Abstract
This analysis examines the impact of AI adoption in enterprises on
employment rates and economic performance in European Union (EU)
countries, recognizing the significant implications of AI for economic
structures and employment dynamics. The primary objective is to
understand how higher AI adoption in enterprises in EU countries affects
the employment rate of people aged 20–64, assuming that AI technologies
complement human labour and create new employment opportunities. A
secondary objective is to investigate whether this relationship exhibits
gender-specific effects, hypothesizing a stronger positive impact on
female employment rates compared to male. The research addresses the
urgent economic problem of managing the disruptive consequences of AI
while harnessing its benefits, especially given the uncertainty
surrounding its full impact and transitional challenges. The methodology
employs a general-to-specific modelling approach using a panel dataset
covering the 27 EU countries from 2021–2024. Regarding the main
hypothesis (H1), the analysis established a robust modelling framework
for assessing AI’s impact on overall employment, consistently
identifying panel and random effects. For the secondary hypothesis (H2),
the results are mixed and very nuanced, indicating a complex and
sector-specific relationship between AI adoption and gender-specific
employment rates. While some sectors, such as water supply, sewerage,
waste management, and information and communication showed a stronger
positive and statistically significant effect of AI adoption on female
employment rates, supporting H2, other sectors contradicted this.
Specifically, the AI hiring index, manufacturing, and professional,
scientific and technical activities showed a negative effect, which was
stronger for female employment. These results underline the need for
targeted policies that address gender-specific challenges and employment
opportunities arising from the adoption of AI across sectors.
Introduction
Analysing AI adoption and its impact on labour market outcomes is
critical because of its implications for economic structures and
employment dynamics across sectors. AI technologies have introduced
significant disparities in job creation and replacement, as some
industries experience automation that makes some roles obsolete while
creating new opportunities in emerging fields. Understanding these
changes is essential for policymakers to develop strategies to mitigate
negative impacts, particularly in low-skilled sectors where job losses
are most evident. Also, integrating AI has potential to increase
productivity and economic growth, although it comes with the risk of
increasing inequality. As companies across sectors, including healthcare
and finance, increasingly adopt AI, analysing sector-specific challenges
and successes becomes crucial to supporting effective workforce
transitions and reskilling initiatives. Therefore, this topic presents
urgent economic problem that requires multifaceted solutions to harness
benefits of AI while addressing its disruptive consequences.
[1,2,3,4,5]
Description Of the
Problem
The problem of sectoral AI adoption and its impact on employment and
economic output in the European Union is multi-faceted due to
uncertainty about the full scope of AI’s impact, significant
transitional challenges, and the complexity of developing adaptive
policy frameworks and supporting social acceptance.
Key aspects of this problem include:
- Uncertain and dual impacts on employment. AI revolution is in its
early stages and is still evolving, presenting both the potential to
replace jobs and create new employment opportunities. There is no clear
consensus in the existing literature on whether AI will ultimately lead
to net job losses or gains. Sectors traditionally dependent on manual
labour are identified as most vulnerable to job replacement. [1,2]
- Transitional challenges and skills gaps. Current phase of AI
adoption is characterized by transitional challenges, where AI-driven
economic growth does not immediately translate into widespread job
creation. A key issue is the need for adaptive policies and education
systems to prepare the workforce for these changes. Therefore, effective
reskilling and upskilling programs are key interventions to mitigate the
negative impacts and equip workers for new AI-enabled roles. [2,4]
- Regulatory and policy ambiguity in the EU context. The European
Union faces particular challenges in establishing a comprehensive AI
policy framework. There are significant geographical differences in
regulatory approaches to AI around the world. A key issue is ensuring
“privacy by design” in the technological infrastructure and clearly
defining responsibilities, liabilities and culpability among human and
AI stakeholders. [5]
- Impact on economic outcomes and the potential for increasing
inequality. While AI offers significant potential for increasing
productivity, there is a significant risk of deepening existing
inequalities if the economic benefits of AI are not shared widely across
society. [2,4,5]
- Implementation barriers and trust. In addition to policy and skills,
there are fundamental barriers to practical AI implementation. “Lack of
integration” with existing systems, difficulties with “information
sharing,” and the broader need to “promote standards” and “build
capacity” in the AI ecosystem. [2,5]
Research
Hypotheses
Based on the literature the following hypotheses are worth
testing.
Main Hypothesis (H1): Higher levels of enterprise AI
adoption in EU countries lead to increased employment rates in the 20-64
age group, as AI technologies complement human labour and create new job
opportunities rather than displacing workers.
Secondary Hypothesis (H2): The relationship between
AI adoption and employment exhibits gender-specific effects, with AI
adoption having a stronger positive impact on female employment rates
compared to male employment rates in EU countries.
Literature Review
Main Hypothesis
(H1)
The core of this hypothesis addresses the dual possibilities
presented by AI: job displacement versus the creation of new employment
opportunities [1,2]. Several sources support the potential for AI to
lead to net job creation or a positive correlation with employment
rates:
- Empirical analysis, specifically regression analysis of data from
2012 to 2022, suggests that AI development generally has a negative
correlation with unemployment rates, implying that increased AI adoption
can be linked to lower unemployment levels [1].
- The popularity of the term “Artificial Intelligence”, as measured by
Google Trends data, correlates with job creation. This indicates that
high public interest can drive investments in AI-driven enterprises,
fostering job growth [1].
- AI integration has the potential to enhance productivity and
stimulate economic growth. This broader economic benefit can contribute
to overall employment growth by fostering a more dynamic and expanding
economy [1,2,3].
- Specific industry implications show increased job creation in the
information technology and consumer goods sectors, illustrating a clear
trend of job growth in industries with rapid AI adoption. This suggests
that AI can act as a catalyst for job creation in specific domains
[1,5].
- In developed economies where AI adoption is higher, there is
evidence of falling unemployment rates, suggesting that AI can generate
new roles as quickly as it makes old ones obsolete. This view is
consistent with the idea that AI can complement human work rather than
completely replace it [1,2].
- While AI adoption and innovation may initially pose challenges
leading to job losses, strategic AI implementation significantly
mitigates these negative effects and improves employment conditions in
the long term. This suggests that this relationship is complex and
dependent on policy and strategic choices [2].
Secondary Hypothesis
(H2)
Despite that cited sources do not provide direct findings supporting
a stronger positive impact on female employment, they indicate that the
gendered impact of AI adoption on employment is a key area of active
research and analysis. Few sources address the intersection of
AI/automation and gender in the labour market. Discussions include
gender as a potential dimension of inequalities, requiring research on
whether the benefits of AI are broadly distributed across socioeconomic
groups [1,2,3]. Thus, considering gender as a variable affecting the
labour market in the context of AI makes this a valid hypothesis.
Data
Datasets
Overview
The analysis combines four datasets covering EU countries for
2021-2024:
- AI
Hiring Index Data(data.csv):
- Source: OECD.AI – LinkedIn-based AI hiring trends
- Coverage: 47 countries, monthly data for years 2018-2025 → monthly
data for years 2021-2024 transformed to annual means
- Selected variables: Country, ref_date,
Relative_AI_hiring_index_pct
- Key variable: AI hiring intensity index – ratio of AI talent hiring
relative to overall hiring of LinkedIn members per country 12-month
moving average going back from December
- EU
Employment Data (estat_lfsi_emp_a_filtered_en.csv):
- Source: Eurostat – Labour Force Survey
- Coverage: 27 EU countries, annual 2021-2024
- Selected variables: sex, geo, TIME_PERIOD, OBS_VALUE
- Key variable: employment percentage of total population
- Labour
productivity and unit labour costs by industry
(estat_nama_10_lp_a21_filtered_en.csv):
- Source: Eurostat
- Coverage: 27 EU countries, annual 2021-2024
- Selected variables: nace_r2, geo, TIME_PERIOD, OBS_VALUE
- Key variable: real labour productivity per hour worked (percentage
change on previous period)
- EU
GDP Data (estat_namq_10_gdp_filtered_en-3.csv):
- Source: Eurostat – National Accounts
- Coverage: EU countries, quarterly data for years 2021-2024 →
quarterly data for years 2021-2024 transformed to annual means
- Selected variables: geo, TIME_PERIOD, OBS_VALUE
- Key variable: contribution to GDP growth in percentage point change
compared to same period in previous year
- EU
AI Adoption Data
(estat_isoc_eb_ain2_filtered_en-6.csv):
Source: Eurostat – Information Society Statistics
Coverage: 27 EU countries, annual data for years 2021-2024, 9
sectors, 2 information society indicators (actually 3 but 2 were
combined into 1 – description below)
Selected variables: nace_r2, indic_is, geo, TIME_PERIOD,
OBS_VALUE
Key variable: percentage of enterprises
Information society indicators E_AI_PANY (data only for 2021) and
E_AI_1PANY (data only for 2023-2024) combined into 1 variable E_AI_PANY
(due to majority of the same purposes covered by both
indicators; the rest indicators which are not included in the
other dataset contain data in the periods as E_AI_PANY or
E_AI_1PANY):
- E_AI_PANY – enterprises using AI technologies for at least one of
the purposes:
- E_AI_PMS – marketing or sales
- E_AI_PPP – production processes
- E_AI_PBA – for organisation of business administration
processes
- E_AI_PME – for management of enterprises
- E_AI_PLOG – logistics
- E_AI_PITS – ICT security
- E_AI_PHR – for human resources management or recruiting
- E_AI_1PANY – enterprises using AI technologies for at least one of
the purposes:
- E_AI_PMS – marketing or sales
- E_AI_PPP – production processes
- E_AI_PBAM – organisation of business administration
processes or management
- E_AI_PLOG – logistics
- E_AI_PITS – ICT security
- E_AI_PFIN – accounting, controlling or finance management
- E_AI_PRDI – research and development (R&D) or innovation
activity
E_AI_TANY – enterprises use at least one of the AI
technologies:
- AI_TTM – performing analysis of written language (text mining)
- AI_TSR – converting spoken language into machine-readable format
(speech recognition)
- AI_TNLG – generating written or spoken language (natural language
generation)
- AI_TIR – identifying objects or persons based on images (image
recognition, image processing)
- AI_TML – use machine learning (e.g. deep learning) for data
analysis
- AI_TPA – use AI technologies automating different workflows or
assisting in decision making (AI based software robotic process
automation)
- AI_TAR – use AI technologies enabling physical movement of machines
via autonomous decisions based on observation of surroundings
(autonomous robots, self-driving vehicles, autonomous drones)
Processing Steps
Applied
- Time alignment – restricted analysis to 2021-2024 period where all
datasets overlap
- Country matching – excluded non-EU countries from AI hiring
data
- Calculate mean of 2021 and 2023 values for missing AI adoption data
for 2022
- Adding mean for each indicator_sector variable
- Join all tables to cover 27 EU countries, annual 2021-2024
dataset
AI hiring
data
Employment
data
## Employment - Original dimensions: 405 rows, 12 columns
## Employment - Preprocessed dimensions: 108 rows, 6 columns
## # A tibble: 108 × 6
## country year emp_rate_female emp_rate_male emp_rate_Total emp_gender_gap
## <chr> <int> <dbl> <dbl> <dbl> <dbl>
## 1 Austria 2021 71.3 79.9 75.6 8.60
## 2 Austria 2022 73.4 81.2 77.3 7.8
## 3 Austria 2023 73.3 81.1 77.2 7.8
## 4 Austria 2024 73.9 80.8 77.4 6.90
## 5 Belgium 2021 66.8 74.5 70.6 7.7
## 6 Belgium 2022 68.1 75.7 71.9 7.60
## 7 Belgium 2023 68.3 75.9 72.1 7.60
## 8 Belgium 2024 68.3 76.3 72.3 8
## 9 Bulgaria 2021 69.1 77.5 73.3 8.40
## 10 Bulgaria 2022 72.1 79.6 75.9 7.5
## # ℹ 98 more rows
## Unique countries in employment dataset: 27
Labour
productivity per hour worked
## Productivity - Original dimensions: 1360 rows, 21 columns
## Productivity - Preprocessed dimensions: 108 rows, 17 columns
## # A tibble: 108 × 17
## country year pc_A pc_B pc_C pc_D pc_E pc_F pc_G pc_H pc_I pc_J
## <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 Austria 2021 5.8 24.3 5.6 -4.3 -9.7 -11 -3.1 -4.6 -19.3 2
## 2 Austria 2022 9.7 -12.6 5.1 7.1 -2.5 -3.7 -3.2 10.8 36.6 1.9
## 3 Austria 2023 6.1 -28.6 -2.6 5.8 -0.7 -6.9 -5.7 -2.4 2 -2.1
## 4 Austria 2024 5.8 NA -4.3 NA NA -1.7 NA NA NA 1.9
## 5 Belgium 2021 -0.3 4.6 -10.5 -10.8 -0.2 -2.1 4.1 -0.8 4.4 1.9
## 6 Belgium 2022 1.8 -9.7 9.9 -32.3 1 -4.4 2.9 -1.5 -8.9 0.5
## 7 Belgium 2023 2.7 -9.1 -1.1 11.6 -0.7 -1.5 0.5 -1 6.1 0.3
## 8 Belgium 2024 3.5 NA 0.3 NA NA 0.9 NA NA NA 3.8
## 9 Bulgaria 2021 38.6 -13.5 0.3 41 -21.1 -12.2 9.4 6.4 43.7 -2.2
## 10 Bulgaria 2022 -3.5 -11.2 26.6 -8.3 20.4 7.2 -7.6 10.3 11.2 -4
## # ℹ 98 more rows
## # ℹ 5 more variables: pc_K <dbl>, pc_M <dbl>, pc_N <dbl>, pc_R <dbl>,
## # pc_S <dbl>
## Unique countries in productivity dataset: 27
GDP data
## GDP - Original dimensions: 400 rows, 11 columns
## GDP - Preprocessed dimensions: 100 rows, 3 columns
## # A tibble: 100 × 3
## country year gdp_growth
## <chr> <dbl> <dbl>
## 1 Austria 2021 5.28
## 2 Austria 2022 5.44
## 3 Austria 2023 -0.835
## 4 Austria 2024 -1.34
## 5 Belgium 2021 6.43
## 6 Belgium 2022 4.28
## 7 Belgium 2023 1.21
## 8 Belgium 2024 1.02
## 9 Bulgaria 2021 7.42
## 10 Bulgaria 2022 4.13
## # ℹ 90 more rows
## Unique countries in gdp dataset: 25
AI adoption
data
## AI Adoption - Original dimensions: 1458 rows, 23 columns
## AI Adoption - Preprocessed dimensions: 108 rows, 20 columns
## # A tibble: 108 × 20
## country year AI_PANY_C AI_TANY_C AI_PANY_E AI_TANY_E AI_PANY_F AI_TANY_F
## <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 Austria 2021 9.06 9.61 5.33 5.33 2.2 3.12
## 2 Austria 2022 10.3 11.0 6.24 6.24 3.16 3.7
## 3 Austria 2023 11.6 12.3 7.16 7.16 4.11 4.28
## 4 Austria 2024 21.9 22.7 17.8 17.8 7.19 7.35
## 5 Belgium 2021 7.46 10.4 4.66 8.83 4.81 8.3
## 6 Belgium 2022 9.59 12.9 NA NA 4.32 6.74
## 7 Belgium 2023 11.7 15.3 NA NA 3.83 5.17
## 8 Belgium 2024 18.0 23.2 NA NA 8.25 11.1
## 9 Bulgaria 2021 1.87 2.88 NA NA NA 1.47
## 10 Bulgaria 2022 1.70 2.72 NA NA NA NA
## # ℹ 98 more rows
## # ℹ 12 more variables: AI_PANY_G <dbl>, AI_TANY_G <dbl>, AI_PANY_H <dbl>,
## # AI_TANY_H <dbl>, AI_PANY_I <dbl>, AI_TANY_I <dbl>, AI_PANY_J <dbl>,
## # AI_TANY_J <dbl>, AI_PANY_M <dbl>, AI_TANY_M <dbl>, AI_PANY_N <dbl>,
## # AI_TANY_N <dbl>
## Unique countries in ai_adoption dataset: 27
Panel Structure
This chapter provides the overview of panel dataset and includes
visualizations to illustrate key aspects of the data. Table 4.2-1
describe structure of dataset, which covers 27 EU countries over four
years, from 2021 to 2024. The dataset comprises 108 total observations.
The analysis includes data for 9 sectors represented by 2 indicators:
E_AI_PANY and E_AI_TANY. The dataset also incorporates other variables
such as Productivity in sectors, GDP growth, AI_PANY_MEAN, and
AI_TANY_MEAN. In total, there are 41 variables in the dataset (Table
4.2-2). The data completeness stands at 91.1%, with 4034 valid
observations out of 4428 total, indicating 394 missing values (Table
4.2-3), which results in an unbalanced panel.
TABLE 4.2-1 PANEL STRUCTURE
| Total Observations |
108 |
| Countries |
27 EU countries: AT, BE, BG, CY, CZ, DE, DK, EE, EL, ES, FI, FR, HR,
HU, IE, IT, LT, LU, LV, MT, NL, PL, PT, RO, SE, SI, SK |
| Time Coverage |
2021-2024 (4 years) |
Sectors
(NACE Rev. 2) |
9 sectors:
- C: Manufacturing
- E: Water supply; sewerage, waste management
- F: Construction
- G: Wholesale and retail trade
- H: Transportation and storage
- I: Accommodation and food service activities
- J: Information and communication
- M: Professional, scientific and technical activities
- N: Administrative and support service activities
|
| Indicators |
2 indicators:
|
| Additional variables |
Productivity in sectors, GDP growth, AI_PANY_MEAN, AI_TANY_MEAN |
| Number of variables |
41 |
| Data Completeness |
91.1% (4034 valid / 4428 total; 394 missing) – unbalanced |
TABLE 4.2-2 Panel Data Statistics
##
## ==========================================================
## Statistic N Mean St. Dev. Min Max
## ----------------------------------------------------------
## emp_rate_female 108 71.827 6.575 52.700 80.900
## emp_rate_male 108 80.689 3.999 72.400 89.000
## emp_rate_total 108 76.289 4.853 62.600 83.500
## emp_gender_gap 108 8.862 5.038 0.200 21.000
## ai_hiring_index 96 9.830 7.433 -5.671 35.545
## ai_pany_c 108 6.678 4.341 0.990 21.900
## ai_tany_c 108 8.889 5.817 1.280 27.280
## ai_pany_e 76 4.389 4.312 0.000 17.770
## ai_tany_e 81 6.597 5.832 0.000 24.770
## ai_pany_f 99 2.878 2.085 0.020 10.100
## ai_tany_f 106 4.496 3.478 0.050 19.080
## ai_pany_g 105 6.179 4.440 0.540 20.780
## ai_tany_g 108 8.190 5.677 0.600 27.340
## ai_pany_h 104 5.509 3.674 0.210 20.540
## ai_tany_h 108 7.324 4.846 0.310 23.030
## ai_pany_i 101 3.035 2.641 0.000 15.580
## ai_tany_i 106 4.299 3.397 0.060 16.850
## ai_pany_j 106 26.520 12.328 7.710 59.150
## ai_tany_j 104 31.850 14.219 8.120 68.300
## ai_pany_m 105 13.591 8.349 1.820 38.700
## ai_tany_m 105 17.835 10.909 3.670 53.710
## ai_pany_n 100 7.412 4.399 1.010 21.460
## ai_tany_n 108 9.566 5.811 1.570 27.840
## pc_a 108 -102.802 1,085.097 -11,273.700 94.800
## pc_b 78 -1.006 25.538 -76.800 86.400
## pc_c 108 1.303 6.821 -21.800 26.600
## pc_d 78 -5.546 28.015 -56.300 102.300
## pc_e 82 0.271 11.698 -25.900 26.900
## pc_f 108 -1.133 7.835 -26.600 31.600
## pc_g 82 0.190 7.485 -15.400 22.300
## pc_h 82 4.339 21.804 -29.900 185.800
## pc_i 82 10.449 19.608 -30.300 85.400
## pc_j 108 1.615 6.852 -17.800 26.100
## pc_k 108 2.006 10.508 -24.600 44.200
## pc_m 82 2.094 7.562 -12.600 35.000
## pc_n 82 3.172 7.750 -14.100 29.200
## pc_r 82 4.790 15.171 -49.300 61.800
## pc_s 82 2.982 23.106 -22.800 193.800
## gdp_growth 100 3.588 3.525 -5.572 15.875
## ai_pany_mean 108 8.867 4.846 1.881 25.760
## ai_tany_mean 108 11.083 6.034 2.296 28.244
## ----------------------------------------------------------
TABLE 4.2-3 Missing Data Summary
##
## Missing Data Summary
## ================================
## variable missing_count
## --------------------------------
## 1 ai_hiring_index 12
## 2 ai_pany_e 32
## 3 ai_tany_e 27
## 4 ai_pany_f 9
## 5 ai_tany_f 2
## 6 ai_pany_g 3
## 7 ai_pany_h 4
## 8 ai_pany_i 7
## 9 ai_tany_i 2
## 10 ai_pany_j 2
## 11 ai_tany_j 4
## 12 ai_pany_m 3
## 13 ai_tany_m 3
## 14 ai_pany_n 8
## 15 pc_b 30
## 16 pc_d 30
## 17 pc_e 26
## 18 pc_g 26
## 19 pc_h 26
## 20 pc_i 26
## 21 pc_m 26
## 22 pc_n 26
## 23 pc_r 26
## 24 pc_s 26
## 25 gdp_growth 8
## --------------------------------
Visualisations
Figure 1 illustrates average total employment rate by country from
2021-2024. On the Figure 2 we can see gender gap and on Figure 3 average
employment rates: total and by gender for each year from panel
dataset.
FIGURE 1 Average Total Employment Rate by Country (2021-2024)
FIGURE 2 Average Gender Employment Gap by Country (2021-2024)
FIGURE 3 Average Employment Rates by Gender (2021-2024)
On the Figure 4 there is average hiring index for each year. While to
2023 it was dropping, it grew in 2024.
FIGURE 4 Average AI Hiring Index by Year
Figure 5 depict average AI adoption rates for both indicators: PANY
and TANY. Both rates grow over time.
FIGURE 5 Average AI Adoption: PANY vs TANY
On the last figures 6 and 7 there are presented average AI adoption
rates in enterprises by sectors. As we can see the highest rates occur
for sectors J: information and communication, and N: administrative and
support service activities. The lowest rates are noted for sectors F:
construction, and I: accommodation and food service activities.
FIGURE 6 Average AI Adoption (PANY) by Sector
FIGURE 7 Average AI Adoption (TANY) by Sector
Method/Model
This chapter describes all steps performed during the analysis of
panel dataset.
- General to specific approach was performed starting
with a comprehensive specification that includes all potential
predictors.
- Model selection through sequential elimination was
performed. This process is based on Wald tests, which are used to ensure
the validity of restrictions at each step of variable removal.
- Hypothesis testing with different models were
tested for both the Main Hypothesis (H1) and the Secondary Hypothesis
(H2).
- For Main Hypothesis (H1), which posits that higher AI adoption leads
to increased employment rates in the 20-64 age group, various models
(Model 1 to Model 6) were evaluated, including different combinations of
AI adoption indicators (e.g., ai_pany_sectors, ai_tany_sectors) and
their interactions with the ai_hiring_index.
- For Secondary Hypothesis (H2), which suggests gender-specific
employment effects with a stronger positive impact on female employment,
models were specifically run with emp_rate_female and emp_rate_male as
dependent variables, incorporating AI adoption indicators based on
purpose (ai_pany_sectors) and technology (ai_tany_sectors).
- Hausman tests were performed to determine the
appropriate estimator for the panel data, specifically to choose between
random effects (RE) or fixed effects (FE) models.
- Diagnostic tests were performed for the final
models:
- F-test for individual effects (FE vs Pooled): used to determine if
panel effects exist.
- Breusch-Pagan test (RE vs Pooled): used to determine if random
effects exist.
- Breusch-Godfrey/Wooldridge test: used to check for autocorrelation
in the residuals.
- Studentized Breusch-Pagan test: used to check for heteroskedasticity
in the residuals.
- Pesaran CD test: used to assess cross-sectional dependence, with
robust standard errors applied (clustered) when dependence was
present.
- Variable significance selection process identified
insignificant variables at both 5% and 10% significance levels and their
potential removal based on Wald tests.
The analysis aimed to progress from more general to specific models,
ensuring that only statistically significant predictors were
retained.
Results
The analysis of AI adoption’s effects on employment and economic
output in the EU presents detailed results for both the main and
secondary hypotheses, derived through a general-to-specific modelling
approach. The general model includes AI adoption rate in all sectors and
interactions of mean from all sectoral AI adoption rate and AI hiring
index. Systematic variable elimination was validated by Wald tests and
lead to preferred specification, which retains only statistically
significant variables.
Main Hypothesis
(H1)
Across the six models tested for H1 (Model 1 to Model 6), the Hausman
tests primarily indicated the use of Random Effects (RE) models. Panel
effects were found to exist across all models, as indicated by a p-value
= 0 from the F-test for individual effects. Random effects were
confirmed to exist across all models, with a p-value = 0 from the
Breusch-Pagan test. Most models (Models 1-4) showed no autocorrelation,
but Models 5 and 6 indicated the presence of autocorrelation.
Heteroskedasticity was present in Models 1, 2, 3, and 6, while Model 4
showed homoskedasticity, and Model 5 was close to homoskedasticity
(p=0.0505). Because of cross-sectional dependence for all models, robust
standard errors (clustered) were applied. Process identified
insignificant variables at 5% and 10% significance levels. However, at
the first step, ai_hiring_index was eliminated. Next remaining variables
which were insignificant at 10% were eliminated. Model 3 and 6 have all
significant variables.
Table 6.1‑1 SUMMARY OF MODEL RESULTS (DEPENDENT VARIABLE:
EMP_RATE_TOTAL)
| Formula |
emp_rate_total ~ ai_hiring_index + ai_pany_c +
ai_pany_e + ai_pany_f + ai_pany_g + ai_pany_h + ai_pany_i + ai_pany_j +
ai_pany_m + ai_pany_n + ai_pany_mean:ai_hiring_index |
emp_rate_total ~ ai_pany_c + ai_pany_e + ai_pany_f
+ ai_pany_g + ai_pany_h + ai_pany_i +
ai_pany_j + ai_pany_m + ai_pany_n + ai_pany_mean:ai_hiring_index |
emp_rate_total ~ ai_pany_c + ai_pany_e + ai_pany_g + ai_pany_j +
ai_pany_m + ai_pany_n + ai_pany_mean:ai_hiring_index |
emp_rate_total ~ ai_hiring_index + ai_tany_c +
ai_tany_e + ai_tany_f + ai_tany_g + ai_tany_h + ai_tany_i + ai_tany_j +
ai_tany_m + ai_tany_n + ai_tany_mean:ai_hiring_index |
emp_rate_total ~ ai_tany_c + ai_tany_e + ai_tany_f
+ ai_tany_g + ai_tany_h + ai_tany_i +
ai_tany_j + ai_tany_m + ai_tany_n + ai_tany_mean:ai_hiring_index |
emp_rate_total ~ ai_tany_c + ai_tany_e + ai_tany_g + ai_tany_j +
ai_tany_m + ai_tany_n + ai_tany_mean:ai_hiring_index |
| Model Type |
RE |
RE |
RE |
FE |
RE |
RE |
| F-test for individual effects FE vs Pooled |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
| Breusch-Pagan test RE vs Pooled |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
| Hausman test RE vs FE |
p = 0.6490 |
p = 0.4770 |
p = 0.2552 |
p = 0.0002* |
p = 0.9998 |
p = 0.7014 |
| Interpretation |
Use Random Effects |
Use Random Effects |
Use Random Effects |
Use Fixed Effects |
Use Random Effects |
Use Random Effects |
| Wald test for variable removal |
|
p = 0.4108* |
p = 0.4108* |
|
p = 0.8869* |
p = 0.8869* |
| Interpretation |
|
Removal justified |
Removal justified |
|
Removal justified |
Removal justified |
| Breusch-Godfrey/Wooldridge test Serial
Correlation |
p = 0.1800 |
p = 0.1913 |
p = 0.1648 |
p = 0.7724 |
p = 0.0213* |
p = 0.0352* |
| Interpretation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
Autocorrelation present |
Autocorrelation present |
| Studentized Breusch-Pagan test
Heteroskedasticity |
p = 0.0482* |
p = 0.0114* |
p = 0.0230* |
p = 0.2088 |
p = 0.0505 |
p = 0.0162* |
| Interpretation |
Heteroskedasticity present |
Heteroskedasticity present |
Heteroskedasticity present |
Homoskedasticity |
Homoskedasticity |
Heteroskedasticity present |
| Pesaran CD test Cross-sectional Dependence |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
| Insignificant variables 5% |
ai_hiring_index, ai_pany_f, ai_pany_h, ai_pany_i |
ai_pany_f, ai_pany_h, ai_pany_i |
All significant |
ai_hiring_index, ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i,
ai_tany_n, ai_hiring_index:ai_tany_mean |
ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i, ai_tany_n |
ai_tany_n |
| Insignificant variables 10% |
ai_hiring_index, ai_pany_f, ai_pany_h, ai_pany_i |
ai_pany_f, ai_pany_h, ai_pany_i |
All significant |
ai_hiring_index, ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i,
ai_hiring_index:ai_tany_mean |
ai_tany_f, ai_tany_h, ai_tany_i, ai_tany_n |
All significant |
##
## Model Comparison: General to Specific
## ========================================================================================
## Dependent variable:
## -----------------------------------------------------------
## emp
## Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
## (1) (2) (3) (4) (5) (6)
## ----------------------------------------------------------------------------------------
## ai_hiring_index -0.017 -0.035
## (0.020) (0.025)
## ai_pany_c -0.464*** -0.480*** -0.416***
## (0.148) (0.151) (0.115)
## ai_pany_e 0.170** 0.170** 0.166**
## (0.070) (0.069) (0.072)
## ai_pany_f 0.231 0.241
## (0.223) (0.222)
## ai_pany_g 0.380*** 0.373*** 0.458***
## (0.119) (0.120) (0.124)
## ai_pany_h -0.047 -0.027
## (0.195) (0.195)
## ai_pany_i 0.032 0.056
## (0.121) (0.118)
## ai_pany_j 0.105*** 0.112*** 0.101***
## (0.032) (0.030) (0.027)
## ai_pany_m -0.227*** -0.229*** -0.185***
## (0.066) (0.067) (0.054)
## ai_pany_n 0.229*** 0.236*** 0.199***
## (0.074) (0.073) (0.069)
## ai_hiring_index:ai_pany_mean -0.008**
## (0.004)
## ai_pany_mean:ai_hiring_index -0.011*** -0.010***
## (0.003) (0.002)
## ai_tany_c -0.298*** -0.282*** -0.252***
## (0.094) (0.092) (0.082)
## ai_tany_e 0.232*** 0.207*** 0.215***
## (0.069) (0.076) (0.078)
## ai_tany_f 0.133 0.073
## (0.143) (0.142)
## ai_tany_g 0.119 0.154* 0.191**
## (0.093) (0.090) (0.077)
## ai_tany_h -0.091 0.010
## (0.111) (0.112)
## ai_tany_i 0.076 0.052
## (0.100) (0.094)
## ai_tany_j 0.096*** 0.098*** 0.097***
## (0.026) (0.026) (0.027)
## ai_tany_m -0.121*** -0.106*** -0.092***
## (0.032) (0.038) (0.035)
## ai_tany_n 0.099** 0.086* 0.070*
## (0.046) (0.052) (0.041)
## ai_hiring_index:ai_tany_mean -0.005
## (0.003)
## ai_tany_mean:ai_hiring_index -0.009*** -0.008***
## (0.002) (0.002)
## Constant 74.834*** 74.622*** 74.289*** 74.187*** 74.046***
## (1.404) (1.373) (1.140) (1.305) (1.189)
## ----------------------------------------------------------------------------------------
## Observations 68 68 68 73 73 73
## R2 0.831 0.827 0.833 0.635 0.817 0.815
## Adjusted R2 0.798 0.797 0.814 0.343 0.788 0.795
## ========================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
## Robust standard errors in parentheses
## Model 1 : Random Effects
## Model 2 : Random Effects
## Model 3 : Random Effects
## Model 4 : Fixed Effects
## Model 5 : Random Effects
## Model 6 : Random Effects
The analysis of the variable coefficients in Model 3 and 6 provides
the following insights (Table 6.1-2).
Variables with a positive and significant impact on the total
employment rate (emp_rate_total), supporting H1:
- Water supply; sewerage, waste management (ai_pany_e) for both showed
a positive and statistically significant impact, for ai_tany_e (0.215),
for ai_pany_e (0.166). This suggests that higher adoption of AI in
sector E is associated with an increase in employment rates.
- Wholesale and retail trade (ai_pany_g) for both displayed a positive
and statistically significant impact, for ai_pany_g (0.458), for
ai_tany_g (0.191). This indicates a strong positive association between
AI adoption in sector G and increased employment rates.
- Information and communication (ai_pany_j) have statistically
significant positive impact, similar for both models ai_pany_j (0.101)
and ai_tany_j (0.097). This also indicates a positive relationship with
higher employment rates.
- Administrative and support service activities (ai_pany_n) for both
models shows a positive association with increased employment rates for
ai_pany_j (0.199) and ai_tany_j (0.070).
These findings support the part of Hypothesis H1 that AI complement
human labour and create new job opportunities, leading to increased
employment rates.
Variables with a negative and significant impact on the total
employment rate (emp_rate_total), contradicting H1:
- Manufacturing for ai_pany_c (-0.416) and ai_tany_c (-0.252) indicate
that higher AI adoption in sector C is associated with a decrease in
employment rates.
- AI in professional, scientific and technical activities ai_pany_m
(-0.185) and ai_pany_m (-0.092) suggests a negative relationship with
employment rates.
These results do not support the overall premise of Hypothesis H1
that AI leads to increased employment. Instead, they suggest that
certain forms of AI adoption may lead to job displacement or
a negative influence on employment rates.
Table 6.1‑3 AI Adoption Impact on Employment Rates (using ai_pany and
ai_tany indicators)
| AI Adoption Indicator (Purpose of AI) |
AI_PANY (Model 3) |
AI_TANY (Model 6) |
| ai_pany_c / ai_tany_c (AI in Manufacturing) |
-0.416 |
-0.252 |
| ai_pany_e / ai_tany_e (AI in Water supply; sewerage, waste
management) |
0.166 |
0.215 |
| ai_pany_g / ai_tany_g (AI in Wholesale and retail
trade) |
0.458 |
0.191 |
| ai_pany_j / ai_tany_j (AI in Information and
communication) |
0.101 |
0.097 |
| ai_pany_m / ai_tany_m (AI in Professional, scientific and
technical activities) |
-0.185 |
-0.092 |
| ai_pany_n / ai_tany_n (AI Administrative and support service
activities) |
0.199 |
0.070 |
Secondary Hypothesis
(H2)
The Secondary Hypothesis (H2) proposes that the relationship between
AI adoption and employment exhibits gender-specific effects, with a
stronger positive impact on female employment rates compared to male
employment rates in EU countries. Models were run separately for
emp_rate_female and emp_rate_male. Similar to H1, diagnostic tests were
performed for these models. Both panel effects and random effects were
found in all models. The Hausman test results suggested use of Random
Effects models. Cross-sectional dependence was present in all models for
H2, thus robust standard errors (clustered) were applied. Most models
showed no autocorrelation. Heteroskedasticity tests showed mixed
results.
AI_HIRING_INDEX +
AI_PANY_SECTORS
Table 6.2‑1 SUMMARY OF MODEL RESULTS (DEPENDENT VARIABLE:
EMP_RATE_FEMALE AND EMP_RATE_MALE)
| Test |
Model1 |
Model2 |
Model3 |
Model4 |
Model5 |
Model6 |
| Formula |
emp_rate_female ~ ai_hiring_index + ai_pany_c + ai_pany_e +
ai_pany_f + ai_pany_g + ai_pany_h + ai_pany_i + ai_pany_j + ai_pany_m +
ai_pany_n + ai_pany_mean:ai_hiring_index |
emp_rate_female ~ ai_hiring_index + ai_pany_c + ai_pany_e +
ai_pany_f + ai_pany_g + ai_pany_h +
ai_pany_i + ai_pany_j + ai_pany_m + ai_pany_n |
emp_rate_female ~ ai_hiring_index + ai_pany_c + ai_pany_e +
ai_pany_g + ai_pany_j + ai_pany_m + ai_pany_n |
emp_rate_male ~ ai_hiring_index + ai_pany_c + ai_pany_e + ai_pany_f
+ ai_pany_g + ai_pany_h + ai_pany_i + ai_pany_j + ai_pany_m + ai_pany_n
+ ai_pany_mean:ai_hiring_index |
emp_rate_male ~ ai_hiring_index + ai_pany_c + ai_pany_e +
ai_pany_f + ai_pany_g + ai_pany_h +
ai_pany_i + ai_pany_j + ai_pany_m + ai_pany_n |
emp_rate_male ~ ai_hiring_index + ai_pany_c + ai_pany_e + ai_pany_g
+ ai_pany_j + ai_pany_m + ai_pany_n |
| Model Type |
RE |
RE |
RE |
RE |
RE |
RE |
| F-test for individual effects FE vs Pooled |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
| Breusch-Pagan test RE vs Pooled |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
| Hausman test RE vs FE |
p = 0.9999 |
p = 0.5408 |
p = 0.9603 |
p = 0.1853 |
p = 0.9991 |
p = 0.9967 |
| Interpretation |
Use Random Effects |
Use Random Effects |
Use Random Effects |
Use Random Effects |
Use Random Effects |
Use Random Effects |
| Wald test for variable removal |
N/A |
p = 0.3284* |
p = 0.5646* |
|
p = 0.1588* |
p = 0.5787* |
| Interpretation |
|
Removal justified |
Removal justified |
|
Removal justified |
Removal justified |
| Breusch-Godfrey/Wooldridge test Serial
Correlation |
p = 0.1703 |
p = 0.2227 |
p = 0.1956 |
p = 0.1580 |
p = 0.2277 |
p = 0.3311 |
| Interpretation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
| Studentized Breusch-Pagan test
Heteroskedasticity |
p = 0.0173* |
p = 0.0030* |
p = 0.0040* |
p = 0.0591 |
p = 0.0613 |
p = 0.0373* |
| Interpretation |
Heteroskedasticity present |
Heteroskedasticity present |
Heteroskedasticity present |
Homoskedasticity |
Homoskedasticity |
Heteroskedasticity present |
| Pesaran CD test Cross-sectional Dependence |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
Cross-sect. dep. present. Robust SEs applied Clustered |
| Insignificant variables 5% |
ai_hiring_index, ai_pany_f, ai_pany_h, ai_pany_i,
ai_hiring_index:ai_pany_mean |
ai_pany_f, ai_pany_h, ai_pany_i |
All significant |
ai_hiring_index, ai_pany_f, ai_pany_h, ai_pany_i, ai_pany_j |
ai_pany_f, ai_pany_h, ai_pany_i, ai_pany_j |
ai_pany_j |
| Insignificant variables 10% |
ai_hiring_index, ai_pany_f, ai_pany_h, ai_pany_i,
ai_hiring_index:ai_pany_mean |
ai_pany_f, ai_pany_h, ai_pany_i |
All significant |
ai_hiring_index, ai_pany_f, ai_pany_h, ai_pany_i |
ai_pany_f, ai_pany_h, ai_pany_i, ai_pany_j |
ai_pany_j |
TABLE 6.2 2 MODEL COMPARISON (DEPENDENT VARIABLE: EMP_RATE_FEMALE AND
EMP_RATE_MALE)
##
## Model Comparison: General to Specific
## ========================================================================================
## Dependent variable:
## -----------------------------------------------------------
## emp emp_rate_male
## Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
## (1) (2) (3) (4) (5) (6)
## ----------------------------------------------------------------------------------------
## ai_hiring_index -0.029 -0.068*** -0.074*** -0.012 -0.055*** -0.055***
## (0.029) (0.020) (0.020) (0.020) (0.014) (0.014)
## ai_pany_c -0.512*** -0.459*** -0.533*** -0.408*** -0.353** -0.394***
## (0.167) (0.164) (0.126) (0.157) (0.160) (0.106)
## ai_pany_e 0.176** 0.177** 0.168** 0.173** 0.174** 0.149**
## (0.080) (0.081) (0.081) (0.067) (0.069) (0.069)
## ai_pany_f 0.273 0.284 0.182 0.187
## (0.253) (0.252) (0.201) (0.201)
## ai_pany_g 0.397*** 0.388*** 0.442*** 0.373*** 0.366*** 0.403***
## (0.136) (0.131) (0.127) (0.119) (0.115) (0.108)
## ai_pany_h -0.166 -0.274 0.056 -0.052
## (0.237) (0.218) (0.162) (0.160)
## ai_pany_i 0.068 0.002 -0.027 -0.096
## (0.144) (0.147) (0.123) (0.130)
## ai_pany_j 0.145*** 0.128*** 0.130*** 0.057* 0.039 0.035
## (0.035) (0.032) (0.029) (0.034) (0.032) (0.026)
## ai_pany_m -0.266*** -0.258*** -0.216*** -0.179*** -0.170*** -0.148***
## (0.071) (0.072) (0.061) (0.066) (0.064) (0.050)
## ai_pany_n 0.272*** 0.252*** 0.206*** 0.178** 0.156** 0.144**
## (0.083) (0.083) (0.077) (0.072) (0.071) (0.065)
## ai_hiring_index:ai_pany_mean -0.007 -0.008**
## (0.005) (0.003)
## Constant 70.061*** 70.711*** 70.049*** 79.790*** 80.473*** 80.509***
## (1.857) (1.716) (1.500) (1.106) (1.054) (0.973)
## ----------------------------------------------------------------------------------------
## Observations 68 68 68 68 68 68
## R2 0.763 0.759 0.758 0.881 0.878 0.864
## Adjusted R2 0.717 0.717 0.730 0.857 0.857 0.848
## ========================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
## Robust standard errors in parentheses
## Model 1 : Random Effects
## Model 2 : Random Effects
## Model 3 : Random Effects
## Model 4 : Random Effects
## Model 5 : Random Effects
## Model 6 : Random Effects
The analysis of the variable coefficients in Model 3 and 6 provides
the following insights (Table 6.2-2).
Supporting H2:
- Water supply; sewerage, waste management (ai_pany_e) for both showed
a positive and statistically significant impact, which was stronger for
female employment (0.168) than male (0.149).
- Information and communication (ai_pany_j) have statistically
significant positive impact for female employment (0.130), while it was
not significant for male employment model.
- Administrative and support service activities (ai_pany_n) for both
showed a positive and statistically significant impact, which was
stronger for female employment (0.206) than male (0.144).
Contradicting H2:
- AI hiring index (ai_hiring_index) for both female (-0.074) and male
(-0.055) employment showed a negative impact, but the negative impact
was stronger for female employment.
- Manufacturing (ai_pany_c) for both genders showed a negative and
statistically significant impact, more for female employment (-0.533)
than male (-0.394).
- Professional, scientific and technical activities (ai_pany_m) for
both exhibited a negative and statistically significant impact, stronger
for female employment (-0.216) than male (-0.148).
- Wholesale and retail trade (ai_pany_g) for both displayed a positive
impact, the magnitudes were very similar, with a small difference
slightly favouring male employment (0.444 vs 0.442).
Table 6.2‑3 Comparison of AI Adoption Impact on Female vs. Male
Employment Rates (using ai_pany indicators)
| AI Adoption Indicator (Purpose of AI) |
Impact on Female Employment (Model 3) |
Impact on Male Employment (Model 6) |
| ai_hiring_index (AI in hiring processes) |
-0.074 |
-0.055 |
| ai_pany_c (AI in Manufacturing) |
-0.533 |
-0.394 |
| ai_pany_e (AI in Water supply; sewerage, waste
management) |
0.168 |
0.149 |
| ai_pany_g (AI in Wholesale and retail trade) |
0.442 |
0.403 |
| ai_pany_j (AI in Information and
communication) |
0.130 |
0 |
| ai_pany_m (AI in Professional, scientific and technical
activities) |
-0.216 |
-0.148 |
| ai_pany_n (AI Administrative and support service
activities) |
0.206 |
0.144 |
AI_HIRING_INDEX +
AI_TANY_SECTORS
Table 6.2‑4 SUMMARY OF MODEL RESULTS (DEPENDENT VARIABLE:
EMP_RATE_FEMALE AND EMP_RATE_MALE)
| Formula |
emp_rate_female ~ ai_hiring_index + ai_tany_c + ai_tany_e +
ai_tany_f + ai_tany_g + ai_tany_h + ai_tany_i + ai_tany_j + ai_tany_m +
ai_tany_n + ai_tany_mean:ai_hiring_index |
emp_rate_female ~ ai_hiring_index + ai_tany_c + ai_tany_e +
ai_tany_f + ai_tany_g + ai_tany_h + ai_tany_i +
ai_tany_j + ai_tany_m + ai_tany_n |
emp_rate_female ~ ai_hiring_index + ai_tany_c + ai_tany_e +
ai_tany_g + ai_tany_j + ai_tany_m |
emp_rate_male ~ ai_hiring_index + ai_tany_c + ai_tany_e + ai_tany_f
+ ai_tany_g + ai_tany_h + ai_tany_i + ai_tany_j + ai_tany_m + ai_tany_n
+ ai_tany_mean:ai_hiring_index |
emp_rate_male ~ ai_hiring_index + ai_tany_c + ai_tany_e +
ai_tany_f + ai_tany_g + ai_tany_h + ai_tany_i +
ai_tany_j + ai_tany_m + ai_tany_n |
emp_rate_male ~ ai_hiring_index + ai_tany_c + ai_tany_e + ai_tany_g
+ ai_tany_j + ai_tany_m |
| Model Type |
FE |
RE |
RE |
RE |
RE |
RE |
| F-test for individual effects (FE vs Pooled) |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
Panel effects exist |
| Breusch-Pagan test (RE vs Pooled) |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
Random effects exist |
| Hausman test (RE vs FE) |
p = 0.0000* |
p = 0.9226 |
p = 0.9460 |
p = 1.0000 |
p = 0.5862 |
p = 0.9397 |
| Interpretation |
Use Fixed Effects |
Use Random Effects |
Use Random Effects |
Use Random Effects |
Use Random Effects |
Use Random Effects |
| Wald test for variable removal |
|
p = 0.4319* |
p = 0.9116* |
|
p = 0.3174* |
p = 0.8361* |
| Interpretation |
|
Removal justified |
Removal justified |
|
Removal justified |
Removal justified |
| Breusch-Godfrey/Wooldridge test (Serial
Correlation) |
p = 0.7053 |
p = 0.0630 |
p = 0.1468 |
p = 0.0152* |
p = 0.0723 |
p = 0.1567 |
| Interpretation |
No autocorrelation |
No autocorrelation |
No autocorrelation |
Autocorrelation present |
No autocorrelation |
No autocorrelation |
| Studentized Breusch-Pagan test
(Heteroskedasticity) |
p = 0.1204 |
p = 0.0107* |
p = 0.0010* |
p = 0.2789 |
p = 0.2490 |
p = 0.0418* |
| Interpretation |
Homoskedasticity |
Heteroskedasticity present |
Heteroskedasticity present |
Homoskedasticity |
Homoskedasticity |
Heteroskedasticity present |
| Pesaran CD test (Cross-sectional Dependence) |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
p = 0.0000* |
| Interpretation |
Cross-sect. dep. present. Robust SEs applied (Clustered) |
Cross-sect. dep. present. Robust SEs applied (Clustered) |
Cross-sect. dep. present. Robust SEs applied (Clustered) |
Cross-sect. dep. present. Robust SEs applied (Clustered) |
Cross-sect. dep. present. Robust SEs applied (Clustered) |
Cross-sect. dep. present. Robust SEs applied (Clustered) |
| Insignificant variables (5%) |
ai_hiring_index, ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i,
ai_tany_n, ai_hiring_index:ai_tany_mean |
ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i, ai_tany_n |
ai_tany_g |
ai_hiring_index, ai_tany_f, ai_tany_h, ai_tany_i, ai_tany_j,
ai_tany_n, ai_hiring_index:ai_tany_mean |
ai_tany_f, ai_tany_h, ai_tany_i, ai_tany_j, ai_tany_n |
All significant |
| Insignificant variables (10%) |
ai_hiring_index, ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i,
ai_tany_n, ai_hiring_index:ai_tany_mean |
ai_tany_f, ai_tany_g, ai_tany_h, ai_tany_i, ai_tany_n |
ai_tany_g |
ai_hiring_index, ai_tany_f, ai_tany_h, ai_tany_i, ai_tany_j,
ai_tany_n, ai_hiring_index:ai_tany_mean |
ai_tany_f, ai_tany_h, ai_tany_i, ai_tany_j, ai_tany_n |
All significant |
Table 6.2‑5 MODEL COMPARISON (DEPENDENT VARIABLE: EMP_RATE_FEMALE and
EMP_RATE_MALE)
##
## Model Comparison: General to Specific
## ========================================================================================
## Dependent variable:
## -----------------------------------------------------------
## emp emp_rate_male
## Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
## (1) (2) (3) (4) (5) (6)
## ----------------------------------------------------------------------------------------
## ai_hiring_index -0.047 -0.078*** -0.077*** -0.030 -0.059*** -0.057***
## (0.032) (0.022) (0.022) (0.024) (0.016) (0.015)
## ai_tany_c -0.343*** -0.301*** -0.302*** -0.234** -0.231** -0.240***
## (0.104) (0.097) (0.088) (0.092) (0.092) (0.088)
## ai_tany_e 0.267*** 0.245*** 0.240*** 0.178** 0.187** 0.188**
## (0.072) (0.078) (0.080) (0.075) (0.073) (0.075)
## ai_tany_f 0.141 -0.002 -0.011 -0.012
## (0.168) (0.170) (0.145) (0.132)
## ai_tany_g 0.071 0.127 0.138* 0.229*** 0.220*** 0.213***
## (0.120) (0.113) (0.083) (0.082) (0.077) (0.062)
## ai_tany_h -0.136 -0.066 0.040 -0.008
## (0.138) (0.143) (0.094) (0.094)
## ai_tany_i 0.107 0.015 -0.016 -0.031
## (0.127) (0.117) (0.090) (0.090)
## ai_tany_j 0.141*** 0.144*** 0.164*** 0.039 0.042 0.060***
## (0.032) (0.030) (0.029) (0.029) (0.027) (0.023)
## ai_tany_m -0.131*** -0.124** -0.132*** -0.084** -0.093*** -0.095***
## (0.038) (0.049) (0.041) (0.037) (0.034) (0.030)
## ai_tany_n 0.110* 0.063 0.062 0.051
## (0.065) (0.059) (0.051) (0.043)
## ai_hiring_index:ai_tany_mean -0.004 -0.004
## (0.004) (0.003)
## Constant 69.621*** 69.224*** 79.903*** 80.260*** 80.054***
## (1.672) (1.563) (0.986) (1.020) (1.006)
## ----------------------------------------------------------------------------------------
## Observations 73 73 73 73 73 73
## R2 0.650 0.739 0.734 0.892 0.868 0.851
## Adjusted R2 0.370 0.697 0.710 0.872 0.847 0.838
## ========================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
## Robust standard errors in parentheses
## Model 1 : Fixed Effects
## Model 2 : Random Effects
## Model 3 : Random Effects
## Model 4 : Random Effects
## Model 5 : Random Effects
## Model 6 : Random Effects
Impact Comparison (AI_TANY indicators - Table 6.2-5):
Supporting H2:
- Water Supply, sewerage, waste management (ai_tany_e): both positive,
and the positive impact was stronger for females (0.240) than males
(0.188).
- Wholesale and retail trade (ai_tany_g) for both displayed a positive
impact, but higher for male (0.213 vs 0.138).
- Information and communication (ai_tany_j): both positive, and the
positive impact was stronger for females (0.164) than males
(0.060).
Contradicting H2:
- AI hiring index (ai_hiring_index) for both negative, but the
negative impact was stronger for females (-0.077 vs. -0.057).
- Manufacturing (ai_tany_c) for both negative, and the negative impact
was stronger for females (-0.302 vs. -0.240).
- Professional, scientific and technical activities (ai_tany_m) for
both negative, and the negative impact was stronger for females (-0.132
vs -0.095).
Table 6.2‑6 Comparison of AI Adoption Impact on Female vs. Male
Employment Rates
| AI Adoption Indicator |
Impact on Female Employment (Model 3) |
Impact on Male Employment (Model 6) |
| ai_hiring_index (AI in hiring processes) |
-0.077 |
-0.057 |
| ai_tany_c (AI in Manufacturing) |
-0.302 |
-0.240 |
| ai_tany_e (AI in Water supply; sewerage, waste
management) |
0.240 |
0.188 |
| ai_tany_g (AI in Wholesale and retail trade) |
0.138 |
0.213 |
| ai_tany_j (AI in Information and
communication) |
0.164 |
0.060 |
| ai_tany_m (AI in Professional, scientific and technical
activities) |
-0.132 |
-0.095 |
Findings
The analysis of AI adoption and its impact on employment in the EU
provides detailed results for both its main and secondary hypotheses,
and highlights critical areas for future consideration and action.
Main Hypothesis
(H1)
The analysis investigated how sectoral AI adoption in EU countries
influences the employment rate for individuals aged 20-64, hypothesizing
that AI technologies complement human labour and create new job
opportunities. The analysis indicates different effect among sectors
with positive dependency in sectors:
- water supply; sewerage, waste management
- wholesale and retail trade
- information and communication
- administrative and support service activities
and negative in sectors:
- manufacturing
- professional, scientific and technical activities.
Liu (2024) supports finding of a negative correlation with
unemployment. His regression analysis (2012-2022) revealed that
increased AI adoption is linked to lower unemployment. He found that
increases in the AI Index (published papers) led to decrease in
unemployment rates, and increased public interest in AI (measured by
Google Trends) correlated with reduction in unemployment. The
information technology and consumer goods sectors increased in job
creation due to the rapid implementation of artificial
intelligence.[1]
Adhikari & Hamal (2024) found that higher AI adoption and
education levels initially lead to job displacement and do not
immediately translate into job creation. However, strategic
implementation of AI significantly mitigates these adverse effects in
the long term. [2]
Secondary Hypothesis
(H2)
The analysis examined the relationship between AI adoption and
employment gender-specific effects, hypothesizing stronger positive
impact on female employment rates compared to male employment rates in
EU countries. The analysis indicates different effect among sectors with
positive dependency in sectors:
- water supply; sewerage, waste management
- information and communication
and stronger negative in sectors:
- manufacturing
- professional, scientific and technical activities.
Effect in the rest of the sectors were mixed.
Also, AI hiring index dependency on employment was negative for both
female and male, but the negative impact was stronger for females.
Cited articles do not contain direct findings on gender-specific
employment impacts of AI adoption. However, the potential for AI to
exacerbate existing labour market inequalities is discussed [2,3]. This
general concern about inequality includes gender as a dimension that
needs to be addressed. Analysis of H2 of gender-specific effects
contributes to this gap and provides empirical data.
Methodological
Similarities and Differences
Selected literature describes utilisation of empirical data and
regression analysis from 2012 to 2022 [1]. The core quantitative
analysis was based on a multiple linear regression model, using the AI
Index and Google Trends data as predictors for the unemployment rate.
Also, Pearson correlation coefficients were calculated and the dataset
was divided into training and test parts. Other source employed a robust
econometric approach for data from 2010 to 2022 [2]. A key technique was
Principal Components Analysis (PCA) to address multicollinearity among
variables, transforming them into uncorrelated principal components for
regression analysis. Heteroskedasticity-robust standard errors were used
and stationarity tests were conducted.
Summary
In summary, my findings on the partially positive correlation between
AI adoption and employment (or negative with unemployment) are supported
by Liu (2024). My study of gender-specific effects provides new
empirical evidence in area that literature identifies as important but
only generally indicate potential inequalities.
Next Possible Ways of
Handling the Topic/Problem
Analysing AI adoption and its impacts is described as an urgent
economic problem requiring multifaceted solutions. Given the growing and
evolving nature of AI, ongoing efforts are needed to manage its
disruptive consequences while leveraging its benefits. It is essential
that policymakers develop adaptive policies and education systems to
prepare the workforce for the changes caused by AI adoption. There is a
need to define responsibilities, liabilities and guilt between human and
AI. Policymakers need to develop strategies to mitigate negative
impacts, especially in low-skilled sectors where job losses are most
visible. Reskilling and upskilling programs are key interventions to
equip workers for new AI-related roles and address the changing nature
of job roles and skills required. Further analysis is needed to
determine which specific skills and sectors will be most affected by AI
adoption. Analysing sector-specific challenges and successes is critical
to support effective workforce transitions. For large-scale AI
implementations, especially in sectors such as healthcare, quantifying
medical and economic outcomes using standard procedures is crucial to
justify investment. Practical AI implementation faces fundamental
barriers, such as lack difficulties in sharing information, and the need
to promote standards. Addressing these challenges requires building
capacity in the AI ecosystem. Fostering trust and confidence among
professionals and the general public is essential for successful
adoption and acceptance of AI across sectors. The complex interplay of
technological, economic, and social factors requires further
interdisciplinary research to provide a holistic understanding and
develop effective strategies to manage implications of AI. This includes
further exploration of gender employment impacts, taking into account
the mixed findings of the secondary hypothesis.
Bibliography
[1] Liu,J. (2024). The Impact of the Development of Artificial
Intelligence on Unemployment Rates. Advances in Economics, Management
and Political Sciences,121,154-163. https://doi.org/10.54254/2754-1169/121/20242410
[2] Adhikari, P., & Hamal, P. (2024). Impact and Regulations of
AI on Labor Market and Employment in USA. Preprints. https://doi.org/10.20944/preprints202407.0906.v1
[3] Ernst, E., Merola, R. and Samaan, D. Economics of Artificial
Intelligence: Implications for the Future of Work. IZA Journal of Labor
Policy, Sciendo, Vol. 9 (Issue 1), (2019) https://doi.org/10.2478/izajolp-2019-0004
[4] Havryk, A., & Nazarova, T. (2024). Artificial intelligence
and its role in the labor market and financial sector itself: US point
of view. International Science Journal of Management, Economics &
Finance, 3(3), 1–9. https://doi.org/10.46299/j.isjmef.20240303.01
[5] Wolff J, Pauling J, Keck A and Baumbach J (2021) Success Factors
of Artificial Intelligence Implementation in Healthcare. Front. Digit.
Health 3:594971. doi: 10.3389/fdgth.2021.594971