1. What IGW is

After Klout shut down in 2018 and Forbes ended its "World's Most Powerful People" ranking the same year, there is no single open cross-domain indicator of influence in public circulation. Lists like Time 100 are qualitative; billionaire rankings only measure wealth; social-media follower counts reward celebrities over politicians. IGW is an attempt to combine six complementary signals into one indicator that is comparable across a president, a billionaire, a pope, an athlete and an artist.

Version 2.0 (2026) is a deliberate philosophical shift from v1.0. IGW v1.0 mixed three phenomena: public visibility, information interest and real decision power. v2.0 separates these dimensions, each has its own component and its own sub-index, and the combined score is presented as one of several possible cross-sections, not "the final truth".

2. How to interpret the score

We present IGW not as a "final ranking of influence" but as an open radar of global public influence. No single indicator will ever answer whether Cristiano Ronaldo, Xi Jinping, Taylor Swift and Elon Musk are "really" comparable, which is why alongside the combined score we publish four sub-indices and a weight-sensitivity test.

3. v2.0 formula

IGW v2.0 is the weighted sum of six normalized components (each 0-100), multiplied by 10, giving a final 0-1000 scale:

IGW v2.0 = 10 × (
    0.20·A   Attention Demand
  + 0.20·M   Media Agenda
  + 0.15·O   Owned Audience
  + 0.25·D   Decision Power
  + 0.10·N   Network Centrality
  + 0.10·T   Temporal Persistence
)

Weights and what they measure

Why v2.0 differs from v1.0: in v1.0 the combined weight of attention components (W+S+G) was 70% and decision power only 10%. In practice IGW v1.0 measured attention with a power correction. v2.0 raises D to 25% and introduces two new dimensions (Network, Temporal) that dampen the premium for a one-off scandal or viral moment.

4. Components

(A) Attention Demand

A composite combining information demand and its dynamics:

A = 0.60·Wikipedia + 0.40·Google Trends Momentum

Wikipedia (60% of A)

Sum of daily biographical pageviews in the ten most-popular languages (en, es, zh, hi, ru, ar, pt, ja, de, fr), aggregated to an annual average with a log10 transform. Source: Wikimedia Pageviews API.

Limitation: the Pageviews API only goes back to 1 July 2015. Years 2010-2014 have a lower data-quality class (C/D) and are reconstructed from Wikipedia stats.grok.se and interpolation around known milestones.

Google Trends Momentum (40% of A)

Google Trends measures the relative dynamics of interest in a person over time; values 0-100 are normalized within the query. It is not an absolute measure of popularity across people. So in v2.0 we treat Trends as momentum, not volume, and use fixed anchor terms (a few globally known names, e.g. "Barack Obama", "football", "COVID") to scale queries into a common comparison frame. Method details: Google Trends Understanding the data.

(M) Media Agenda

The person's presence in the media agenda, normalized to the internet's growing volume:

M = log(person_mentions / total_GDELT_mentions_in_year)
    × source_diversity
    × language_diversity

Source: GDELT Project, specifically Global Knowledge Graph 2.0 (available from 1 April 2013). For 2010-2012 we use GDELT Event Database 1.0 with a class-C confidence flag.

Why normalize by yearly volume: without it, 2024 automatically looks "bigger" than 2012 because the internet and media monitoring are larger, not because a given person has more influence. source_diversity and language_diversity guard against over-rewarding people popular in one country or one source type.

(O) Owned Audience

The person's own communication channel:

O = 0.70·log(total_followers)
  + 0.15·multi-platform breadth
  + 0.15·engagement proxy

Followers summed across X/Twitter, Instagram, TikTok, YouTube, Weibo, Facebook (public profile), measured at January each year. multi-platform breadth = count of platforms with active presence. engagement proxy (where available) = median like-rate of the last 30 posts. Where engagement cannot be retrieved historically we use the simplified variant:

O = 0.80·log(followers) + 0.20·number of active platforms

Sources: Social Blade, archives of public counters, Wayback Machine for 2010-2014.

Deliberate limitation: component O rewards publicly active people. Closed leaders (Xi Jinping, Ali Khamenei) score near zero on O despite real power, which is why the weight is only 15% (not 40-50% as Klout had) and is balanced by a high D weight.

(D) Decision Power

The most important component philosophically in v2.0 and the one most strongly reworked from v1.0. Influence can be formal, capital-based, cultural, infrastructural, epistemic or mobilizational; D tries to capture all these dimensions through an explicit scoring matrix instead of a single expert number.

Every person receives a 0-100 score per the matrix assigned to their type:

Politicians

CEOs / Founders

Religious leaders

Artists / Athletes

Activists

The full D table with per-person justifications is published together with the data in influence.json. We welcome disagreement; D is the component most exposed to criticism, which is why it has an explicit structure rather than a black box.

(N) Network Centrality

The person's centrality in the network of other people, organizations and topics, computed from the GDELT GKG co-occurrence graph. The more often a person is co-mentioned with diverse institutions and other people, the higher N. The component penalizes isolated celebrities and rewards people who operate inside networks, which is part of why supreme-court justices, central-bank governors and diplomats are better surfaced in v2.0 than in v1.0.

(T) Temporal Persistence

A 3-year discounted weighted average (0.5/0.3/0.2 for current/previous/two-years-back) applied to the A+M+O+D+N total. The component penalizes one-year spikes (scandal, viral) and rewards sustained presence. Without T the ranking would be highly vulnerable to seasonal swings, e.g. Greta Thunberg 2019 or Putin 2022 would dominate the entire period.

5. Normalization

Each component is brought to a 0-100 scale by percentile-rescaling against a broad reference universe of ~600 public figures, not only the 19 narrative figures. This is the key change from v1.0.

Normalization procedure per component per year:

1. Collect raw values for all ~600 people in the benchmark universe.
2. Apply a log10 transform for heavy-tailed components (Wikipedia, GDELT, followers).
3. Winsorize outliers at the 1st and 99th percentiles.
4. Rescale percentile to 0-100 with fixed anchors: 0 = 5th percentile, 50 = median, 100 = 99th percentile.
5. Display the 19 narrative figures in the animation.

Anchors are stable across editions; a score of 70 in component A in 2024 and 2026 means the same position relative to the reference population, not relative to the current top 19.

6. Data-quality classes

Every (component × person × year) combination has an assigned data-quality class:

7. Sub-indices

Alongside the combined score we publish four influence profiles. Same set of components, different weights, showing which type of influence dominates for a person:

Sub-indices make it visible that Taylor Swift dominates in Audience and Attention while Xi Jinping dominates in Power and Network, and there is no need to pretend they are comparable on a single flat type of influence.

8. Confidence score

Every IGW result has an A/B/C/D confidence level based on data completeness, number of sources, quality of historical data, agreement across weight variants and result stability across methods:

Example: Donald Trump 2024: IGW 880 / confidence A · Xi Jinping 2024: IGW 780 / confidence B · Dalai Lama 2024: IGW 400 / confidence C+ · Pope Leo XIV 2025: IGW 780 / confidence C (short tenure, sparse post-election data).

9. Weight-sensitivity test

Because IGW is a composite indicator, the result depends on decisions about weights, normalization and source selection. So with every major edition we publish seven alternative variants and compare position stability:

1. Baseline v2.0 weights (reference)
2. Equal weights (each component 1/6)
3. No Owned Audience (O = 0, redistributed)
4. No Decision Power (D = 0, redistributed)
5. Attention-only (A + M, rest = 0)
6. Power-heavy (D = 50%, rest proportional)
7. Media-heavy (M = 40%, rest proportional)

People whose position stays stable across most variants get confidence A or B. People whose position depends heavily on a single weight get C or D and are clearly marked on the chart.

10. Person selection

v2.0 separates two things:

Calibration universe (~600 people)

A broad benchmark of global public figures used to normalize every component. It comes from the intersection of: top 1000 Wikipedia pageviews 2015-2025 (10 languages), top 200 GDELT GKG persons per year, and people from Time 100 / Forbes Powerful People / Bloomberg Billionaires 2010-2026 lists. The calibration universe is not shown in the animation; it exists only to stabilize the scale.

Narrative list (19 people)

The people shown in the animation and ranking. Selection criteria: (1) global recognition, (2) at least one year in the top of Time/Forbes/Atlantic 2010-2026, (3) geographic balance (USA, China, Europe, India, South America, Middle East), (4) profession diversity (politics, business, religion, sport, culture, activism, technology).

Full list: Barack Obama, Donald Trump, Xi Jinping, Vladimir Putin, Narendra Modi, Angela Merkel, Pope Francis, Pope Leo XIV, Dalai Lama, Elon Musk, Bill Gates, Mark Zuckerberg, Jeff Bezos, Sam Altman, Taylor Swift, Cristiano Ronaldo, Lionel Messi, Greta Thunberg, Beyoncé.

11. Scope & limitations

Indicator scope

IGW measures the global public influence of living and publicly measurable people. It does not cover: anonymous individuals, intelligence officers, undisclosed decision-makers, or people whose influence is real but invisible in open data (e.g. hedge-fund heads operating outside the media, oligarchs avoiding publicity, anonymous intelligence chiefs).

Known biases

• Language bias. Wikipedia, GDELT and Trends are more heavily populated in Western languages; people active primarily on the Chinese internet (Weibo, Baidu, Toutiao) are underestimated despite three Asian languages in the Wikipedia basket.
• Recency bias. M, A and Trends reward current events. T (Temporal Persistence) softens this but does not eliminate it.
• Social-media bias. Component O rewards publicly active people, which is why its weight is only 15% and is balanced by a high D.
• 2010-2014 reconstruction. Most APIs from that period don't exist. Those years have data-quality class B-C, marked separately on the chart (lower bar opacity).
• Component D is an expert assessment. Despite the scoring matrix, it remains subjective. That is why we publish a "no-D" variant in the sensitivity test.
• Trends anchor terms. Scaling to a common frame through an anchor basket reduces but does not eliminate the relativity problem of Google Trends.

12. Changelog

13. Citation

If you use IGW in an article, presentation or chart, please link back to this page and cite the version, e.g.:

IGW v2.0, WyścigPKB, https://wyscigpkb.pl/metodologia

14. Sources & verification points

• Wikimedia Analytics / Pageviews API · data from 1 July 2015.
• Wikimedia Pageviews FAQ · no data older than July 2015.
• GDELT Project · many datasets; v2.0 uses GKG 2.0.
• GDELT Data · Global Knowledge Graph 1.0 available from 1 April 2013.
• Google Trends Understanding the data · 0-100 values are normalized, not absolute.
• Google Trends FAQ · data for popular terms, with its own normalization and filtering.
• OECD/JRC Handbook on Constructing Composite Indicators · the standard for composite indicators: normalization, weighting, aggregation, imputation, sensitivity analysis and transparency.
• Social Blade · historical social-media counters.