Note

The data used in this project is derived from a restricted subset of the Semantic Scholar Open Research Corpus (S2ORC). From an original dataset of over 130 million papers, multiple filters were applied to focus on relevant articles:

• Only papers with abstracts and parsable PDFs/LaTeX files (2000–2020) were included.
• Fields like Economics, Medicine, and Sociology were prioritized, excluding others like Biology and Physics.
• A random convenience sample of ~1 million articles was used due to computational limits.

These restrictions mean the dataset does not comprehensively represent global academic research. Additionally, the Natural Language Processing (NLP) methods used for classification may introduce biases or errors. Interpret results with caution.

Project Setup

Clone the project : git clone https://github.com/AbeyHurtis/METAEXPLORE.git
Change directory : cd METAEXPLORE

---

Download the following datasets into project directory

Dataset : results_completed.csv(rename the dataset from results_completed_updated_20231003.csv to results_completed.csv ) [https://datacatalogfiles.worldbank.org/ddh-published/0065200/DR0092801/results_completed_updated_20231003.csv?versionId=2024-05-13T14:24:55.8520476Z]

data_use_country_scores_annual.csv [https://datacatalogfiles.worldbank.org/ddh-published/0065200/DR0092803/data_use_country_scores_annual.csv?versionId=2024-05-13T14:25:01.7623013Z]

---

Install required libraries : pip install requirements.txt

---

Running the project

you can open the stats.ipynb Jupyter notebook
or
you can run the following comman for a non interactive result : python stats.py
project developed on Python 3.11.4

---

Introduction

The Data Use in Academia Dataset is a subset of the Semantic Scholar Open Research Corpus (S2ORC) [@DataUseinAcademiaDataset] [@lo2020], comprising over 130 million English-language academic papers across multiple disciplines, sourced from publishers, open archives like arXiv and PubMed, and web crawls. To enhance usability and relevance, restrictions were applied: only articles with an abstract and a parsed PDF or LaTeX file were included, resulting in approximately 30 million articles. Additionally, only articles published between 2000 and 2020 were considered. Articles from fields less likely to utilize data from national statistical systems, such as Biology and Engineering, were excluded, narrowing the dataset to around 10 million articles. Natural Language Processing (NLP) techniques were employed to extract the countries of study and data usage from the text, utilizing two main approaches: regular expression searches based on ISO3166 country names and Named Entity Recognition (NER) via the spaCy library.

Analysis

Observing the top Journal and amount of papers published.

China's Research Boom and the COVID-19 Disruption

The dataset providees an interesting look at the global distribution of research output over a span of nearly two decades.

	Year	Top Country	Top Country Count	Median	Difference from Median	Percentage Difference
11	2001	USA	229	4	225	5625
18	2002	USA	264	4	260	6500
15	2003	USA	282	4.5	277.5	6166.67
6	2004	USA	433	6.5	426.5	6561.54
12	2005	USA	431	5	426	8520
7	2006	USA	458	7	451	6442.86
2	2007	USA	510	7	503	7185.71
5	2008	USA	543	7	536	7657.14
3	2009	USA	612	8	604	7550
14	2010	USA	657	8	649	8112.5
8	2011	USA	701	11	690	6272.73
13	2012	USA	782	11	771	7009.09
17	2013	USA	790	14.5	775.5	5348.28
0	2014	CHN	897	16	881	5506.25
9	2015	CHN	1083	16	1067	6668.75
4	2016	CHN	1212	16	1196	7475
16	2017	CHN	1346	15	1331	8873.33
1	2018	CHN	1221	13	1208	9292.31
10	2019	CHN	1168	12.5	1155.5	9244
19	2020	CHN	291	4	287	7175

China's Rise to Dominance (Post-2014)

• From 2001 to 2013, the United States consistently published more research papers than any other country, peaking at 790 papers in 2013. During this period, the difference in output between U.S. and other countries remained substantial, with the U.S. having the heighest counts year after year.

• 2014 - The Shift: In 2014, China overtakes the U.S. for the first time, with 897 papers publised compared to the U.S. . This markks the beginning of a clear shift in global research dominance. The differeence between the U.S. and China continues to grow in the following years.

• 2015-2020: China's Contined Growth: By 2015, China's output rises sharply to 1083 papers, overtaking the U.S. . The Gap widens further in 2016 (1,212 papers), 2017 (1,346 papers), and peaks in 2018 with 1,221 papers. The rapid growth in Chaina's research output highlights a clear thrend: China has increasingly become a global research leader, with signigicant investments in science and technology, a investments in science and technology, a growing research community, and a focus on higher education and innovation.

• China's Reseach Growth: China's percentage difference from the median (Percentage increase from the median) has consistently been high, especially post-2014. this indicates thath China has not only produceed more papers but has incresingly outpaced the global median, with a higher concentration of research output compared to other nations.

Imapct of COVID-19(2020)

• 2020: A Drastic Decline: The year 2020 standsd out as an anomaly. while the dataset shows China as still having the higheest count (291 papers), there is a dramatic drop in the total numbers of published papers from 2019 (1168 papers). This suggests a significant impact from the COVID-19 pandamic, which disrupted the global research ecosystem. Travel restrictions, lab clousers, and social distancing measures likely hindered researchers ability to conduct experiments, collaborate internationally, and publish their findings.

• Percentage Drop: The percentage difference from the mediann for China drops dramatically in 2020, reflecting a sharp deviation from the prior years' trends. This sharp decrease may be reflective of the broader global slowdown in academic and research activities, as institutions and researchers worldwide adjusted to realities of the pandemic.

• Global Imapct: The decline in research output in 2020 is not unique to China but is evident accross many countries. The world saw a shift in priorites as pandemic became the new norm. However, the magnitude of the decreasee in publishedd papers underscores the fragility of research prductivity in an unexpected global crises.

Other Interesting Insights

• Median Values: The median numbers of papers published fluctuates across the yearrs, reflecting shifts in global research output. The median starts at alower value in earlier years but gradually increases as the overall research ouput grows globally.

sensitivity

	Above 1000%	0% to 1000%	Below 0%
2001	8	58	62
2002	13	58	71
2003	13	63	76
2004	16	74	90
2005	18	67	77
2006	12	63	83
2007	14	72	88
2008	15	72	84
2009	17	73	94
2010	21	71	91
2011	20	73	95
2012	20	74	95
2013	18	81	99
2014	19	76	95
2015	20	78	97
2016	17	79	96
2017	23	79	103
2018	25	76	97
2019	25	79	104
2020	10	60	66

Income level and Research Output

Postive 1000 percent difference from Median

	Income Level	Region	Count
0	High income	East Asia & Pacific	13800
1	High income	Europe & Central Asia	17963
2	High income	Middle East & North Africa	139
3	High income	North America	16542
4	Low income	Sub-Saharan Africa	335
5	Lower middle income	Middle East & North Africa	3203
6	Lower middle income	South Asia	6586
7	Lower middle income	Sub-Saharan Africa	2334
8	Upper middle income	East Asia & Pacific	13928
9	Upper middle income	Europe & Central Asia	1898
10	Upper middle income	Latin America & Caribbean	4292
11	Upper middle income	Sub-Saharan Africa	2955

Greater than 0 and 1000 percent difference from Median

	Income Level	Region	Count
0	High income	East Asia & Pacific	4547
1	High income	Europe & Central Asia	15816
2	High income	Latin America & Caribbean	853
3	High income	Middle East & North Africa	2400
4	Low income	East Asia & Pacific	17
5	Low income	Middle East & North Africa	182
6	Low income	South Asia	368
7	Low income	Sub-Saharan Africa	2713
8	Lower middle income	East Asia & Pacific	2402
9	Lower middle income	Europe & Central Asia	468
10	Lower middle income	Latin America & Caribbean	252
11	Lower middle income	Middle East & North Africa	2331
12	Lower middle income	South Asia	3928
13	Lower middle income	Sub-Saharan Africa	4643
14	nan	Latin America & Caribbean	100
15	Upper middle income	East Asia & Pacific	3017
16	Upper middle income	Europe & Central Asia	3181
17	Upper middle income	Latin America & Caribbean	4473
18	Upper middle income	Middle East & North Africa	822
19	Upper middle income	Sub-Saharan Africa	343

Below 0 percent difference from Median

	Income Level	Region	Count
0	High income	East Asia & Pacific	212
1	High income	Europe & Central Asia	486
2	High income	Latin America & Caribbean	653
3	High income	Middle East & North Africa	389
4	High income	North America	9
5	High income	Sub-Saharan Africa	34
6	Low income	East Asia & Pacific	91
7	Low income	Middle East & North Africa	144
8	Low income	South Asia	2
9	Low income	Sub-Saharan Africa	893
10	Lower middle income	East Asia & Pacific	394
11	Lower middle income	Europe & Central Asia	143
12	Lower middle income	Latin America & Caribbean	271
13	Lower middle income	Middle East & North Africa	154
14	Lower middle income	South Asia	65
15	Lower middle income	Sub-Saharan Africa	608
16	nan	Latin America & Caribbean	67
17	Upper middle income	East Asia & Pacific	153
18	Upper middle income	Europe & Central Asia	712
19	Upper middle income	Latin America & Caribbean	493
20	Upper middle income	Middle East & North Africa	84
21	Upper middle income	South Asia	25
22	Upper middle income	Sub-Saharan Africa	212

Overview of the Bubble Chart Distribution

The bubble chart provides a clear representation of research output across different income levels and regions. The x-axis orders income levels from low to high, while the y-axis captures the number of published papers. Bubble sizes correspond to the magnitude of publications, making it easy to compare regional research activity at each income level. The visualization highlights both expected patterns, such as the dominance of high-income regions, and unexpected trends, such as the variability within each income group.

Research Output in High-Income Regions

The "Positive 1000 Percent Difference from Median" chart prominently features high-income regions like Europe & Central Asia and North America with the largest bubbles, indicating their leading positions in global research output. Europe & Central Asia, in particular, stands out with an impressive 17,963 papers, positioned at the upper end of the y-axis. However, within high-income regions, there are surprising data points, such as the Middle East & North Africa, which shows a significantly smaller bubble with just 139 papers. This contrast illustrates that while wealth enables higher research productivity, other factors such as research funding, policies, and political stability play crucial roles.

Lower and Middle-Income Regions: Emerging Contributions

South Asia, in particular, displays a sizable bubble, representing 3,928 papers, reflecting significant progress in academic output despite economic constraints. Similarly, Sub-Saharan Africa, with contributions from both low-income and lower-middle-income categories, demonstrates growth in research. These findings suggest that some regions are building robust academic frameworks.

Challenges in Low-Income Regions

low-income South Asia shows the smallest bubble with just 2 papers, emphasizing the significant challenges faced by underdeveloped regions in contributing to global research. This is mirrored in other low-income regions, such as Sub-Saharan Africa, which produces relatively modest outputs. Even within high-income areas, there are outliers, such as North America’s unexpectedly small bubble of 9 papers in this category.

Median Skewness:

The skewness for the median values of the number of papers published by each country varies from around 4.14 to 6.63 across the years. This indicates a positive skewness, meaning that the distribution of the number of papers published by countries is right-skewed in most years.

In simple terms, a positive skew means that in most years, a small number of countries are publishing a disproportionately large number of papers compared to the majority of countries. The larger the skewness value, the more pronounced this effect is, with a few countries contributing a significant proportion of total scientific publications.

Key Observations:

• High Skewness in 2020 (6.63): The year 2020 exhibits the highest skewness, suggesting that a very small number of countries (perhaps due to the pandemic or other factors) contributed most of the research output. This might reflect global shifts in scientific focus or funding during that period.
• Stable Skewness (around 5): Most years have skewness values clustered around 5, indicating a similar distribution across multiple years. This pattern suggests that in general, the global scientific output has been concentrated in the hands of a few countries.

Average Skewness:

The skewness for the average number of papers published by each country is somewhat lower, ranging from around 4.09 to 4.75. While this is still a positive skew, it is less extreme compared to the skewness of the median. The skewness for the average papers published is consistently in the 4.x range, which suggests a more balanced distribution but still highlights the presence of outliers in some years.

Key Observations:

• Consistent but Lower Skewness: The average skewness values are generally lower than the median skewness, reflecting that the calculation of the average dampens the influence of extreme outliers. The average can be affected by very high values, but not as much as the median, which is more resistant to extreme values.
• Gradual Decrease in Skewness: There is a slight decrease in skewness from 4.75 (2020) to 3.95 (2001), indicating that the concentration of scientific output among a few countries might have been slightly more pronounced in the later years. The skewness decrease towards earlier years suggests a somewhat more even distribution in scientific publications.

Comparing Median and Average Skewness:

• The median skewness is consistently higher than the average skewness across almost all years. This is expected because the median, being a measure of central tendency, is more robust to extreme outliers, whereas the average can be heavily influenced by a small number of countries publishing a large number of papers.
• The larger values of skewness in the median data show that, even when outliers are accounted for, the global scientific community is characterized by a few dominant countries, particularly in recent years.

Interpretation:

Impact of Outliers: The skewness of the median being higher than that of the average suggests that extreme outliers (countries with very high paper counts) have a stronger impact on the distribution when considering the median. This could indicate that a few countries dominate scientific publications in each year.

Global Trends: The increase in skewness values over time, particularly in the last few years (with 2020 being the highest), indicates a growing gap between countries in terms of research productivity. ##Error Bars:

• The error bars (in orange) represent the standard deviation, which shows how spread out the number of papers is across countries.

• A higher standard deviation means more variation in the number of papers published across countries for that year.

• The standard deviation starts relatively low (~24), increases significantly (~130), and fluctuates in the later years, reflecting growing disparities among countries.

• The median generally increases over time, which might suggest a global upward trend in paper production.

• The standard deviation also grows, indicating that while some countries are contributing more papers, disparities between countries are also becoming more pronounced.

References

[1.] Data Use in Academia Stacy, B., Kitzmüller, L., Wang, X., Mahler, D. G., & Serajuddin, U. (2023). Data Use in Academia. Accessed: Oct 9, 2024.

[2.] S2ORC: The Semantic Scholar Open Research Corpus Lo, K., Wang, L. L., Neumann, M., Kinney, R., & Weld, D. S. (2020). S2ORC: The Semantic Scholar Open Research Corpus. arXiv:1911.02782.