Stack Overflow 2023 salary survey explained with Machine Learning
SHAP values of country, years of coding, and many more!
The purpose of this article is to analyze and determine the main factors that contribute to the current salaries of various types of developers. The study relies on the Stack Overflow survey from May 2023 with the dataset already publicly available at Kaggle. Similarly to previous stories, the analysis is based on SHapley Additive exPlanations (SHAP) values. Full details of the analysis are also available in a public notebook on Kaggle.
Table of contents:
· Step 1 — data preprocessing · Step 2 — setting a Machine Learning model to predict the yearly gross salaries · Step 3 — explanation of the obtained Machine Learning model. ∘ How do you learn to code? ∘ What online resources do you use to learn to code? ∘ What online courses or certifications do you use to learn to code? ∘ Which programming, scripting, and markup languages have you done extensive development work in over the past year? ∘ Which database environments have you done extensive development work in over the past year? ∘ Which cloud platforms have you done extensive development work in over the past year? ∘ Which web frameworks and web technologies have you done extensive development work in over the past year? ∘ Which other frameworks and libraries have you done extensive development work in over the past year? ∘ Which developer tools for compiling, building, and testing have you done extensive development work in over the past year? ∘ Which development environments and collaboration tools did you use regularly over the past year? ∘ What is the primary operating system in which you work? ∘ Which collaborative work management and/or code documentation tools did you use regularly over the past year? ∘ Which communication tools did you use regularly over the past year? ∘ Which AI-powered search tools did you use regularly over the past year? ∘ Which AI-powered developer tools did you use regularly over the past year? ∘ Which professional technology tools has my company? ∘ Which of the following best describes your current employment status? ∘ What is your age? ∘ Which best describes your current work situation? ∘ Which of the following best describes the highest level of formal education that you’ve completed? ∘ Including any education, how many years have you been coding in total? ∘ NOT including education, how many years have you coded professionally (as a part of your work)? ∘ Which of the following describes your current job, the one you do most of the time? ∘ Approximately how many people are employed by the company or organization you currently work for? ∘ Where do you live? ∘ Would you like to participate in the Professional Developer Series? ∘ Are you an individual contributor or a people manager? ∘ How many years of working experience do you have? ∘ What industry is the company you work for?
Step 1 — data preprocessing
Here, data preprocessing consists of the following steps:
- rescaling the label column (yearly total compensation) to kUSD/year;
- removing records with no salary reported or reporting anomalously small salaries (< 1 kUSD/year);
- removing 2% (2%) responders with the largest (smallest) compensations;
- replacing null values for numerical and categorical columns;
- extracting the LearnCode, LearnCodeOnline, LearnCodeCoursesCert, LanguageHaveWorkedWith, DatabaseHaveWorkedWith, PlatformHaveWorkedWith, WebframeHaveWorkedWith, MiscTechHaveWorkedWith, ToolsTechHaveWorkedWith, NEWCollabToolsHaveWorkedWith, OpSysProfessional use, OfficeStackAsyncHaveWorkedWith, OfficeStackSyncHaveWorkedWith, AISearchHaveWorkedWith, AIDevHaveWorkedWith, ProfessionalTech, and Employment columns, and encoding their different values with CountVectorizer using at least 200 records present in the dataset;
- encoding rare categorical variables (in Age, RemoteWork, EdLevel, DevType, OrgSize, Country, TBranch, ICorPM, and Industry columns) with no more than 60 different categories in each column and at least 20 data samples in each category;
- finally, dropping columns was not used for this analysis.
After these transformations, we obtained a cleaned dataset containing 45,430 records and 381 columns (including the label column).
Step 2 — setting a Machine Learning model to predict the yearly gross salaries
The data obtained from the previous step are divided randomly into training and test samples. These samples are then used to build a model using the CatBoostRegressor, which is specifically designed to handle categorical features. The resulting model has a root mean squared error (RMSE) of approximately 33.3 kUSD/year. This represents a significant improvement compared to the baseline model, which has an RMSE of about 58.9 kUSD/year. It’s worth noting that the baseline model assumes a consistent salary of around 89.0 kUSD/year for every record.
Step 3 — explanation of the obtained Machine Learning model.
In this study, we employ the SHapley Additive exPlanations (SHAP) method, which is widely used to investigate the interpretability of Machine Learning models. The SHAP values are thus expressed in thousands of US dollars per year (kUSD/year). This method allows us to gain insights into how different features contribute to the model’s predictions and quantify their impact on the output. By using SHAP, we can better understand the factors influencing the results and provide a clearer explanation of the model’s behavior.
First, we look into the span of SHAP values for every feature of our interest:
According to the SHAP method, several key features significantly influence the yearly total compensation. These features include:
- Developers’ country: The country in which developers are located, as indicated by the Country column.
- Years of professional coding experience: The number of years developers have spent in professional coding, is represented by the YearsCodePro column.
- Organization size: The size of the organization in which developers work, captured by the OrgSize column.
Now, let’s delve into more specific information about all available features to gain a better understanding of their impact on compensation.
How do you learn to code?
Regarding the methods used to learn coding, as discussed in the LearnCode column, it is evident that the highest annual compensations are associated with participation in Hackathons, whether virtual or in-person. The next most effective methods are learning through Books and other physical media, as well as on-the-job training:
To elaborate further, Hackathons provide a platform for developers to showcase their skills and problem-solving abilities in a competitive environment. By participating in these events, coders not only gain valuable experience but also have the opportunity to network with industry professionals and potential employers, which can significantly impact their earning potential.
Books and physical media remain reliable and accessible resources for learning coding. They offer comprehensive and structured content that allows individuals to grasp programming concepts at their own pace. Furthermore, these resources can serve as long-term references for continuous skill enhancement.
On-the-job training is another valuable method for acquiring coding knowledge. Working in a coding-related role provides practical hands-on experience, exposure to real-world projects, and the guidance of experienced professionals. This immersive learning approach enables individuals to gain proficiency in coding while actively contributing to meaningful projects within a professional setting.
It is important to note that while these methods have shown positive correlation with higher compensations, a combination of multiple learning approaches often yields the best results. Therefore, aspiring coders should consider diversifying their learning methods to enhance their skill set and increase their prospects for higher salaries and career advancement.
What online resources do you use to learn to code?
Regarding the mentioned online resources for learning to code, specifically the LearnCodeOnline column, we see that the largest yearly compensations are associated with Blogs with tips and tricks:
These blogs usually provide valuable insights and guidance for individuals seeking to enhance their coding skills. By exploring these resources, aspiring coders can gain practical knowledge and stay updated with the latest industry trends. Investing time in these blogs can significantly contribute to one’s proficiency in coding and potentially lead to lucrative career opportunities.
What online courses or certifications do you use to learn to code?
When it comes to online courses or certifications for learning coding (LearnCodeOnlineCert column), we see that the largest yearly compensations are associated with Coursera, followed by edX, Udacity, and Pluralsight:
Coursera, known for its extensive range of courses and programs, offers a wide selection of coding courses taught by experts in the field. Students who complete these courses and earn certifications have the potential to earn lucrative salaries in the coding industry.
edX is another popular online learning platform that offers coding courses and certifications. With partnerships with top universities and educational institutions, edX provides high-quality coding education that can lead to rewarding career opportunities.
Udacity is renowned for its focus on practical, hands-on learning experiences. The platform offers nanodegree programs specifically designed to equip individuals with the skills necessary to succeed in the coding industry. These nanodegree programs have been recognized for producing highly skilled professionals who command competitive salaries.
Pluralsight, known for its vast library of coding courses and tutorials, offers comprehensive learning resources for aspiring coders. By completing courses and earning certifications, individuals can position themselves for well-paying coding jobs.
Importantly, all these platforms not only provide quality coding education but also offer opportunities to network with professionals, access job boards, and participate in coding challenges and competitions.
Which programming, scripting, and markup languages have you done extensive development work in over the past year?
When it comes to programming, scripting, and markup languages (LanguageHaveWorkedWith column), the largest reported developer salaries are associated with OCaml, followed by Go, TypeScript, Scala, Apex, Clojure, Rust, Haskell, and C++:
OCaml is a statically typed functional programming language that combines functional programming paradigms with imperative and object-oriented features. It stands for Objective Caml and is widely known for its strong type inference system, which allows developers to write concise and expressive code. OCaml is highly regarded for its emphasis on safety, performance, and concurrency.
Go is a statically typed programming language known for its simplicity and efficiency. It was designed to be easy to learn and use, with built-in support for concurrent programming and robust error handling, making it a popular choice for scalable and reliable software development.
TypeScript is a statically-typed superset of JavaScript that compiles to plain JavaScript. It adds type annotations, enhanced tooling, and improved scalability to JavaScript projects, making them more robust and easier to maintain.
Scala is a statically typed programming language that combines object-oriented and functional programming paradigms. It provides concise syntax, strong type inference, and interoperability with existing Java code, making it a popular choice for building scalable and maintainable applications.
Apex is a proprietary programming language specifically designed for the Salesforce platform. It combines the syntax of Java and the database querying capabilities of SQL, allowing developers to build custom business logic and interact with the Salesforce data model.
Clojure is a dynamic, functional programming language that runs on the Java Virtual Machine (JVM). It combines the simplicity and elegance of Lisp with the power and interoperability of Java, making it an attractive choice for building robust and scalable applications.
Rust is a modern programming language known for its strong emphasis on memory safety and performance. It provides developers with the ability to write efficient, reliable, and secure software, making it increasingly popular for systems programming and building high-performance applications.
Haskell is a statically typed functional programming language known for its strong type system and lazy evaluation. It emphasizes immutability and purity, making it a popular choice for academic research and the development of high-assurance systems.
C++ is a powerful and versatile programming language known for its efficiency and high performance. It combines features of both low-level and high-level languages, making it suitable for a wide range of applications, from system programming to game development.
Which database environments have you done extensive development work in over the past year?
Among database environments (DatabaseHaveWorkedWith column), we see that the largest yearly total compensation is associated with Snowflake, Cosmos DB, CockroachDB, BigQuery, and ClickHouse:
Snowflake is a cloud-based data warehousing platform that offers a highly scalable and flexible solution for storing and analyzing large volumes of data. It provides a unique architecture that separates storage and computing, enabling efficient and cost-effective data processing. With its ability to handle diverse data types and support complex analytics, Snowflake has gained popularity as a powerful tool for modern data management and analytics in organizations.
Cosmos DB is a globally distributed, multi-model database service provided by Microsoft. It offers high scalability, low latency, and comprehensive data consistency models, making it suitable for building highly responsive and globally available applications.
CockroachDB is a distributed SQL database that is designed for scalability, resilience, and consistency. It provides ACID-compliant transactions across multiple nodes and offers horizontal scaling capabilities to handle high volumes of data and traffic.
BigQuery is a cloud-based data warehouse provided by Google Cloud. It offers a scalable and highly efficient solution for storing and analyzing large volumes of data, enabling organizations to derive valuable insights and make data-driven decisions.
ClickHouse is a columnar database management system designed for high-performance analytics processing. It is known for its ability to handle large volumes of data with fast query execution, making it popular for data warehousing and real-time analytics applications.
Which cloud platforms have you done extensive development work in over the past year?
Among cloud platforms (PlatformHaveWorkedWith column), we see that the largest yearly total compensation is associated with Fly.io and Amazon Web Services (AWS), followed by colocation, Cloudflare, Google Cloud (GCP), and Microsoft Azure:
Fly.io is a powerful cloud platform designed for deploying and running applications globally. It offers a developer-friendly environment with features like automatic scaling, load balancing, and global distribution. With its focus on simplicity and performance, Fly.io enables developers to easily deploy and manage their applications across multiple regions, ensuring reliable and efficient performance for end users.
Amazon Web Services (AWS) is a comprehensive cloud computing platform offered by Amazon. It provides a wide range of services, including computing power, storage, and databases, as well as tools for analytics, machine learning, and IoT. With its scalable and flexible infrastructure, AWS allows businesses to easily deploy and manage their applications and services in the cloud, enabling cost efficiency and enhanced productivity.
A colocation data center is a service that allows businesses to securely store and manage their physical servers and hardware in a third-party data center. It provides reliable infrastructure, connectivity, and security features, enabling businesses to focus on their core operations while leveraging the benefits of a robust and professionally managed data center environment.
Cloudflare is a robust cloud platform that provides a wide range of services, including content delivery, DDoS protection, and DNS management, to optimize and secure websites and applications. It offers businesses a scalable and reliable infrastructure to enhance their online presence and protect against cyber threats.
Google Cloud Platform (GCP) is a comprehensive cloud computing platform offered by Google, providing a wide range of services for building, deploying, and managing applications and infrastructure. It offers scalable computing power, storage options, and advanced data analytics capabilities, making it a popular choice for businesses seeking reliable and efficient cloud solutions.
Microsoft Azure is a comprehensive cloud platform that provides a wide range of services and tools for building, deploying, and managing applications and infrastructure. It offers scalability, security, and flexibility, empowering businesses to leverage the power of the cloud to drive innovation and achieve their digital transformation goals.
Which web frameworks and web technologies have you done extensive development work in over the past year?
Regarding web frameworks and web technologies (marked as WebframeHaveWorkedWith column), the largest yearly total compensations are associated with React, Elm, and Next.js:
React is a popular JavaScript library used for building user interfaces. It allows developers to create interactive and dynamic web applications efficiently. With its component-based architecture and virtual DOM, React simplifies the process of developing reusable UI components and efficiently updating the user interface based on changes in data.
Elm is a functional programming language designed for building web applications. It focuses on reliability and robustness by enforcing a strong type system and eliminating runtime errors. Elm’s architecture promotes a predictable and maintainable codebase, making it a popular choice for developers seeking a dependable and efficient web framework.
Next.js is a popular web framework built on top of React.js that simplifies the process of building scalable and performant web applications. It provides server-side rendering and automatic code splitting, allowing for faster page loads and improved SEO. With its robust ecosystem and support for serverless functions, Next.js offers a versatile and efficient solution for developing modern web applications.
Which other frameworks and libraries have you done extensive development work in over the past year?
Among other frameworks and libraries (MiscTechHaveWorkedWith column), we see that the largest total compensation is associated with Apache Spark, followed by Hugging Face Transformers, and Apache Kafka:
Apache Spark is a powerful open-source data processing framework that provides high-speed and distributed computing capabilities. It supports various data processing tasks, including batch processing, real-time streaming, machine learning, and graph processing. Spark’s in-memory processing and fault-tolerance features make it a popular choice for big data analytics and large-scale data processing applications.
Hugging Face Transformers is a popular open-source library that provides a wide range of tools for various tasks in machine learning. It provides state-of-the-art pre-trained models and a user-friendly interface for fine-tuning models on specific tasks. With its extensive support for different tasks and its large community, Hugging Face Transformers has become a popular choice for developers seeking efficient and effective machine-learning solutions.
Apache Kafka is an open-source distributed streaming platform that enables the real-time processing of high-volume data streams. It provides a highly scalable, fault-tolerant, and reliable infrastructure for handling streaming data. With its publish-subscribe messaging model and robust architecture, Kafka has become a popular choice for building data pipelines, real-time analytics, and event-driven applications.
Which developer tools for compiling, building, and testing have you done extensive development work in over the past year?
When it comes to developer tools for compiling, building, and testing (ToolsTechHaveWorkedWith column), the ones associated with the highest reported yearly compensations are Google Test, Terraform, Puppet, Homebrew, Kubernetes, LLVM’s Clang, and Dagger:
Google Test is a widely used testing framework for C++ code. It provides a comprehensive set of tools and APIs for writing and running unit tests. With its robust features and strong community support, Google Test is favored by developers seeking reliable and efficient testing solutions for their C++ projects.
Terraform is a popular infrastructure-as-code tool that enables developers to provision and manage cloud resources across various platforms. It allows for declarative configuration, making it easier to define and maintain infrastructure as code. With Terraform, developers can efficiently create and modify infrastructure setups, automating the provisioning process and ensuring consistency in their deployments.
Puppet is an automation tool that allows for the management and configuration of computer systems. It enables administrators to define desired system states and automatically enforce them, streamlining the process of managing infrastructure at scale.
Homebrew is a popular package manager designed for macOS systems. It allows users to easily install and manage software packages, libraries, and dependencies from a command-line interface, enhancing the convenience and efficiency of software management on macOS.
Kubernetes is an open-source container orchestration platform that simplifies the deployment and management of containerized applications. It enables organizations to scale and automate the management of their applications across multiple environments. With features like automatic scaling, load balancing, and self-healing capabilities, Kubernetes helps ensure high availability and efficient resource utilization.
LLVM’s Clang is a compiler front end that supports multiple programming languages, including C, C++, and Objective-C. It is known for its high performance, extensive language compatibility, and robust error diagnostics, making it a popular choice among developers for compiling code.
Dagger is a dependency injection framework for Java, Kotlin, and Android. It enables developers to efficiently manage and inject dependencies into their applications, promoting modular and scalable code.
Which development environments and collaboration tools did you use regularly over the past year?
About development environments and collaboration tools (NEWCollabToolsHaveWorkedWith column), the largest yearly compensations are associated with CLion, followed by IntelliJ IDEA and RubyMine:
CLion is a powerful integrated development environment (IDE) specifically designed for C and C++ programming languages. It offers advanced features like intelligent code completion, code analysis, and debugging capabilities to enhance developer productivity.
IntelliJ IDEA is a powerful IDE designed for Java development, offering advanced code editing, debugging, and productivity features. It provides a user-friendly and efficient environment for developers, enabling them to streamline their workflow and enhance their productivity.
RubyMine is a powerful IDE specifically designed for Ruby and Rails applications. It offers a comprehensive set of tools and features to enhance productivity and streamline the development process for Ruby developers.
What is the primary operating system in which you work?
Regarding operating systems at work (OpSysProfessional use column), we see that the largest reported yearly compensations are associated with macOS, followed by Other Linux-based, and Red Hat:
macOS is the operating system developed by Apple exclusively for its Mac computers, providing a seamless and user-friendly experience with a focus on integration, security, and a wide range of productivity and creative tools. It offers a stable and intuitive environment that caters to the needs of both casual users and professionals alike.
Red Hat Linux is a popular and widely-used open-source operating system known for its stability, security, and enterprise-level support. It provides a robust and reliable platform for businesses and individuals to run their applications and services with confidence.
Which collaborative work management and/or code documentation tools did you use regularly over the past year?
About collaborative work management and/or code documentation tools (OfficeStackAsyncHaveWorkedWith column), we see that the largest reported yearly compensations are associated with Linear, followed by Stack Overflow for Teams, Confluence, Smartsheet, and Monday.com:
Linear is a collaboration tool designed for engineering and product teams. It offers a streamlined workflow for managing and tracking projects, allowing teams to prioritize tasks, track progress, and collaborate effectively. With its intuitive interface and powerful features like issue tracking, real-time updates, and integrations with popular tools, Linear simplifies project management and enhances team productivity.
Stack Overflow for Teams is a collaboration tool designed to help organizations improve knowledge sharing and problem-solving among their teams. It provides a platform where team members can ask questions, share expertise, and access a repository of collective knowledge, enhancing productivity and fostering collaboration.
Atlassian’s Confluence is a powerful collaboration tool that enables teams to create, organize, and share content in a centralized and accessible platform. It facilitates seamless collaboration, knowledge sharing, and efficient project management, enhancing productivity and communication within organizations.
Smartsheet is a powerful collaboration tool that enables teams to streamline their work processes, track projects, and manage tasks efficiently. With its intuitive interface and comprehensive features, Smartsheet facilitates seamless collaboration and coordination among team members, improving productivity and ensuring project success.
Monday.com is a collaborative project management tool that helps teams streamline their work and track progress in real-time. It offers customizable workflows and intuitive features to enhance team communication and productivity.
Which communication tools did you use regularly over the past year?
Regarding communication tools (OfficeStackSyncHaveWorkedWith column), the largest reported yearly compensations are associated with Slack, followed by Signal and Zoom:
Slack is a cloud-based collaboration software that simplifies communication and teamwork within organizations. It offers real-time messaging, file sharing, and integrations with various productivity tools. With its user-friendly interface and powerful features, Slack helps teams streamline their workflows and improve collaboration efficiency.
Signal Messenger is a secure and private messaging app known for its end-to-end encryption and emphasis on user privacy. It offers a simple and user-friendly interface while prioritizing data protection, making it a popular choice for those seeking secure communication.
Zoom is a popular video conferencing platform that enables people to connect and communicate remotely through virtual meetings. It offers features such as video and audio capabilities, screen sharing, and chat functionality, making it a versatile tool for remote collaboration and communication.
Which AI-powered search tools did you use regularly over the past year?
About AI-powered search tools (AISearchHaveWorkedWith), we see that the largest total yearly compensation is associated with ChatGPT and Google Bard AI:
ChatGPT is a large language model (LLM) chatbot developed by OpenAI based on the GPT architecture. It is trained on a wide range of text data to generate human-like responses and assist users in various tasks. With its ability to understand and generate natural language, ChatGPT serves as a powerful tool for communication, information retrieval, and creative content generation.
Google Bard AI is a LLM chatbot developed by Google. It is based on the LaMDA language model, which has been trained on a dataset of trillions of words.
Which AI-powered developer tools did you use regularly over the past year?
Regarding AI-powered developer tools (AIDevHaveWorkedWith column), the largest reported yearly salaries are associated with GitHub Copilot and AWS CodeWhisperer:
Github Copilot is an AI-powered code completion tool developed by OpenAI in collaboration with GitHub. It integrates with popular code editors and assists developers in writing code by suggesting whole lines or blocks based on context. By leveraging machine learning, Copilot aims to boost productivity and help developers write code more efficiently.
AWS CodeWhisperer is a specialized service offered by Amazon Web Services (AWS) that helps developers enhance their coding skills and optimize their code. It provides personalized guidance and recommendations based on best practices and industry standards, enabling developers to write efficient and high-quality code.
Which professional technology tools has my company?
About professional technology tools (ProfessionalTech column), the largest yearly total compensations are associated with Observability tools, AI assisted technology tools, and Automated testing:
Which of the following best describes your current employment status?
Regarding current employment status (Employment column), we see that the largest reported compensation is associated with Independent contractor (freelancer) or self-employed status:
What is your age?
Regarding age category (Age column), the largest yearly total compensations are associated with 35–44 years old, followed by 25–34 years old, 45–54 years old, and 55–64 years old:
Which best describes your current work situation?
About remote work possibility (RemoteWork column), the largest reported yearly compensations are associated with Remote and Hybrid (some remote, some in-person) work possibilities:
Which of the following best describes the highest level of formal education that you’ve completed?
About the highest completed level of formal education (EdLevel column), we see that the largest yearly salaries are associated with Master’s degree (M.A., M.S., M.Eng., MBA, etc.), followed by Bachelor’s degree (B.A., B.S., B.Eng., etc.), and Professional degree (JD, MD, Ph.D, Ed.D, etc.):
This behaviour contradicts to the results of previous studies, so more detailed analysis is required to understand this apparent difference.
Including any education, how many years have you been coding in total?
Regarding the total years of coding (YearsCode column), we see that the largest reported salaries are associated with 39–46 years of coding:
NOT including education, how many years have you coded professionally (as a part of your work)?
Regarding the total professional years of coding (YearsCodePro column), we see that the largest reported salaries are associated with 25+ years of coding professionally:
Which of the following describes your current job, the one you do most of the time?
About the current job description (DevType column), the largest reported yearly compensations are associated with Senior Executives (C-Suite, VP, etc.), followed by Engineering managers, Site reliability engineers, Developer Experience roles, and Security professionals:
Approximately how many people are employed by the company or organization you currently work for?
About current organisation size (OrgSize column), we see that the largest yearly total compensations are associated with the largest organisations (10,000+ employees), following by organisations with 5,000–9,999 employees, and organisation with 1,000–4,999 employees:
Where do you live?
Regarding current country of residence (Country column), the largest reported salaries are associated with United States of America, followed by Switzerland, Israel, Denmark, Australia, Ireland, and Canada:
Would you like to participate in the Professional Developer Series?
Regarding responders willingness to participate in the Professional Developer Series (TBranch column), the largest yearly compensations are associated with the negative answer:
Are you an individual contributor or a people manager?
Regarding being an individual contributor or people manager (ICorPM column), we see that the largest yearly compensations are associated with being people managers:
How many years of working experience do you have?
About total years of working experience (WorkExp column), we see that the largest reported salaries are associated with 25+ years of total working experience:
What industry is the company you work for?
Finally, about company’s industry (Industry column), we see that the largest yearly total compensations are associated with Financial Services and Healthcare industries:
I hope these results can be useful for you. In case of questions/comments, do not hesitate to write in the comments below or reach me directly through LinkedIn or Twitter.
You can also subscribe to my new articles, or become a referred Medium member.