to regulate intelligence cooperation -referred to in the law as liaison -on all issues.It also created the National Counterterrorism Center (NCTC) as a separate agency specifically focused on combatting international terrorism that has forged international liaison agreements with counterterrorism centers across the Five Eyes countries. 28At the same time, established agencies like the CIA and FBI significantly increased their foreign liaison relationships specifically in response to the ongoing threat from al-Qaida. 29efebvre (2003) finds that four factors impact effective intelligence cooperation, (1) threat perception, (2) the distribution of power, (3) human rights records, and ( 4) other legal issues (e.g., bureaucratic steps required to engage in cooperation), suggesting that the United States may be weary of increasing liaison with authoritarian countries.However, building on Tuzuner's dissertation, Aydinli and Tuzuner (2011) argue that the United States has been more likely to seek intelligence cooperation with non-democratic states during 2000-2009.Others similarly note the importance of security context to the formation of international intelligence cooperative agreements, as in the case of the short-lived TRIDENT agreement between Israel, Turkey, and pre-revolutionary Iran. 30The value to American intelligence can be significant, even from smaller intelligence services.Taillon (2002)  notes, '[O]n occasion, some smaller nations can have access to important human intelligence sources, and therefore these states can be attractive partners in intelligence-gathering activities abroad'. 31This type of relationship was likely important during the Global War on Terrorism, which largely addressed the issue of Islamist terrorism in remote regions of Africa, the Middle East, and Asia.

The data: intelligence capacity
The data required to conduct a comprehensive study on the value of intelligence to security undoubtedly exists, though it may be concealed from public view.Given the limited accessibility to these data for scholars, we must make do with the resources at our disposal.Although certain public data, such as metrics of intelligence agency personnel strength and budgets, can be used in addressing the research question, additional data collection falls outside the purview of this study. 32n this context, the USIIB dataset, which records the annual count of cooperative events between the U.S. intelligence community and foreign government entities per country-year (2000-2006), stands as the most comprehensive dataset available on intelligence organizations' behavior.As intelligence capacity is a single-factor latent variable in this study, it is necessary to again ask whether intelligence cooperation can be operationalized as intelligence capacity.The literature is replete with discussions of the positive impact of cooperation on capacity, and regularly highlights the emphasis on international intelligence cooperation in combating terrorism post-9/11.These points suggest that cooperation aptly fits within the definition of the Intelligence Capacity latent variable.
To further elucidate the relationship between cooperation and capacity, let's draw upon game theory's Stag Hunt.In this scenario, hunters must cooperate to successfully hunt a stag and ensure sustenance for all.Should even one member of the group fail to cooperate, perhaps opting to pursue an easier, yet less rewarding catch, the entire group risks going without the substantial reward of the stag.This scenario is analogous to intelligence capacity, where cooperation is integral to reaching substantial goals, such as preventing terrorist attacks.This dynamic, encapsulated by the post-9/11 intelligence cooperation recorded in the USIIB dataset, stresses the mutual benefits of international intelligence collaboration.From a constructivist standpoint, the outcomes of cooperation or defection, like the hunters' decision to collaborate or go solo, are themselves constructs shaped by perceptions, interpretations, and beliefs.Outcomes cannot be definitively known in advance, so states or agencies that choose to engage in international cooperation are likely motivated by an expectation (read: construct) of increased operational capacity and improved security.However, it's critical to clarify that a lack of cooperation does not directly translate to an increase in terrorism.Rather, non-cooperating states or agencies might simply forego potential benefits that could have reduced their level of threat below its current baseline.In this light, the construct of intelligence capacity as measured through cooperation in our dataset is nuanced and multifaceted, reflecting the complexities inherent in international relations and intelligence work.
To head off some of the expected criticism of using USIIB to measure potentially secret behavior, I will address the question of how it can accurately do so.First, the activities of intelligence agencies are not as invisible as we intelligence officers may sometimes assume.U.S. intelligence agencies have officers stationed across the globe.Some collect intelligence overtly, others less so, but most with the general knowledge of local authorities, media, and sometimes the populations.The clearest example of this is the presence of a CIA base operating in Benghazi following the first Libyan Civil War, which was well known to their neighbors, and even marked on the crowdsourced mapping site Wikimapia!Second, this is a low-resolution study that asks a very broad question.It relies on the statistical assumption that even if every event cannot be counted, the cumulative changes in event counts over time reflect real changes in behavior.
The USIIB project used the KEDS_Count software program and an archive of AFP news wire articles to generate weighted quarterly data of U.S. international intelligence cooperative events for 191 countries for each year between 2000 and 2006.(The number of countries was limited to those recognized by the United States during the studied period.)The quarterly events were totaled to calculate the yearly U.S. intelligence cooperative events, and then the number of U.S. international intelligence cooperative events for each country-year was totaled.These cooperative events range in significance from one agency appealing to another for cooperation to agencies engaging in material cooperation across state lines. 33USIC (i.e., the count of cooperative events between the United States and other states) is the only factor contributing to the intelligence capacity latent variable.It is lagged by one year here (t-1), as the effect of an increase in intelligence capacity on the degree of terrorism is expected to be delayed.Thus, the following hypotheses are tested:

Dependent variable: degree of terrorism
Although there is no internationally accepted definition of terrorism, it can be described as 'the deliberate use or threat of violence against civilians by a nonstate entity (individual or group) in pursuit of a political or religious goal'. 34Several factors have been considered in the literature as potential keys to combatting terrorism, particularly economic, political, fractionalization, and geographic factors.In analyzing these factors (using data from the World Market Research Center's Global Terrorism Index, which assesses the risk of terrorism), only support for the political and geographic arguments can be found.Studies on foreign aid and terrorism similarly find little evidence of an economic impact on terrorism, but find (using GTD data) that other types of aid (i.e., governance and civil society aid) have slowed the increase of terrorism in some states, suggesting that 'democracy aid' should be used as a counterterrorism tool. 35he literature on regime type and counterterrorism can be divided into two opposing camps, those who believe (1) democracies are more prone to terrorism, or (2) democracies are less prone to terrorism. 36Those in the first camp 37 generally agree that the relative liberty afforded to individuals in democratic societies creates a permissive environment for terrorism.Amichai (2022), however, is critical of earlier studies, arguing that terrorism data drawn from the late 20 th century are biased because they mostly consist of attacks by nationalist and leftist movements that no longer exist, instead arguing that 21 st century trends show more terrorist attacks in less democratic states (using GTD).She further criticized the dichotomous treatment of democratic status in previous studies, and instead opted for a six-category ordinal classification of states based on their degree of liberality, ranging from liberal democracy to closed autocracy. 38e data: degree of terrorism In accordance with H 1 and H 2 , the degree of terrorism latent variable is constructed in two different models, with Model 1 formed to capture total negative outcomes for the cooperating country, and Model 2 formed to capture negative outcomes for Americans abroad.All factors used were compiled from the GTD dataset.This is an improvement on the USIIB, which includes less comprehensive information on terrorism outcomes drawn from the now defunct Terrorism Knowledge Base.Model 1 of the degree of terrorism latent variable includes three factors, (1) the number of successful terrorist attacks per country-year (success), (2) the number of people killed in terrorist attacks per country-year (nkill), and (3) the number of people wounded in terrorist attacks per country year (nwound).On the other hand, Model 2 includes the factors successus, nkillus, and nwoundus, which limits the model to the number of Americans and American targets affected by successful terrorist attacks outside the United States per country-year.Other terrorism datasets could be suitably employed for this study, but only if they provided data for the years leading to, during, and following the examined period (2007-2020), such as the RAND Database of Worldwide Terrorism Incidents (RDWTI), 39 or the Political Instability Task Force Worldwide Atrocities Dataset. 40However, these datasets contain significantly fewer incidents, and are cited less in the literature.

Moderating variable: regime type
Intelligence cooperation can be limited by circumstance.Taking inspiration from the dyadic peace literature, Bock (2015) argues that Anglo-Soviet intelligence cooperation during World War II suffered from a lack of shared norms.He derives the 'depth of intelligence cooperation' based on three factors, (1) frequency of contact between liaison officers, (2) the volume of intelligence exchanged, and the (3) granularity of that intelligence. 41Overall, the relationship could be described in terms of realist theory, such that initial cooperation, which was based on need, did not continue to grow as power relationships shifted throughout the war, and ultimately ceased following the war. 42 Brown and Farrington (2017) directly critique Bock's approach, and are less moved by the Anglo-Soviet experience.They offer a different framework that predicts intelligence sharing based on three factors, 'a reason to exchange (information and friendship), a route for exchange (close personal ties), and a record of exchange (reliably meeting expectations)'. 43ccording to Allen Dulles (2006), the veteran spy and first civilian Director of Central Intelligence, 'most totalitarian countries have, in the course of time, developed not one but two intelligence services with quite distinct functions'. 44One service is concerned with the collection of military intelligence, while the other, typically referred to as a 'security' service, is the creation of those in power, rooted in efforts to stifle dissent. 45 Gill (2008) argues that the meaning and operation of intelligence varies as regimes differ 'in terms of the centrality or otherwise of security intelligence concerns'. 46He presents a spectrum of regimes along a spectrum with the 'counterintelligence state' on one extreme, and 'security states' on the other. 47In both extremes, intelligence is central to the state's functioning, but internally focused in the former, and outwardly in the latter.This roughly equates to the traditional democratic-authoritarian division of presented by Matei and Halladay (2019) in The Conduct of Intelligence in Democracies, based on the work of Linz and Stepan (1996).Intelligence transforms as democracy is consolidated (i.e., liberalized) -it is itself democratized.Democratic reform of intelligence requires (1) democratic control over intelligence, and (2) effective democratic management of intelligence.Democratized intelligence means that the civilian leadership has developed intelligence doctrine, intelligence institutions, and a multicomponent resourcing mechanism. 48atfield (2022) cautions against 'treating authoritarian intelligence as merely a degenerate form of "legitimate" Western practices' in his post-structural critique of previous attempts to quantify intelligence by regime type, and notes that authoritarian intelligence differs from democratic intelligence in that they focus on suppressing dissent and controlling information flows, and have varying strengths and weaknesses in the intelligence process.Andrew (2004)  used the experiences of Soviet Russia and Baathist Iraq to argue that the intelligence apparatus of one-party states behaves differently from their democratic counterparts in two distinct ways.First, authoritarian intelligence communities are 'central to the structure of the one-party state and to the systems of repression and social control which seek to suppress all challenges to its authority'.Second, they also act as mechanisms 'for reinforcing the regime's misconceptions of the outside world'.Thus, Andrew sees authoritarian, particularly one-party regime intelligence as more inward looking than their counterparts.
The literature suggests a difference in purpose between democratic and authoritarian states' intelligence services.Broadly speaking, this difference is described in terms of focus, where authoritarian intelligence is more internally focused and concerned about threats to the regime, and democratic intelligence more outwardly focused and concerned about more traditional security threats.It is expected that regime type will impact any security outcomes associated with changes to a state's capacity to conduct intelligence, thus the final hypothesis of this study is presented: H 3 : Democratic states will experience increased counterterrorism benefits associated with increased intelligence capacity compared to less democratic states.

The data
The regime type latent variable, constructed from a single factor (level of democracy), measures democracy using data extracted from 'Freedom in the World' reports by Freedom House.This factor encompasses aspects of freedom of political rights and civil rights, with countries scored on a continuous scale of 1 to 7. To enhance intuitiveness, the scale is inverted so that lower scores represent less free countries, and higher scores denote countries with greater freedom.In their research, Roberts and Tellez (2020) highlight the significant impact of the scores in the Freedom House's Freedom in the World report on the perception and treatment of nations.The duo terms this as the 'Scarlet Letter effect', where a country designated as 'Not Free' by Freedom House is subjected to increased verbal criticism from democracies, particularly in the aftermath of the report's annual release.
They also contend that the Freedom House scores can steer international discourse and diplomatic relations among countries.Interestingly, they separate the effect of the Freedom House's indicator from the underlying factors the indicator measures by exploiting a discontinuity in the assignment of a country's freedom status.This implies that countries on either side of a certain threshold are assigned different labels, despite having similar levels of political and civil liberties.The conclusions presented in this article indicate that democratic and authoritarian regimes may be more similar (in terms of intelligence operations) than previously believed, and no value judgement is being made about authoritarianism in this context.Nevertheless, the issue points to a potential weakness of this study -that it relies on event data.Since events are coded as positive or negative, it is possible that datasets like USIIB may capture some events from less 'free' countries as more negative than reality.

Control variables
Findings from the two previous studies using the USIIB dataset are central contributors to the selection of control variables.Using variables available in USIIB, Tuzuner (2009)(who also created the dataset) found statistically significant correlations between five categories of variables and the likelihood that a state would engage in intelligence cooperation with the United States in a given year.Three of these categories included some variables with no or negligible effects on the dependent variable.Thus, the remaining two categories, Military Cooperation and Economic indicators, are further examined here.Both categories include multiple variables.Military Cooperation is represented by two variables, U.S. Military Joint Operations (MJO), and U.S. Military Deployment (MDP).To calculate MJO, a comprehensive list of all U.S. military deployments was obtained using the CRS Report for Congress, which was then cross-referenced with official web pages of the United Nations, NATO, and the Multinational Force in Iraq to identify joint operations.Similarly, MDP was constructed by consulting annual reports on the Department of Defense Personnel and Procurement Statistic web site, which provided information on the total number of US military personnel deployed in different countries each year.
The Economic indicators category consists of GDP, level of trade with the United States (US_TRADE), and U.S. Foreign Aid (US_F_AID).The GDP per capita variable was created by collecting data from the International Monetary Fund's official webpage, where GDP per capita data were downloaded through World Economic and Financial Surveys.Missing data for 16 countries were supplemented with the CIA World Fact Book.However, an outlier was observed due to the high GDP per capita of Luxembourg.The U.S. trade variable was calculated using the 'Foreign Trade Statistics' data from the United States Census Bureau web site.Finally, the U.S. foreign aid variable was constituted using the 'Country Reports on Human Rights Practices' prepared annually by the Bureau of Democracy, Human Rights, and Labor and made available on the U.S. Department of State webpage.All control variables are included as single-factor latent variables.

Analytic method: partial least squares structural equation Modeling
Partial Least Squares Structural Equation Modeling (PLS-SEM) is a multifaceted statistical technique chosen for this study.This approach is advantageous for examining intricate relationships between multiple independent and dependent variables by transforming them into what are referred to as 'latent variables'.These latent variables can then be regressed to gain insights into their interconnectedness. 49One key concept this technique addresses is endogeneity, a condition where the predictor variables are correlated with the error term.In this study, for example, there is potential endogeneity if increases in terrorism lead to increases in intelligence cooperation, the central component of the intelligence capacity latent variable.PLS-SEM minimizes endogeneity by comparing latent variables (e.g., degree of terrorism) instead of directly analyzing variables themselves.
The PLS-SEM method also assists with handling statistical issues like high dimensionality (where we have many variables to analyze), small sample sizes, multicollinearity (when predictor variables are highly correlated), and non-normal data distribution (where the data does not follow the bell curve of normal distribution).By using latent variables and lagged variables, PLS-SEM allows the simultaneous estimation of both the measurement model (the relationships between observed factors and their latent variables) and the structural model (relationships among the latent variables).The primary objective of using PLS-SEM is to identify underlying causal relationships among latent variables and to test theoretical models.

Data preparation
When preparing the data, it is important to remember that PLS-SEM is robust against violations of normality and multicollinearity. 50By standardizing variables through z scale transformation, the technique enhances data distribution, mitigates the impact of outliers, and reduces the problems associated with multicollinearity.In the context of this study, data from GTD and USIIB datasets were compiled and prepared, removing zero values and insignificant observations.Each factor was tested for normal distribution using the Kolmogorov-Smirnov test (see Table C2).All showed significant deviation from normal distribution. 51Similarly, the collinearity statistics were checked to ensure that multicollinearity, a condition where predictor variables are highly correlated, does not bias the findings.Collinearity can inflate the variances of the parameter estimates and make the estimates very sensitive to minor changes in the model.(PLS-SEM is used to minimize these effects.)However, despite efforts, Breusch-Pagan and White tests revealed significant heteroskedasticity across the dataset.Heteroskedasticity is a situation where the variability of the error term, or 'noise' around the line of best fit, differs across the values of an independent variable.This issue was addressed using bootstrapping, a non-parametric resampling technique that offers robust estimation of standard errors of model parameters, even when classical statistical assumptions are violated.Finally, the concept of endogeneity was addressed using tests for data stationarity.The Augmented Dickey-Fuller (ADF) tests were conducted for all variables to ensure that they were not characterized by unit roots, which could potentially lead to misleading results in the presence of endogeneity.The test results confirmed the rejection of the null hypothesis of non-stationarity, implying that the factors are stationary, and the problem of endogeneity has been addressed.PLS-SEM is not without limitation.There has been some discussion in the literature about the usefulness of single factor variables in PLS-SEM.However, an examination of Cronbach's α for this set of models indicates that they can be expected to perform similar to models with multi-indicator latent variables. 52Thus, while there is some concern about single factor PLS-SEM models, the model examined here has been constructed such that this issue is minimized.Overall, this method of addressing normality, collinearity, heteroskedasticity, and endogeneity aims to enhance the readability and reliability of the study results.

Estimating model parameters
Parameters were estimated for each model to determine fit using PLS-SEM (see Table A).Construct reliability and validity were appropriate across all models, however model fit was low-to-moderate.Central to this problem is likely the inclusion of the regime type as a moderating latent variable; its removal improves fit markedly.One possible reason that regime type is affecting fit could be that, as Annex A shows, it has little estimated moderating effect between intelligence capacity and degree of terrorism in contradiction to H 3 .Additionally, VIF statistics show strong collinearity among the degree of terrorism indicators, which may further affect model performance.The chosen solution for this problem is to remove one of the degree of terrorism factors after multiple attempts to alter the variable.The number of killed by terrorism per countryyear factors (nkill and nkillus) were thus removed for models 1A and 2A, reducing multicollinearity to acceptable levels.Since the remaining factors, number of successfully attacks and number of wounded per country-year are closely correlated to the removed factor, the overall degree of terrorism latent variable continues to measure the relative impact of terrorism on each country.Consequently, the inner model is changed to remove the regime type variable (see Figure 3), and the outer models for Models 1B and 2B are refined to remove the variable for number of people killed in terrorist attacks per country-year.
Finally, simultaneous Gaussian Copula (GC) estimates are calculated for each model path to detect the presence of endogeneity (see Table 2).The GC models hidden relationships that could be affecting both independent and dependent variables; when significant, a GC indicates the presence of endogeneity.Models 1B and 2B show significant copulas for the path U.S. Military Deployment ➔ Degree of Terrorism, and Model 2A shows a significant copula for the path GDP ➔ Degree of Terrorism, which indicates endogeneity is present.Since Model 1A is functional, analysis moves forward using the GC approach proposed by Hult et al. (2018); results for the models showing endogeneity are reported leaving the GC in place.As a secondary measure, Models 2A and 2B are estimated without the control variables (see Figure B) that showed significant copulas (GDP and USMD respectively), which appears to reduce endogeneity in those models (i.e., re-estimated GCs are not significant).Results, reported in Table B, are not significantly different from the models reported with GCs in place.
Table 1 compares Model 1 and Model 2 in terms of reliability, validity, collinearity, effect sizes, and model fit statistics for both the original models (A) and those without outliers (B).Model 1A exhibits strong reliability with a Cronbach's α of 0.936, composite reliability (rho_a) of 0.947, and AVE of 0.94, while Model 2A has slightly lower values (Cronbach's α of 0.878, rho_a of 0.925, and AVE of 0.889).Model 2B, however, shows substantially weaker reliability (Cronbach's α of 0.435 and AVE of 0.471).Discriminant validity, assessed using HTMT, is acceptable with all values falling below .850,indicating no problematic overlap between constructs.Nevertheless, values approaching this threshold, like 0.718 for U.S. Military Deployment ➔ Degree of Terrorism in Model 1A, are notable.Collinearity statistics (VIF) show no significant multicollinearity concerns, with all values close to or at 1 in both the outer and inner models.The f-squared values reveal varying effect sizes of the predictors on degree of terrorism.For example, in Model 1A, U.S. Military Deployment has a large effect size of 0.496, while GDP has a small effect size of 0.014.Model fit statistics (SRMR and NFI) indicate an acceptable fit for Model 1 (SRMR of 0.02 and NFI of 0.877 for Model 1A), but a worse fit for Model 2 (SRMR of 0.039 and NFI of 0.82 for Model 2A), with the poorest fit seen in Model 2B (SRMR of 0.086 and NFI of 0.635).

Results and analysis
Table 2 shows the correlation matrixes for each latent variable combination in the 4 models.In Model 1A, GDP has a significant negative relationship with degree of terrorism (β = −0.09,p < 0.001), while intelligence capacity is positively but not significantly related (β = 0.168, p =  0.114).U.S. joint operations have a marginally significant negative relationship (β = −0.045,p = 0.055), and U.S. military deployment has a significant positive relationship (β = 0.608, p < 0.001).When removing outliers in Model 1B, the results remain consistent, with the exception that intelligence capacity becomes significant (β = 0.370, p < 0.001).In Model 2A, all variables have significant relationships with degree of terrorism, with GDP and U.S. joint operations being negative and intelligence capacity and U.S. military deployment being positive.Finally, in Model 2B, GDP becomes marginally significant (β = −0.059,p = 0.089) while U.S. joint operations remain not significant (β = −0.053,p = 0.197), but intelligence capacity and U.S. military deployment maintain their significant relationships with degree of terrorism.

Discussion
For the primary aim of this study, measuring the impact of intelligence on security, three of the four models show significant, positive correlations between intelligence capacity and degree of terrorism, with the strongest model (1A) showing no statistically significant relationship.That is, the data indicate that an increase in a state's intelligence capacity has, at best, no impact on its security -potentially a negative effect!These findings contradict H 1 and H 2 , which expect a positive impact on security.As hinted above, results from the latter three models should be interpreted with caution due to the likelihood that endogeneity is present, although steps were taken to reduce it.Furthermore, the failure of regime type to serve as a significant moderating variable between intelligence capacity and degree of terrorism contradicts H 3 , that democratic states will enjoy increased benefits from intelligence compared to their authoritarian counterparts.Together, these findings present challenges to the assumptions that (1) intelligence is a critical to security, and that (2) intelligence leads to fundamentally different outcomes in democratic versus authoritarian states.
In this case, it is prudent to develop a new hypothesis to explain why the data are inconsistent with the core assumption that intelligence is critical to security. 53Alternative theories, such as that leaders typically ignore intelligence, are inadequate to explain these results within the context of the theoretical framework, as they refute the assumption that more intelligence means more security.As this is just one study, and its results an apparent outlier in the intelligence studies literature, there is not yet cause to challenge the core assumption.Instead, we can build from the positive heuristic of the theoretical assumption that intelligence is necessary to security, which provides hints at how to change theory in a progressive way. 54he finding that intelligence may not be critical to security, then, directs us to evaluate the parameters of the model within the broader context of the study (i.e., within the literature indicating a positive relationship between intelligence capacity and security outcomes).Beyond the potential weaknesses of the model already discussed, there are three further parameters to examine, (1) the data, (2) the theory, and (3) the model constructs.First, the data.Public intelligence cooperation data simply may not be able to capture the potential relative difference between a U.S. cooperative event with the United Kingdom, a state having extensive preexisting intelligence relationships with the United States, and, say, Russia.There can be little doubt that a single public display of bilateral cooperation (as captured by the dataset) between the United States and any member of the Five Eyes or NATO communities pales in comparison to the daily intelligence flow among these international partners.The potentially extreme gulf in meaning between cooperative events may render USIIB data less useful for this kind of analysis.Additional data collection, and expansion of the USIIB dataset could help to further differentiate cooperative states based on their degree of security cooperation with the United States.
In terms of theory, Kent's tripartite conception of intelligence does not account for the inclusion of clandestine action as a component of intelligence.This is, unfortunately, inconsistent with the organization of U.S. intelligence today.Responsibility for covert action was transferred from the military to the intelligence community during the 20 th century for a variety of reasons, including their shared tradition of secrecy.This pairing of action and intelligence suggests that any measure of U.S. intelligence as a single enterprise will necessarily include covert action.Regarding the study at hand, this means that the USIIB dataset captures features of two constructs in one variable (USIC), (1) intelligence capacity, and (2) covert action capacity.In this study, the intelligence capacity latent variable consists of this single factor, and therefore the results cannot be used to differentiate between intelligence operations (e.g., collection, and analysis) and covert action.This indicates a need to further refine the intelligence capacity construct through additional theory development and data collection.

Intelligence and regime type
The role of regime type in this study is similarly evaluated.The data for regime type are based on the Freedom of the World index, which is a commonly used measure for the degree of democracy present in a given country-year.The theory supporting H 3 is not fully developed within intelligence studies, 55 although there are numerous works in the comparative politics literature on variations in institutional effectiveness between democratic and authoritarian regimes.Although the methods literature indicates single-factor latent variables can be appropriately used in PLS-SEM, the regime type latent variable could still be improved by including multiple indicators of democracy from more comprehensive datasets like Varieties of Democracy (V-Dem).

Conclusion
Both models assess the relationships between several factors (GDP, Regime Type, U.S. Joint Operations, and U.S. Military Deployment) and DEGREE OF TERRORISM In both models, Cronbach's α, composite reliability (rho_a and rho_c), and average variance extracted (AVE) show high values, indicating good internal consistency and convergent validity of the latent variables.Discriminant validity is assessed using the HTMT values, and most of them are below the common threshold of 0.85 or 0.90, which indicates acceptable discriminant validity.However, the value 0.

Appendix B. Models without variables potentially causing endogeneity
In Model 2A, Intelligence Capacity has a positive and significant relationship with Degree of Terrorism (β = 0.347, p < 0.001), U.S. Joint Operations has a negative and significant relationship with Degree of Terrorism (β = -0.128,p < 0.001), and U.S. Military Deployment has a positive and significant relationship with Degree of Terrorism (β = 0.408, p = 0.002).GDP is not included in this model.In Model 2B, Intelligence Capacity has a positive and significant relationship with Degree

Figure 2 .
Figure 2. PLS-SEM model of intelligence capacity and degree of terrorism.

H 1 :
An increase in Intelligence Capacity through cooperation with the United States will decrease the Degree of Terrorism in the cooperating country.

H 2 :
An increase in Intelligence Capacity through cooperation with the United States will decrease the Degree of Terrorism targeting Americans abroad.

Figure 3 .
Figure 3. Revised model of intelligence capacity and degree of terrorism.
719 for U.S. Military Deployment ➔ Degree of Terrorism in Model 1A is relatively high, suggesting potential concerns in distinguishing these two constructs.Regarding the inner model, the f-squared values indicate the effect size of the relationships, with higher values representing a stronger effect.U.S. Military Deployment ➔ Degree of Terrorim show the most substantial effect across both models, while GDP ➔ Degree of Terrorism and Regime Type ➔ Degree of Terrorism have relatively smaller effects.The collinearity statistics, measured by VIF values, indicate that multicollinearity is not an issue for most of the factors, as their VIF values are below the threshold of 5 or 10.However, some variables like Number of People Killed and Number of People Wounded in Model 1A and Number of Americans Killed in Model 2A have VIF values above the threshold, indicating potential multicollinearity issues.The model fit is evaluated using SRMR and NFI, with lower SRMR values and higher NFI values indicating a better fit.Model 1A and 1B have better fit values than Model 2A and 2B, but there is room for improvement in all models, especially in Model 2B, which has the lowest NFI value (0.416).
Figure B. Models altered to remove potentia endogeneity.

Table 1 .
Measures of model quality and fit.

Table 2 .
Correlation coefficients for the tested models.