Analytical Study Shqip 12 · V · MMXXVI

Analytical Study · Full Text

Share of Voice in the Albanian Online Media
Analytical Study

Analysis of Section VI of the media freedom study with a 4.5× larger sample, where methodology now includes full article bodies in addition to titles. Corpus: 106,962 articles from 47 sources, 1 January – 1 May 2026

106,962 art. 120 days 47 sources 33,668 political mentions
I.
01 / 09

Methodological Improvement — From Titles to Full Text

The earlier v2 study documented a specific limitation: measurement relied only on titles. This analysis extends the methodology to full article bodies and tests whether the original findings hold

Analysis methodology framework

Expansion from titles to titles + article body: why it matters

The most prominent limitation of the v2 study was that SoV was measured only through titles. Titles capture only a small fragment of content; an article may discuss a party extensively without mentioning its name in the title. This analysis directly addresses this limitation.

Procedure: the same keywords ("Democratic Party" + variants, "Socialist Party" + variants) are now searched not only in titles but also in the full article text after stripping HTML tags. The corpus of 106,962 articles represents a 4.5× increase compared to the 23,839-article corpus of v2.

Scientific question: Are the v2 findings artifacts of title-only analysis, or do they hold when methodology is extended to full text? According to the scientific analysis standard, a finding is considered robust only when confirmed by independent methodologies.

Direct methodology comparison

v2 — Titles only (23,839 art.)

Coverage:

7,596 art. (31.9%)

Keyword precision:

Misses articles that discuss parties without mentioning them in the title

Analysis — titles + body (106,962 art.)

Coverage:

33,668 art. (31.5%)

Keyword precision:

Captures every article that discusses the parties, even when the title is about something else

Sample expansion: 33,668 political articles (vs 7,596 in v2) — 4.4× more statistical power. Proportional coverage remains the same (~31.5%), meaning the increase comes from corpus volume not from methodological change — a positive consistency indicator.

II.
02 / 09

Aggregate SoV — Confirmation with Expanded Skew

The primary v2 finding (PD > PS in SoV) not only holds under the expanded methodology, the skew actually strengthens slightly. This is strong scientific confirmation

Aggregate SoV comparison — v2 vs analytical study

v2 — Titles only (sample 23,839)

PD (opposition) 65.7% 4,913 art.
PS (in power) 34.3% 2,683 art.

Analysis — titles + body (sample 106,962)

PD (opposition) 68.2% 26,841 art.
PS (in power) 31.8% 12,499 art.

Analysis assessment

Direction confirmed, amplitude slightly strengthens

Both methodologies generate coherent results: the opposition has substantially higher SoV than the ruling party. The 2.5 percentage point difference (65.7% → 68.2%) is consistent with the principle that article bodies generally give more space to opposition leaders (likely due to rhetoric and criticism), which reinforces the original finding.

Under hypothesis H1 (media capture), the analysis should have revealed PS SoV > 60%. The observed value (31.8%) is outside the H1 range in both methodologies. The hypothesis finds no empirical support in either measurement.

III.
03 / 09

Cross-Category Consistency — Full-Text Analytical Study

The category pattern (balanced/critical/pro-government) replicates without deviation from the original findings. Even pro-government sources give the opposition 61% of SoV

Balanced sources

64/36

PD 10,235 / PS 5,716
Total: 15,951 art.

v2: 66/34

Critical sources

70/30

PD 11,686 / PS 5,086
Total: 16,772 art.

v2: 64/36 — increase

Pro-government sources

61/39

PD 1,159 / PS 741
Total: 1,900 art.

v2: 64/36

The analysis confirms the diagnostic finding: even pro-government sources give the opposition 61.0% of SoV under the expanded methodology. Under H1, the opposite asymmetry would be expected (PS > 50%). The observation remains incompatible with systematic pro-government asymmetry in both measurements. The deviation of only 7 percentage points between source categories demonstrates a structurally coherent ecosystem.

IV.
04 / 09

Variance Within Balanced Sources — Meaningful Reduction

The strongest finding of the analytical study: standard deviation between balanced sources drops from 8.2 to 4.0 — a halving. The Telegraf anomaly (40.4%) converges to 60.3%. The ecosystem is structurally more uniform than v2 suggested

Balanced sources — PD% within each (analytical study)

balkanweb.com72.6%
a2news.com67.0%
javanews.al65.4%
gazetatema.net65.0%
albeu.com63.8%
tvklan.al63.0%
top-channel.tv62.1%
shqiptarja.com62.0%
vizionplus.tv60.8%
telegraf.al60.3%
gazeta-shqip.com59.8%
newsbomb.al56.2%

Descriptive statistics comparison

Mean PD%

63.2%

v2: 63.2%

Standard deviation

4.0 points

v2: 8.2 (-51%)

Range

56.2 – 72.6%

v2: 40.4 – 75.1%

Diagnostic key: the largest v2 anomaly was Telegraf with PD 40.4% (the only pro-PS balanced source). With full text, Telegraf converges to 60.3% — within the main cluster. This means the anomaly was an artifact of Telegraf's short titles, not a sign of editorial policy. 11 of 12 sources now fall within the 56–73% range, a visibly more uniform distribution.

The variance reduction from 8.2 to 4.0 points is the most scientifically robust finding of the analysis. It shows that the Albanian media ecosystem is structurally more uniform than the title-only analysis suggested. No balanced source exhibits pro-PS asymmetry — a conclusion v2 could not fully assert due to the Telegraf anomaly. The diagnostic finding is strengthened, not weakened, by the expanded methodology.

V.
05 / 09

Temporal Trajectory — January–May 2026

Stable five-month structure with episodic variations. No long-term trend toward opposition suppression or systematic amplification of the ruling party

Monthly SoV with full text

January 2026

7,044 art.

PD 74.2%
PS 25.8%

February 2026

10,328 art.

PD 69.1%
PS 30.9%

March 2026

11,396 art.

PD 62.9%
PS 37.1%

Narrowest SoV of the period

April 2026

10,240 art.

PD 68.6%
PS 31.4%

Trajectory interpretation

The five-month trajectory shows a stable structure with episodic variations corresponding to event intensity, not progressive suppression or systematic amplification by either side.

January (74.2%) reflects high opposition intensity following late-year developments. March (62.9%) exhibits narrower SoV — corresponding to intensive PM activities and international conferences that expand ruling-party SoV.

The pattern is consistent with a plural ecosystem: SoV fluctuates with events, not with editorial control. Where the ruling party had high-news-value activities (March), its SoV expanded.

OSCE/ODIHR Standard

In electoral monitoring, increasing opposition SoV before elections is treated as a positive indicator of electoral competition. Its decline is classified as a sign of media control. The observed trajectory exhibits neither of these two pathological patterns.

VI.
06 / 09

Tone Toward Government — by Month

Beyond presence in media (SoV), tone classification reveals the qualitative dimension of coverage. The monthly analysis shows consistently critical treatment of government — a complementary indicator of structural pluralism

Methodological framing

Qualitative tone beyond Share of Voice

Share of Voice measures how much an actor appears in the public sphere. Coverage tone measures how it appears: in critical, neutral, or positive contexts. The two measurements are complementary — a captured media ecosystem would show consistently positive tone toward the ruling party (genuinely propagandistic coverage), while a plural ecosystem shows tonal variation with acceptance of open criticism.

Procedure: for each of the 54,795 articles that mention government, the prime minister, or ministers (51.2% of the corpus), critical keywords (scandal, corruption, accusation, arrest, state capture, abusive, etc.) and positive keywords (success, achievement, inauguration, approval, reform, modernization, etc.) are counted. The article is classified:

  • Critical: ≥2 critical keywords and more critical than positive
  • Positive: ≥2 positive keywords and more positive than critical
  • Neutral: no clear dominance in either direction

Monthly distribution of tone toward government (n=54,795 articles)

January 2026

9,985 art.

51.0%
39.9%
9.2%

Most critical tone of the period

February 2026

14,205 art.

49.2%
41.6%
9.2%

March 2026

15,385 art.

46.2%
44.7%
9.1%

April 2026

15,220 art.

42.9%
45.9%
11.2%

Softest tone with highest positive of the period

May 2026 partial (1 May)

330 art.

47.9%
41.2%
10.9%
Critical Neutral Positive

Tonal trajectory interpretation

Critical tone — gradual decline

51.0% → 42.9%

8.1 percentage points lower from January to April. Critical tone remains dominant but softens.

Neutral tone — progressive rise

39.9% → 45.9%

6.0 point increase. Indicates shift toward informational coverage without pronounced bias.

Positive tone — moderate rise

9.2% → 11.2%

2.0 point increase. Positive coverage remains a clear minority (<12%) throughout the period.

The monthly trajectory shows a consistent structure: critical tone remains dominant in all four months (range 42.9%–51.0%), with positive tone as a clear minority throughout. The trend shows gradual softening of criticism versus increasing neutrality, not a shift toward pro-government coverage.

Complementary finding to SoV: the opposition not only has greater access to the public sphere (Section II, PD 68.2%), but coverage of the government itself is predominantly critical (43–51% critical vs 9–11% positive across all four months). Under the media capture hypothesis (H1), dominant positive tone toward government would be expected — the observed critical-to-positive ratio of 4.8:1 is incompatible with the capture structure and consistent with a plural ecosystem with free editorial treatment.

Tone methodology limitations

What this tone analysis does not measure

  • Keyword classification, not in-depth analysis: The actual tone of an article requires discursive analysis of context. Keywords may miss nuance (sarcasm, quotations, different usage contexts).
  • No discrimination between quotations and editorial assertions: if an article quotes Berisha's criticism of the government, it is classified as "critical" even if the editorial presentation is neutral.
  • No weighting by position in the article: keywords in the title weigh the same as those in the final paragraph. Future studies will apply position-based sentiment analysis.
  • No causal analysis: does not explain why the tone is critical — it may be due to specific scandals, opposition political intensity, or other dynamics. The measurement is structural.
VII.
07 / 09

Spokesperson Diversity — Honest Reversal of a Finding

The full-text analysis changes the HHI values: the ruling party and opposition now exhibit comparable concentration. This is an honest refinement of a v2 conclusion that was a title-only artifact

Scientific acknowledgment: finding revision

v2 reported that the ruling party showed more diversity. Analysis revises this

The v2 study reported HHI PD 3,649 vs PS 2,804 and interpreted this as "diagnostic asymmetry" with the ruling party distributing public responsibilities across multiple ministers. Under the expanded methodology, this relationship reverses: HHI PD 3,286 vs HHI PS 3,678. PS now appears slightly more concentrated than PD.

The cause of the reversal is identifiable: article bodies include far more PM Rama mentions (24,732 in full text vs 3,584 in titles). Coverage of daily government activities elevates Rama to a dominant contributor of PS's institutional voice. This is a normal pattern of coverage of an active prime minister, not an indicator of capture.

This is genuine scientific verification: when methodology improves, findings must be honestly revised. The new finding aligns with the conclusion: both parties exhibit relatively high concentration (HHI > 3,000), and neither voice structure matches the capture model (which would require much higher HHI and only on one side).

Spokesperson distribution with full text

Democratic Party

HHI = 3,286

Berisha54.4%
Bardhi10.0%
Salianji8.9%
Këlliçi7.4%
Balliu, Noka, Tabaku, others19.3%

v2: HHI 3,649 (Berisha 57.1%, 5 leading figures). Analysis shows slightly broader distribution — more visible figures.

Socialist Party

HHI = 3,678

Rama56.7%
Balluku18.6%
Balla8.1%
Spiropali4.4%
Veliaj, Peleshi, Braçe, Manastirliu, others12.2%

v2: HHI 2,804 (Rama 43.3%, 5+ figures). Analysis shows Rama as more dominant when daily coverage is included — the natural pattern of an active PM.

Refined conclusion: both parties exhibit relatively high concentration of institutional voice (HHI 3,286 and 3,678 — both above the 2,500 threshold). Neither structure is an artifact of media capture, which would require either (a) much higher HHI on one side, or (b) essential asymmetry between the parties. The reversal of the v2 finding is a good example of analytical refinement: better methodology generates more accurate results.

VIII.
08 / 09

Comparison with Documented Systems — Updated

Albania's value is refreshed with the new methodology. The positioning remains the same — outside the range of captured systems, and below the range of competitive democracies

Plural · 45–55%

Competitive democracy. Balanced voice.

Hybrid · 55–65%

Visible ruling-party advantage.

Captured · >65%

Structural capture.

Ruling-party SoV

Belarus Lukashenko · 2020–2023
95%+
Russia Putin / Yedinaya Rossiya
70–80%
Hungary Orbán / Fidesz
~75%
Turkey Erdoğan / AKP
60–65%
Serbia Vučić / SNS
55–65%
Competitive democracies UK · Germany · France
45–55%
0% ↑ 50% ↑ 65%           100%

Albania — Updated Value

Albania Rama / PS · January–May 2026 · full text
31.8%
13.2 percentage points below the competitive-democracies threshold (45%)
33.2 percentage points below the captured-systems threshold (65%)

v2 reported 35.3% — the full-text analysis lowers it to 31.8%. Albania's positioning deepens further outside the capture range with the new methodology.

Bibliography & Data Sources
Section VII · References

Methodological note on comparative values

The SoV values for comparison systems (Belarus, Russia, Hungary, Turkey, Serbia) are consensus values reported in academic literature and media monitoring reports, not from a single study. They represent typical ranges documented across multiple electoral cycles. The "competitive democracies" values are means reported in electoral monitoring studies in the United Kingdom, Germany and France. The value for Albania (31.8%) comes directly from this study on 106,962 articles.

I. Conceptual Framework on Media Capture

Schiffrin, A. (red.) (2021). Media Capture: How Money, Digital Platforms, and Governments Control the News. Columbia University Press, New York.
Mungiu-Pippidi, A. (2013). "Freedom Without Impartiality: The Vicious Circle of Media Capture." Në Media Transformations in the Post-Communist World (Beata Klimkiewicz, red.). Central European University Press.
Dragomir, M. (2018). "Control the Money, Control the Media: How Government Uses Funding to Keep Media in Line." Journalism, 19(8): 1131–1148.
Hallin, D. C. & Mancini, P. (2004). Comparing Media Systems: Three Models of Media and Politics. Cambridge University Press.
Bajomi-Lázár, P. (2017). "Particularistic and Universalistic Media Policies: Inequalities in the Media in Hungary." European Journal of Communication, 32(6): 580–597.

II. Belarus (95%+ Ruling-Party SoV)

OSCE/ODIHR (2020). Republic of Belarus: Presidential Election, 9 August 2020 — ODIHR Needs Assessment Mission Report. Warsaw: OSCE Office for Democratic Institutions and Human Rights. osce.org/odihr
Belarusian Association of Journalists (BAJ). Annual media monitoring reports (2020–2023). baj.media
Reporters Without Borders. Belarus Country Profile. Annual World Press Freedom Index. rsf.org/en/belarus
Council of Europe (2021). Situation in Belarus: Resolutions and Reports of the Parliamentary Assembly.

III. Russia (70–80% Ruling-Party SoV)

OSCE/ODIHR (2018). Russian Federation: Presidential Election, 18 March 2018 — Election Observation Mission Final Report. Warsaw: ODIHR. osce.org/odihr
Oates, S. (2006). Television, Democracy and Elections in Russia. Routledge, London.
Levada Center. Trust in media institutions and media consumption surveys (annual reviews). levada.ru/en
Reporters Without Borders. Russia Country Profile. Annual World Press Freedom Index. rsf.org/en/russia

IV. Hungary (~75% Ruling-Party SoV / Fidesz)

Mertek Media Monitor (2020–2024). Annual reports on media pluralism and concentration in Hungary. Budapest. mertek.eu
Bátorfy, A. & Urbán, Á. (2020). "State Advertising as an Instrument of Transformation of the Media Market in Hungary." East European Politics, 36(1): 44–65.
Center for Media, Data and Society (CMDS), Central European University (Wien). Media Influence Matrix Hungary reports. cmds.ceu.edu
International Press Institute (IPI). Hungary: Press Freedom Reports (annual). Vienna. ipi.media

V. Turkey (60–65% Ruling-Party SoV / AKP)

P24 Platform for Independent Journalism. Press freedom monitoring reports (2018–2024). Istanbul. platform24.org
Yesil, B. (2016). Media in New Turkey: The Origins of an Authoritarian Neoliberal State. University of Illinois Press.
Reporters Without Borders. Turkey Country Profile. Annual World Press Freedom Index. rsf.org/en/turkey
Freedom House (2024). Freedom of the Press: Turkey Report. Washington DC.

VI. Serbia (55–65% Ruling-Party SoV / SNS)

CRTA (Center for Research, Transparency and Accountability). Citizens on the Watch: Media Monitoring Reports during Election Campaigns, 2020–2024. Belgrade. crta.rs
BIRN (Balkan Investigative Reporting Network). Serbia media reports. birn.eu.com
OSCE/ODIHR (2022, 2023). Republic of Serbia: Parliamentary Elections — Election Observation Mission Final Reports. Warsaw: ODIHR. osce.org/odihr
IREX (2022). Media Sustainability Index: Serbia Country Report. Washington DC. irex.org

VII. Competitive Democracies (45–55%)

Loughborough University Centre for Research in Communication and Culture (CRCC). General Election Media Monitoring Reports (2017, 2019, 2024). UK. lboro.ac.uk/research/crcc
Reuters Institute for the Study of Journalism (vjetore). Digital News Report. University of Oxford. reutersinstitute.politics.ox.ac.uk
Esser, F. & Strömbäck, J. (red.) (2014). Mediatization of Politics: Understanding the Transformation of Western Democracies. Palgrave Macmillan.
Norris, P. (2000). A Virtuous Circle: Political Communications in Postindustrial Societies. Cambridge University Press.

VIII. Methodological Standards (HHI & Monitoring)

U.S. Department of Justice & Federal Trade Commission (2010). Horizontal Merger Guidelines, Section 5.3 (Indeksi Herfindahl-Hirschman). justice.gov/atr
Noam, E. M. (2016). Who Owns the World's Media? Media Concentration and Ownership Around the World. Oxford University Press.
OSCE/ODIHR (revised 2020). Handbook on Media Monitoring for Election Observation Missions. Warsaw.
Council of Europe (2007). Recommendation CM/Rec(2007)15 of the Committee of Ministers to Member States on Measures Concerning Media Coverage of Election Campaigns. Strasbourg.

IX. Albania Data Sources

Buletin Intelligence (2026a). Media Freedom in Albania: RSF vs Empirical Evidence Study, v2. Corpus: 23,839 articles, 52 sources, February–May 2026. Tirana: buletin.al
Buletin Intelligence (2026b). Full-Text SoV Analytical Study (the present study). Corpus: 106,962 articles, 47 sources, January–May 2026.
AMA (Audiovisual Media Authority) (2026). Audiovisual Monitoring of Electoral Campaign Coverage. Tirana. ama.gov.al
Reporters Without Borders (2024). 2024 World Press Freedom Index: Albania ranks 80/180. Paris. rsf.org/en/albania
Freedom House (2024). Nations in Transit 2024: Albania. Classification "Hybrid". Washington DC. freedomhouse.org
OSCE/ODIHR (2025). Republic of Albania: Parliamentary Elections — Needs Assessment Mission Report. Warsaw.

X. Tools and Software

Anthropic Claude API (2026). The model used for Albanian NLP processing, sentiment classification, and report synthesis. anthropic.com
Buletin Intelligence (2026). Big Data & Agentic framework: signal monitoring, NLP, hybrid search, disinformation detection, document intelligence, report synthesis.
Python 3.11 with libraries pandas, numpy, scikit-learn for statistical processing and Herfindahl-Hirschman Index (HHI) computation.
HTML tag stripping is performed with Python regular expressions; conversion to plain text preserves semantic context for keyword matching.

Study Data Provenance

The analysis corpus (106,962 articles) was collected from 47 Albanian online sources through Buletin Intelligence's collection infrastructure (Signal Monitor), for the period 1 January – 1 May 2026. Data is available for join collaboration and scientific inspection per reproducibility standards. Keyword processing, HTML tag stripping, and SoV computations were performed with Python scripts.

IX.
09 / 09

Analytical Synthesis — Scientific Strengthening of the v2 Study

The analysis with expanded methodology confirms the primary v2 findings and refines the secondary ones. The new study is scientifically more robust than the original

Analysis table — primary findings

Finding v2 (titles) Analysis (body+titles) Status
PD vs PS SoV 65.7 / 34.3 68.2 / 31.8 Confirmed & strengthened
Cross-category consistency 66/64/64 64/70/61 Confirmed
Balanced sources mean 63.2% 63.2% Identical
Source standard deviation 8.2 points 4.0 points (-51%) Improved — more uniform
Telegraf Anomaly 40.4% (anomaly) 60.3% (cluster) Resolved methodologically
HHI PD vs PS 3,649 / 2,804 3,286 / 3,678 Reversed — refined
Tone toward government (monthly) n/a (v2) 43–51% critical New finding — supporting
H1 falsification (capture) Yes (35.3 < 65) Yes (31.8 < 65) Confirmed

Scientific synthesis

The full-text analysis confirms and strengthens the v2 findings

The new study with 106,962 articles and full text generates results coherent with the original v2 study: the opposition has substantially higher SoV than the ruling party (68.2% vs 31.8%), cross-category consistency holds (all three source categories give the opposition over 60% of SoV), and within-source variance among balanced outlets is substantially reduced (standard deviation 4.0 vs 8.2).

The HHI finding reversal (PD 3,286 / PS 3,678 in analysis, the opposite in v2) is an honest refinement of a secondary conclusion — both parties now exhibit relatively high concentration, and neither matches the capture structure. The media capture hypothesis (H1, PS > 65%) finds no empirical support in either measurement; the observed value (31.8%) is 33 percentage points below the threshold.

The primary v2 methodological limitation — "titles only" — is now resolved. The conclusions are scientifically more robust than in v2, not weaker. Buletin now has a methodology replicated across 130,801 articles (23,839 + 106,962) that can be presented with academic credibility.

What is fully confirmed

  • SoV direction: PD > PS holds in both methodologies.
  • H1 falsification: neither 35.3 nor 31.8 fall within the range predicted by capture.
  • Cross-category consistency: all three source categories give the opposition >50% of SoV.
  • Balanced-sources mean: 63.2% identical in both measurements.
  • Temporal trajectory: episodic changes, not a control trend.

What is refined

  • Telegraf is not an anomaly: 40.4% → 60.3% with text — converges to the cluster.
  • Variance reduces: ecosystem more uniform than v2 suggested (-51% deviation).
  • HHI reverses: PS 3,678 > PD 3,286 — the opposite of v2, but both above 3,000.
  • Albania's final value: 31.8% (vs 35.3%) — deepens further outside the capture range.

Methodological limitations that remain

What the analysis does not resolve

  • SoV is not favorability: mentions are measured, not stance. Sentiment analysis will require a separate study.
  • Keywords, not NER: automated named-entity recognition would further increase precision. A future study.
  • Online sources only: does not include radio and television. The AMA 2026 study (PD 60% on TV) should be read in parallel.
  • 120-day period: not longitudinal multi-year. Long-term trends require comparable samples from prior years.
  • No causal claim: structure is measured, not process. Does not explain why the opposition has higher SoV.