index.qmd

---
title: "Final Project (Due Wednesday, May 10th)"
author: "Jerry Cai"
institute: "University of Pennsylvania"
date: last-modified
toc: true

bibliography: references.bib

format: 
  html:
    self-contained: true

editor: source
---

This is the final project submission on Quarto

Jerry Cai

# R Code

```{r}
library(ggplot2)
library(readr)
library(ggdag)
library(tidyverse)
library(tidyr)
library(dplyr)
library(gt)
library(modelsummary)

```


# Introduction
Widespread adoption of online marketplace platforms and artificial intelligence have created tremendous economic opportunities for freelance gig workers, enabling remote work and flexible work arrangements. However, an equivalent amount of skilled labor is being replaced by technological systems, creating an entirely new employment landscape to navigate for creative, cultural, and entrepreneurial sectors (Sytsma, 2023). The World Economic Forum has defined the digital economy as “With digital knowledge and information as key production factors, modern information networks as important carriers, and information and communication technology as an essential component of increasing productivity through economic activities”. Nowadays, the digital economy represents an emerging industrial revolution that is reliant on high-tech digital technology and rapidly expanded globally (Zhang).  

Although the opportunities of digital economy are expanding, the inequality within countries has become worse. According to the United Nations, 71 percent of the world’s population live in countries where inequality has grown. 
Inequalities of opportunity affect a person’s life expectancy and access to basic services such as healthcare, education, water, and sanitation. These inequalities also entrench uncertainty, vulnerability and insecurity, undermines trust in institutions and government, increases social discord and tensions and trigger violence and conflicts. As economic and financial systems become increasingly digitized, an individual’s overall wellness becomes increasingly tied to their financial stability. For the traditional artisanal communities that are not fully technologically integrated with global financial systems,  such a change creates inevitable challenges that threaten artisan and students’ future economic wellness (United Nations).

In spite of the growing economic disparities across developing nations, participation in the online gig economy has proven effective for certain groups to improve access to crucial economic opportunities, while maintaining geographic and temporal accessibility. According to Huang et al (2019), the growth of the gig economy has been partly attributed to technological advancements that enable flexible work environments, along with economic downturns, and associated financial stressors in the offline economy. Huang et al (2019) focused on how unemployment in the traditional labor market is likely to have a significant impact on the supply of labor in online markets be-cause associated financial stressors (i.e., loss of in-come) provide sufficient incentive for unemployed and underemployed workers to engage in or experiment with a new form of employment.  

When it comes to understanding factors that moderate the relationship between offline unemployment and the supply of labor in online gig economy, one must consider key features that characterize the workers in the local labor market (e.g., population demographics and geographic dispersion of county residents’ social connections) and those characterizing the geographic context (e.g., internet connectivity). These moderating factors help us to better understand heterogeneity in the relationship between unemployment and participation in the online labor markets. (Huang 2019). Online labor markets involve work that is easily outsourced and delivered online, such as software development and data entry. Accordingly, workers recently laid off from positions in information technology (IT)-related industries—who we would expect can perform virtual work that does not require physically being on site and that also requires relatively little interpersonal interaction—should be particularly likely to move online when faced with job loss.

The existing body of research seeks to understand the drivers of the gig economy would enable effective policy efforts that increase the expansion gig economy opportunities and its overall efficacy improving financial stability for more people. However, there seems to be less substantial understanding of the effects of online gig economy participation on an individual’s overall financial wellness and economic outcomes. There seems to be strong promise for governments to bolster digital gig economy employment opportunities to effectively improve financial stability for vulnerable groups. Findings that motivate the consequential effects of digital gig economic participation on financial stability can translate into effective policy for affecting financial health with increased digital gig economy participation.  In this study, I aim to investigate the effective strength of gig economy participation on individual financial wellness and stability. 

# Theory and Hypotheses

Hypothesis 1: Countries with greater digital gig economy participation tend to see greater individual financial health. 

Hypothesis 2: Countries with high access to internet technologies and higher overall wage/salaried employment will experience the highest frequency of digital gig platform usage.

Rationale 1: Online gig/freelance work platforms are more efficient mechanisms of value creation, enabling supply to quickly meet demand through digital means. Greater efficiency takes less energy and time to create economic value, allowing for more overall value creation and therefore long-term freedom and flexibility to pursue activities that create greater fulfillment.

Marginal economic earnings measures the comparative difference in outcomes between digital gig economy workers. Ex. A gig worker in Africa would derive greater marginal benefit from economic earnings than a gig worker in the US. Greater economic disparities/instability in a given country will cause more individuals to pursue freelance/gig marketplace platform work, since the marginal benefit is comparatively higher. In countries with lower wage/salary employment rates, gig platform employment is the next best alternative for economic opportunities. Additionally, countries of greater information technology (IT) are correlated with greater technology investment, growth and overall ecosystem dynamics that enable efficient value creation and exchange. 

Rationale 2:
A report by Manyika et al (2016), which documented 160+ million individuals participating in the digital gig economy, estimates that a substantial fraction (30%) of workers earn income via gig workout of necessity (because of a lack of better employment options). Specifically, an estimated 1% increase in a county’s reported unemployment rate is associated with an approximate 21.8% increase in the number of residents active on the platform and an approximate 15% increase in the total volume of bids submitted by those residents, translating to approximately $1.8 million in additional annual wages earned on the platform. It has been observed that counties with better internet access and digital skills are most likely to experience gig-economy opt-in following local unemployment shocks. Participation in the gig economy may also be influenced by policy interventions (e.g., education levels may be increased via training initiatives, and internet access may be improved through investment in internet infrastructure).

# Data Description

## A - Data Description
The variables I will examine in my dataset combined_data are:
totalSupply, totalDemand
adjIncomeCapita
commExports, serviceICTExports
totalUnemployment, totalWageSalWorkers

From the OLI dataset, variables I will look at the total daily new listings and filled listings across the 5 largest online gig work sites: Freelancer.com, Upwork.com, Guru.com, Peopleperhour.com, Mturk.com. Using an AWS web scraper method, the OLI researchers sampled the sites in 24 hour intervals to observe newly added job postings and filled job postings. Each posting’s geographic information and employment sector were noted. 
 
Annual total count : annualTotalCount
Software development : softwareDev
Writing and translation : writingTranslation
Clerical and data entry : clericalDataEntry
Creative and multimedia : creativeMultimedia
Professional services : professionalServices 
Sales and marketing support : salesMarketingSupport

From the WHO dataset, variables I will look at are

Financial Wellness:  adjusted income per capita (adjIncomeCapita), 
Gig Economy Participation:  Gig economy supply (totalSupply), Gig economy demand (totalSupply)
Internet accessibility and infrastructure: service communications/technology exports (serviceICTExports), 
Total employment rates of wage/salaried workers: total unemployment (totalUnemployment)


To test hypothesis 1, I chose variables that represented digital gig economy participation and indicators of  financial health: which are  gig economy worker participation count (totalSupply), gig economy job listing count (totalSupply) and adjusted income per capita (adjIncomeCapita). Gig economic participation count is calculated via survey data from the OLI database, which surveyed workers on digital platform about their country of residence and which day this interaction was recorded. Such was replicated each day over 4 years, which serve to reflect to total amount of gig economy activity in a given year. Adjusted income per capita is a standard WHO development indicator, which is mean income per capita adjusted into USD. 

For further empirical extension in hypothesis 2, I chose variables that are effectors of gig economy participation count (totalSupply), such as digital technology service exports (serviceICTExports) and overall total unemployment (totalUnemployment). I will note there is a possible interaction between the later 2 variables.

## Loading in dataset
```{r}
#| echo: false
#| warning: false


# Write combined_data as csv
input_file_path_combined <- "/Users/jycai/Documents/GitHub/psci3200_jerrycai/combined_data.csv"

# write_csv(combined_data, input_file_path_combined)

# Read combined_data csv
combined_data <- read_csv(input_file_path_combined)

# Create a named vector for the mappings
code_to_name <- c(
  "NY.ADJ.NNTY.PC.CD" = "adjIncomeCapita",
  "BX.GSR.NFSV.CD" = "serviceICTExports",
  "SL.UEM.TOTL.ZS" = "totalUnemployment",
  "totalSupply" = "totalSupply",
  "totalDemand" = "totalDemand"
)

```

## Data variables description
```{r}
#| echo: false
#| warning: false


combined_data_2 <- combined_data %>%
  filter(indicator %in% c("adjIncomeCapita", "serviceICTExports", "totalUnemployment", "totalDemand", "totalSupply"))
  # filter(country %in% c("Chile", "Argentina", "Brazil", "Bolivia", "Peru", "Venezuela", "Uruguay", "Ecuador", "Mexico", "Panama", "Nicaragua", "Costa Rica", "Honduras"))

# Reshaping the data to have one row per country-year with separate columns for each indicator
combined_data_2_wide <- combined_data_2 %>%
  pivot_wider(names_from = indicator, values_from = value) %>%
  filter(!is.na(adjIncomeCapita) & !is.na(serviceICTExports)& !is.na(totalUnemployment)& !is.na(totalDemand) & !is.na(totalSupply))
  
# combined_data_2a_wide <- combined_data_2a_wide[combined_data_2a_wide$totalDemand <= 2.5e+05, ]

summary_df <- combined_data_2_wide %>%
  summarise(
    Mean_adjIncomeCapita = mean(adjIncomeCapita, na.rm = TRUE),
    Min_adjIncomeCapita = min(adjIncomeCapita, na.rm = TRUE),
    Max_adjIncomeCapita = max(adjIncomeCapita, na.rm = TRUE),
    SD_adjIncomeCapita = sd(adjIncomeCapita, na.rm = TRUE),

    Mean_serviceICTExports = mean(serviceICTExports, na.rm = TRUE),
    Min_serviceICTExports = min(serviceICTExports, na.rm = TRUE),
    Max_serviceICTExports = max(serviceICTExports, na.rm = TRUE),
    SD_serviceICTExports = sd(serviceICTExports, na.rm = TRUE),

    Mean_totalUnemployment = mean(totalUnemployment, na.rm = TRUE),
    Min_totalUnemployment = min(totalUnemployment, na.rm = TRUE),
    Max_totalUnemployment = max(totalUnemployment, na.rm = TRUE),
    SD_totalUnemployment = sd(totalUnemployment, na.rm = TRUE),

    Mean_totalSupply = mean(totalSupply, na.rm = TRUE),
    Min_totalSupply = min(totalSupply, na.rm = TRUE),
    Max_totalSupply = max(totalSupply, na.rm = TRUE),
    SD_totalSupply = sd(totalSupply, na.rm = TRUE),

    Mean_totalDemand = mean(totalDemand, na.rm = TRUE),
    Min_totalDemand = min(totalDemand, na.rm = TRUE),
    Max_totalDemand = max(totalDemand, na.rm = TRUE),
    SD_totalDemand = sd(totalDemand, na.rm = TRUE)
  )

# Now, you can create a GT table directly from this summary_df
summary_table <- gt(summary_df) %>%
  tab_header(title = "Summary Statistics for Economic Data")

# Print or view the table in RStudio's Viewer pane or save as HTML
print(summary_table)

```

# Research Design

## B - Specify the main regression model 
The regression model used will be the standard linear regression model. This is sufficient for such application given the relatively large sample and population size of individuals sampled from respective countries. There were a total of  n = 149 different countries surveyed, with data across 4 years (2016:2020), representing 596 observations for each indicator. The regression model is robust for assessing effect strength of gig economy participation on individual financial health, given that financial health has multiple other external covariate effectors such as individual spending rates, consumer price index, cost of living, loan/credit lines and strength of individual long-term investments. Also affecting individual financial health, potential covariates for gig economy participation include total unemployment rate and strength of traditional employment sectors. An individual’s financial health is greatly effected by outflows, such as living expenses, tax rates which influence the remaining net balance for long-term financial stability. Given the complex effects that external financial institutions have on an individual’s financial health,  we are controlling strictly for income sources, in which adjusted income per capita can reasonably represent average incomes. 

### Total Demand Model
$$
\text{adjIncomeCapita} = \beta_0 + \beta_1 \times \text{totalDemand} + \epsilon
$$
# Model_2_a: adjIncomeCapita` ~ `totalDemand
```{r}
#| echo: false
#| warning: false
#| 
# Model_2_a

# Check if both indicators are available for each country and year
combined_data_2a <- combined_data %>%
  filter(indicator %in% c("adjIncomeCapita", "totalDemand"))

# Reshaping the data to have one row per country-year with separate columns for each indicator
combined_data_2a_wide <- combined_data_2a %>%
  pivot_wider(names_from = indicator, values_from = value) %>%
  filter(!is.na(adjIncomeCapita) & !is.na(totalDemand))
  
combined_data_2a_wide <- combined_data_2a_wide[combined_data_2a_wide$totalDemand <= 2.5e+05, ]

# Run the regression model
regression_model_2_a <- lm(`adjIncomeCapita` ~ `totalDemand`, data = combined_data_2a_wide)

# Display the summary of the regression model
summary(regression_model_2_a)

# Plot diagnostics
plot(regression_model_2_a)

# Create a scatter plot
library(ggplot2)

scatter_plot_2_a <- ggplot(combined_data_2a_wide, aes(x = totalDemand, y = adjIncomeCapita )) +
  geom_point(alpha = 0.6) +  # Adjust point transparency with alpha
  labs(title = "Effect of gig totalDemand on adjIncomeCapita",
       x = "totalDemand",
       y = "adjIncomeCapita") +
  theme_minimal()  # Use a minimal theme for a cleaner look

# Display the plot
print(scatter_plot_2_a)

```

### Total Supply Model
$$
\text{adjIncomeCapita} = \beta_0 + \beta_1 \times \text{totalSupply} + \epsilon
$$

# Model_2_b: adjIncomeCapita` ~ `totalSupply
```{r}
#| echo: false
#| warning: false
#| 

# Model_2_b

# Check if both indicators are available for each country and year
combined_data_2b <- combined_data %>%
  filter(indicator %in% c("adjIncomeCapita", "totalSupply"))

# Reshaping the data to have one row per country-year with separate columns for each indicator
combined_data_2b_wide <- combined_data_2b %>%
  pivot_wider(names_from = indicator, values_from = value) %>%
  filter(!is.na(adjIncomeCapita) & !is.na(totalSupply))
  
combined_data_2b_wide <- combined_data_2b_wide[combined_data_2b_wide$totalSupply <= 1e+06, ]

# Run the regression model
regression_model_2_b <- lm(`adjIncomeCapita` ~ `totalSupply`, data = combined_data_2b_wide)

# Display the summary of the regression model
summary(regression_model_2_b)

# Plot diagnostics
plot(regression_model_2_b)

# Create a scatter plot
library(ggplot2)

scatter_plot_2_b <- ggplot(combined_data_2b_wide, aes(x = totalSupply, y = adjIncomeCapita )) +
  geom_point(alpha = 0.6) +  # Adjust point transparency with alpha
  labs(title = "Effect of gig totalDemand on adjIncomeCapita",
       x = "totalSupply",
       y = "adjIncomeCapita") +
  theme_minimal()  # Use a minimal theme for a cleaner look

# Display the plot
print(scatter_plot_2_b)

```



Additive model
$$
\text{totalSupply} = \beta_0 + \beta_1 \times \text{serviceICTExports} + \beta_2 \times \text{totalUnemployment} +\epsilon
$$
Interaction model
$$
\text{totalSupply} = \beta_0 + \beta_1 \times \text{serviceICTExports} \times \beta_2 \times \text{totalUnemployment} +\epsilon
$$



## Shortcoming Analysis
In terms of the model’s shortcomings, a single indicator such as adjusted income per capita does not fully capture an individual’s financial wellness or ability to maintain financial stability. Even amongst total available income streams, would only have proportionally significant effects on lower adjusted income per capita. Only approximately 10%-30% of sampled countries’ populations have secondary income streams from the gig economy , meaning that the proportional contribution of digital gig economy income would be diminished in a per-capita metric that applies to millions of non-gig workers. 

# Findings

Given the multivariate effects on adjusted income per capita, the study found no strong conclusive evidence that participation in the digital gig economy significantly impacts individual financial health through adjusted annual income. This effect was weaker than previously anticipated,  due to the influence of larger macro variables such as taxation policy, overall economic health, and local industrial specialization. Additionally, the presence of outlier clusters displaying both low participation in the gig economy and varying levels of adjusted income per capita complicates the ability to discern a clear relationship. A further refined analysis, focusing specifically on Latin American and Asian countries, provided more controlled data yet still did not yield a validly significant relationship. The regression model, as indicated by poorly representative residuals, suggests that the impact of gig economy worker participation on individual wellness cannot be considered statistically significant.

# Empirical Extension

The sample size from the World Bank's development indicators spans individuals of varying financial wellness, encompassing the most vulnerable groups as characterized by the United Nations. These groups include rural women, female-led micro, small, and medium enterprises (MSMEs), female entrepreneurs, and gender-based violence (GBV) victims, along with age-vulnerable groups such as the elderly or youth, and those from specific geographical locations like rural, urban, or peri-urban areas, including farmers. Additionally, those in low-income or financially distressed situations are also considered vulnerable. However, the dataset lacks additional controls to filter for these vulnerable populations, which necessitates further investigation. There are covariate relationships noted, such as digital gig workers predominantly being male, who have access to internet infrastructure and possess basic digital literacy skills—as indicated by Huang. Conversely, females and geographically isolated populations are less likely to benefit from these prerequisites, thereby impacting their participation in the online gig economy.

Hypothesis 2 serves to focus on the effect of higher internet service exports and the total unemployment on gig economy worker participation.

To study their effects on gig economy participation, it appears that access to internet technology, higher internet export rates and the overall unemployments are strong effectors of totalSupply. Gig economy participation requires robust digital internet infrastructure to support computers and devices used for working remote jobs. The majority of gig economy jobs across the globe are online service exports (as reflected in the statistic) in roles such as software engineering, data entry and creative multimedia. Additionally, its important to examine surrounding macroeconomic contexts, such as overall employment rate in wage salaried jobs. It stands to reason that in countries with lower employment in wage/salaried jobs and higher access to digital internet technologies, that gig economy participation will be much higher, given individual needs to be economically stable. These comparative alternative income sources would be appealing to individuals who cannot find a traditional wage/salaried job. 

Income from online gig economy positions are assumed to be strongly correlated to a country’s internet/technology service exports, given that 80%-90% of such exports from developing countries fall under the digital gig economy criteria. We are hoping to discern the effector interaction for total gig economy supply and demand, which seem to be correlated with internet/technology service exports and total nationwide unemployment rate. Such insight into the interactions would help inform policy decisions on the degree of resource allocation. 

# Model_2_c: totalSupply ~ serviceICTExports * totalUnemployment

$$
\text{totalSupply} = \beta_0 + \beta_1 \times \text{serviceICTExports} \times \beta_2 \times \text{totalUnemployment}+ \epsilon
$$

```{r}
# Model_2_c

# Check if both indicators are available for each country and year
combined_data_2c <- combined_data %>%
  filter(indicator %in% c("totalSupply", "serviceICTExports", "totalUnemployment" ))
  # filter(country %in% c("Chile", "Argentina", "Brazil", "Bolivia", "Peru", "Venezuela", "Uruguay", "Ecuador", "Mexico", "Panama", "Nicaragua", "Costa Rica", "Honduras"))

# Reshaping the data to have one row per country-year with separate columns for each indicator
combined_data_2c_wide <- combined_data_2c %>%
  pivot_wider(names_from = indicator, values_from = value) %>%
  filter(!is.na(totalSupply) & !is.na(serviceICTExports) & !is.na(totalUnemployment))
  
combined_data_2c_wide <- combined_data_2c_wide[combined_data_2c_wide$totalSupply <= 1e+06, ]

# Run the regression model
regression_model_2_c <- lm(`totalSupply` ~ `serviceICTExports` + `totalUnemployment`, data = combined_data_2c_wide)

# Display the summary of the regression model
summary(regression_model_2_c)

# Plot diagnostics
plot(regression_model_2_c)

# Create a scatter plot
library(ggplot2)

scatter_plot_2_c <- ggplot(combined_data_2c_wide, aes(x = serviceICTExports, y = totalSupply )) +
  geom_point(alpha = 0.6) +  # Adjust point transparency with alpha
  labs(title = "Effect of serviceICTExports on gig economy totalSupply",
       x = "serviceICTExports",
       y = "totalSupply") +
  theme_minimal()  # Use a minimal theme for a cleaner look

# Display the plot
print(scatter_plot_2_c)

scatter_plot_2_c_i <- ggplot(combined_data_2c_wide, aes(x = totalUnemployment, y = totalSupply )) +
  geom_point(alpha = 0.6) +  # Adjust point transparency with alpha
  labs(title = "Effect of total Unemployment on gig economy totalSupply",
       x = "totalUnemployment",
       y = "totalSupply") +
  theme_minimal()  # Use a minimal theme for a cleaner look

# Display the plot
print(scatter_plot_2_c_i)

```

# Model_2_d: totalSupply ~ serviceICTExports + totalUnemployment
$$
\text{totalSupply} = \beta_0 + \beta_1 \times \text{serviceICTExports} + \beta_2 \times \text{totalUnemployment}+ \epsilon
$$
```{r}
# Model_2_d

# Check if both indicators are available for each country and year
combined_data_2d <- combined_data %>%
  filter(indicator %in% c("totalSupply", "serviceICTExports", "totalUnemployment" ))
  # filter(country %in% c("Chile", "Argentina", "Brazil", "Bolivia", "Peru", "Venezuela", "Uruguay", "Ecuador", "Mexico", "Panama", "Nicaragua", "Costa Rica", "Honduras"))

# Reshaping the data to have one row per country-year with separate columns for each indicator
combined_data_2d_wide <- combined_data_2d %>%
  pivot_wider(names_from = indicator, values_from = value) %>%
  filter(!is.na(totalSupply) & !is.na(serviceICTExports) & !is.na(totalUnemployment))
  
combined_data_2d_wide <- combined_data_2d_wide[combined_data_2d_wide$totalSupply <= 1e+06, ]

# Run the regression model
regression_model_2_d <- lm(`totalSupply` ~ `serviceICTExports` + `totalUnemployment`, data = combined_data_2d_wide)

# Display the summary of the regression model
summary(regression_model_2_d)

# Plot diagnostics
plot(regression_model_2_d)

# Create a scatter plot
library(ggplot2)

scatter_plot_2_d <- ggplot(combined_data_2d_wide, aes(x = serviceICTExports, y = totalSupply )) +
  geom_point(alpha = 0.6) +  # Adjust point transparency with alpha
  labs(title = "Effect of serviceICTExports on gig economy totalDemand",
       x = "serviceICTExports",
       y = "totalSupply") +
  theme_minimal()  # Use a minimal theme for a cleaner look

# Display the plot
print(scatter_plot_2_d)

scatter_plot_2_d_i <- ggplot(combined_data_2d_wide, aes(x = totalUnemployment, y = totalSupply )) +
  geom_point(alpha = 0.6) +  # Adjust point transparency with alpha
  labs(title = "Effect of total Unemployment on gig economy total supply",
       x = "totalUnemployment",
       y = "totalSupply") +
  theme_minimal()  # Use a minimal theme for a cleaner look

# Display the plot
print(scatter_plot_2_d_i)

```


# Discussion and Policy Implications
To enhance the potential benefits of digital gig economy participation, policy should continue investing in internet and technology service exports, alongside providing incentives for workers in participating countries. Particularly, countries with unstable economic environments are encouraged to attract more digital gig economic opportunities to increase individual participation. The next steps involve further segmenting the countries surveyed into emerging economies, such as those in Latin America, Africa, and Asia. Subsequent research will include conducting similar regression analyses but focusing on financial institution metrics such as total savings, credit and loan rates, and investment strengths. Additionally, a difference of difference analysis between males and females working in the gig economy will be conducted to determine if the effect on financial health is greater for vulnerable groups. Further research will also explore specific community interventions, such as distributing free computer devices, constructing internet infrastructure, and enhancing digital education, to support economic participation and benefit accrual from the gig economy.