The Fragility Gap | Quanta Analytica Process™

00 // DATASET

What the Data Is

The dashboard behind this product assembles five independent index families into one country level table. Each family measures something different, and the differences matter more than the overlap.

Family	Publisher	Vintage	Fields Used	Scale and Direction	What It Actually Measures
GPI	Institute for Economics and Peace	2024	1	1 to 5, higher is worse	Negative peace. The observed absence of violence and militarisation right now.
PPI	Institute for Economics and Peace	2024	1	1 to 5, higher is worse	Positive peace. The attitudes, institutions and structures associated with sustaining peace.
FSI	Fund for Peace	2023	12 sub-indicators	0 to 10 each, higher is worse	Cohesion, economic, political and social pressures on the state. Totals recomputed as the sum of the twelve fields, range 0 to 120.
BTI	Bertelsmann Stiftung	2024	14 indicators plus regime coding	1 to 10, higher is better, except Q13.3 conflict intensity where higher is worse	Expert coded political and economic transformation quality, governance performance and conflict intensity.
HDI	UNDP	2023	2	0 to 1 and years, higher is better	Human development outcome and mean years of schooling.

Twenty-two states appear in the dashboard. Nineteen carry values across all five families and form the analytic sample. Gambia appears in the BTI table only. Cote d'Ivoire and Guinea-Bissau are absent from the FSI panel. Those three states are retained in the dataset and in the map product, and excluded from any statistic that requires the full indicator set. That exclusion is stated wherever it applies rather than silently absorbed.

Direction normalisation. The five families do not share a sign convention. GPI, PPI and FSI increase with risk. BTI and HDI increase with performance, with the single exception of BTI Q13.3. Every derived statistic in this product operates on explicitly signed variables, and no composite score mixes directions without inversion. Getting this wrong is the most common failure mode in multi-index dashboards, and it is invisible in the output.

01 // BLUF

Bottom Line Up Front

Fragmentation of the coercive apparatus, not legitimacy deficit, tracks conflict intensity across this sample.

FactFSI C1 security apparatus correlates with BTI conflict intensity at r = +0.81, p below 0.001, 95 percent bootstrap interval +0.62 to +0.94. FSI P1 state legitimacy correlates at r = +0.39, p = 0.10, with a bootstrap interval whose lower bound sits at +0.005. The legitimacy association is marginal on both readings, and in a two predictor regression its coefficient reverses sign and cannot be distinguished from noise.

Two clusters share an identical legitimacy deficit and differ by 3.5 points of conflict intensity.

FactThe seven state Contested Monopoly group and the five state Coercively Stable Rentier group both average 8.74 on FSI state legitimacy. Their mean conflict intensity is 8.14 and 4.60 respectively. AssessmentWhat separates them is intact coercive capacity, resource financed patronage, and lower external intervention, not popular consent.

Eight states in the sample carry structural deficits larger than their current violence levels imply.

AssessmentEquatorial Guinea, Mauritania, Republic of the Congo, Liberia, Chad, Guinea, Sierra Leone and the Central African Republic sit above the parity line between standardised positive peace deficit and standardised current violence. CaveatThis is a static positional statement about one year of data. It is not a prediction that these states will deteriorate.

Mali and Burkina Faso show the opposite signature.

AssessmentBoth record violence well above what their structural indicators alone would place them at, with gap scores of minus 1.45 and minus 1.78. The most parsimonious reading is that conflict in the central Sahel has become partly self-sustaining, driven by armed actor dynamics rather than by the underlying structural profile that produced it.

02 // METHOD

How the Numbers Were Produced

Techniques applied: direction normalisation, z-standardisation, residual gap construction, Pearson correlation with non-parametric bootstrap intervals (8,000 resamples), ordinary least squares with two predictors, k-means partitioning (k = 4, 50 initialisations, fixed seed), and principal component analysis for two dimensional projection. All computation in Python using NumPy, SciPy, pandas and scikit-learn. Quality control: every value extracted from the source dashboard by coordinate parsing rather than manual transcription, then row checked against the rendered table.

The Fragility Gap Construction

The gap is deliberately simple, because a simple statistic survives scrutiny better than a weighted composite whose weights nobody can defend. Both the Global Peace Index and the Positive Peace Index are standardised to zero mean and unit variance within the nineteen country sample. The gap is the standardised positive peace score minus the standardised negative peace score.

A positive gap means the state looks structurally weaker than its current level of violence would suggest. A negative gap means violence is running ahead of structure. The two indices correlate at r = +0.52 within this sample, which is high enough to be meaningful and low enough that the residual carries information.

CaveatStandardising within a nineteen country regional sample means every score is relative to this region, not to the world. A gap of zero here describes a state that is typical for West and Central Africa. It does not describe a stable state.

03 // FINDING 01

Where Structure and Violence Come Apart

The scatter below places every state in the analytic sample on two axes. The horizontal axis is current violence and militarisation as measured by the Global Peace Index. The vertical axis is structural peace deficit as measured by the Positive Peace Index. Both are standardised within the sample. The diagonal is the parity line where the two are equal.

Figure 1 // Structural Deficit Against Current Violence

Standardised scores, n = 19. Points above the diagonal carry more structural deficit than present violence. Hover any point for detail.

Contested Monopoly Post-Transition Low Capacity Coercively Stable Rentier Comparatively Consolidated

Axes standardised within the 19 country sample. GPI and PPI are both inverted scales where higher values indicate worse conditions. Diagonal is the line of equality between standardised scores.

Read the upper left quadrant carefully. Equatorial Guinea, Mauritania and the Republic of the Congo sit there because they combine low observed violence with severe structural deficit. That combination is the classic profile of coercive stability. The state is quiet because it is capable of suppressing dissent and can pay for that capability, not because grievances have been resolved.

Figure 2 // The Fragility Gap, Ranked

Standardised positive peace deficit minus standardised current violence. Positive values indicate structural deficit running ahead of observed violence.

Derived statistic. Computed for this product from GPI 2024 and PPI 2024. Not published by the source institutions in this form.

The lower end of that ranking is as informative as the upper end. Burkina Faso and Mali record the most negative gaps in the sample. Their structural indicators are bad but not regionally exceptional. Their violence is regionally exceptional. AssessmentWhere violence outruns structure by this margin, the conflict system has usually acquired its own drivers: armed actor competition, war economies, and retaliation cycles that no longer depend on the conditions that started them. That reading is consistent with the region's trajectory but it is inference from position, not from event data, and it is held at moderate confidence.

04 // FINDING 02

Legitimacy Is Not the Variable

A large share of humanitarian and security risk writing on this region, including work I have produced myself, leads with state legitimacy failure as the organising explanation for operational risk. This dataset does not support that ordering. It supports a different one.

Figure 3 // Correlation With BTI Conflict Intensity

Pearson r with 95 percent bootstrap confidence intervals, 8,000 resamples, n = 19. Slate markers indicate coefficients that are not distinguishable from zero at p below 0.05.

Bootstrap intervals are percentile intervals. With n = 19 they are wide by construction, which is exactly why they are shown rather than bare coefficients. Note the disagreement on P1 state legitimacy: the percentile interval clears zero by 0.005 while the parametric test returns p = 0.10. Where two defensible tests disagree at the margin the correct reading is that the evidence is weak, not that one test wins.

Security apparatus fragmentation and external intervention both correlate with conflict intensity above +0.80 with intervals that exclude zero comfortably. State legitimacy sits at +0.39 with a lower bound at zero. Rule of law sits at minus 0.34 with an interval that spans zero. Those two results should change how the causal story is written.

Holding Fragmentation Constant

Correlation alone would leave the question open, because legitimacy and fragmentation are themselves related. A two predictor regression settles it within this sample.

Model	Predictor	Coefficient	t	Model fit
Model A	C1 security apparatus	+1.011	+5.04	R² = 0.671, adjusted 0.630
	P1 state legitimacy	-0.197	-0.97	R² = 0.671, adjusted 0.630
Model B	C1 security apparatus	+0.495	+2.47	R² = 0.760, adjusted 0.730
	X1 external intervention	+0.598	+2.68	R² = 0.760, adjusted 0.730

In Model A the legitimacy coefficient reverses sign and cannot be distinguished from zero. In Model B, fragmentation of the coercive apparatus and the presence of external intervention together account for roughly three quarters of the variance in conflict intensity across the sample.

CaveatNineteen observations, one time slice, and two predictors chosen after inspecting correlations. This is a description of structure in one dataset. It is not a causal model, and the external intervention indicator is partly endogenous, since intervention follows conflict as well as accompanying it. The finding is held at moderate confidence and is offered as a reordering of priors, not as a settled result.

05 // FINDING 03

Four Operating Environments, Not One Region

Ten standardised indicators spanning all five families were partitioned with k-means at k = 4. Two principal components retain 83.1 percent of the variance, which makes a two dimensional projection an honest summary rather than a decorative one.

Figure 4 // Principal Component Projection With Cluster Assignment

PC1 61.9 percent of variance, PC2 21.2 percent. PC1 loads positively on fragility, violence and fragmentation. PC2 separates human development and state legitimacy from governance quality.

Contested Monopoly Post-Transition Low Capacity Coercively Stable Rentier Comparatively Consolidated Diamond marker = cluster centroid

k-means, k = 4, 50 initialisations, fixed random seed for reproducibility. Cluster labels are analyst assigned after inspecting the group means. The algorithm produced the partition, not the names.

Cluster	n	Members	C1 Fragmentation	P1 Legitimacy	Conflict Intensity	HDI	FSI Total	Mean Gap

Figure 5 // Identical Legitimacy Deficit, Divergent Violence

Contested Monopoly against Coercively Stable Rentier, cluster means across six indicators.

Both clusters average 8.74 on FSI state legitimacy. The divergence appears in coercive fragmentation, external intervention, conflict intensity and socioeconomic development.

This is the central visual argument of the product. Two groups of states, indistinguishable on the indicator that most analysis leads with, separated by three and a half points of conflict intensity. Legitimacy deficit is common to both. What varies is whether the state retains a coherent monopoly on force and whether it can finance patronage from rents.

Analyst override on record. The algorithm placed Niger in the Post-Transition Low Capacity cluster alongside Liberia, Sierra Leone and Togo. That placement is driven by BTI governance scores drawn from a review window that closes before the July 2023 seizure of power, and by the resulting Democracy regime coding. On present conditions Niger belongs with the Contested Monopoly group. The clustering output is reported as computed and the override is recorded here rather than applied silently to the data.

06 // DESIGN

Why the Visualisations Are Built This Way

Three decisions shaped every figure in this product.

Position over colour

Every quantitative claim is encoded as position on a common scale. Colour carries category only. Readers decode position accurately and decode colour intensity poorly, and a choropleth of a composite index invites exactly the misreading this product is arguing against.

Uncertainty is drawn, not disclaimed

Figure 3 shows bootstrap intervals rather than bare coefficients, because with nineteen observations the interval is the finding. A bar chart of correlation coefficients alone would have overstated the evidence by a wide margin.

The comparison is the argument

Figure 5 exists because the strongest claim in this product is a comparison between two groups on one indicator. Building a figure whose only job is to carry the argument is more useful than a dashboard that shows everything and asserts nothing.

Derived values are labelled

The fragility gap and the FSI totals are computed for this product. Neither is published by the source institutions in this form. Every figure note states whether a value is sourced or derived, so a reader can audit the chain without opening the code.

07 // LIMITS

Limitations and Falsifiability

What This Establishes

The association between coercive fragmentation and conflict intensity is strong and robust to bootstrap resampling in this sample.
State legitimacy carries no independent association with conflict intensity once fragmentation is controlled.
Two clusters with identical legitimacy profiles differ sharply in violence output.
Positive peace deficit and current violence diverge systematically in eight states.

What It Suggests

Operational risk models weighted toward governance quality may misrank rentier autocracies.
Coercively stable states carry deferred rather than absent risk.
Central Sahel violence has decoupled from its structural baseline.
Regime type coding is a weak operational discriminator relative to security sector coherence.

What It Cannot Support

Any causal claim. This is one cross-section with nineteen observations.
Any forecast of onset, timing or location of violence.
Sub-national inference. Every value is a national aggregate.
Currency. The most recent input closes in 2024.

What Would Falsify the Central Finding

ID	Test	Result That Would Overturn the Finding
F1	Extend the panel to multiple years and regress change in conflict intensity on lagged C1 and lagged P1.	Lagged legitimacy predicts change and lagged fragmentation does not.
F2	Repeat the analysis on a global sample rather than one region.	The fragmentation coefficient collapses outside West and Central Africa, indicating a regional artefact.
F3	Substitute event based conflict measures for the BTI expert coded intensity score.	The association reverses or disappears, indicating the result reflects coder correlation rather than a real relationship.
F4	Track the eight positive gap states forward for thirty-six months.	Deterioration rates are no higher than in the negative gap group, indicating the gap carries no early warning content.

08 // SO WHAT

Operational Consequence for Security Risk Management

For an NGO security risk manager, the practical consequence is a change in what gets monitored. If fragmentation of the coercive apparatus is the variable that tracks violence, then the indicators worth watching are the ones that describe that fragmentation directly.

Immediate

Re-weight country risk models away from composite governance scores and toward security sector coherence. Command chain integrity, parallel force structures, auxiliary and militia proliferation, and foreign force presence are the leading terms.

Near term

Treat the eight positive gap states as deferred risk rather than low risk. Acceptance strategies built on the current calm in Equatorial Guinea, the Republic of the Congo or Mauritania rest on coercive capacity that can change faster than any structural indicator will register.

Medium term

Build the panel. A single cross-section cannot separate structural drivers from conflict feedback. Three or more index vintages would allow the lagged tests in F1 and turn a positional finding into a directional one.

09 // SOURCES

Source Register

#	Source	Type	Band	Independence Note	Locator
1	Global Peace Index 2024, Institute for Economics and Peace	Composite index	Green	Independent institute. Methodology published. Shares some input data with PPI, so GPI and PPI are not fully independent of one another.	visionofhumanity.org
2	Positive Peace Index 2024, Institute for Economics and Peace	Composite index	Green	Same publisher as source 1. Treated as partially dependent for the gap statistic, which is why the gap is presented as a positional descriptor rather than as two independent measurements.	visionofhumanity.org
3	Fragile States Index 2023, Fund for Peace	Composite index	Green	Independent publisher. Content analysis driven, so partially sensitive to media coverage volume.	fragilestatesindex.org
4	Bertelsmann Transformation Index 2024, Bertelsmann Stiftung	Expert coded index	Amber	Expert assessment rather than measurement. Two year review window closing before the 2023 coup sequence, which affects regime coding for Niger and Gabon.	bti-project.org
5	Human Development Report, UNDP, HDI 2023	Official statistic	Green	Intergovernmental. Relies on national statistical reporting, which is weak in several sample states.	hdr.undp.org
6	ARAC Global Stability Analytics Dashboard v1	Analyst compiled	Amber	The compilation is the analyst's own. Values were re-extracted by coordinate parsing and checked against the rendered tables rather than accepted on trust.	Internal

Band key. Green: established methodology, published documentation, independent publisher. Amber: credible but with a structural limitation stated in the note. Red: not used in this product. Independence test: no key judgment in this product rests on a single source. The central finding in section 04 draws on FSI and BTI simultaneously, which are produced by different institutions using different methods.