The United States technology industry continues to struggle to recruit new talent. According to the US Bureau of Labor Statistics, the number of people employed in technology is not increasing quickly.
Tech jobs pay well and don’t have the drawbacks of some other in-demand jobs, such as the travel schedule of a truck driver or the physically taxing labor required in oil fields.
Tech jobs are sometimes touted as a guarantee of having a comfortable and rewarding career, but the reality is not that simple.
Economics suggests that high wages would eliminate labor shortages, but that’s not the case in tech work. Why?
In this paper, authors Joy Buchanan and Henry Kronk propose a set of factors that have been overlooked and apply broadly to the tech sector.
Individuals with high-status tech jobs report burnout, anxiety, depression, and other mental health issues at higher rates than the general population. They also have to deal with the constant threat of becoming obsolete. Because technology changes so quickly, they must constantly work to update their skills in order to remain competitive.
The authors offer several recommendations for tech companies, educators, and policymakers:
Political and community leaders can provide more accurate messaging such as communicating clearer expectations about the difficulties of entering the tech workforce.
The tech industry could benefit from improvements in computer education. The authors cite a need for more pre-college exposure to computer occupations as well as a need to add communication skills to computer science curriculums.
Teachers, parents, and tech companies can all find ways to inform young people at an age-appropriate level about opportunities. Computer science is abstract and hard to understand. Young people who have some exposure to computer science through a class or camp are more likely to become CS majors in college.
Company leaders can improve their recruitment and development strategies to reflect the labor market realities including paying enough to compensate employees for the mental challenges of demanding technical work and alleviating their own talent shortages by investing in training and education.
Tech companies may be able to attract more women and minorities by improving their scheduling and management practices.
Henry and I examined public data and the existing literature to get a better understanding of the current state of knowledge on this issue. I hope our paper can be helpful, however we partly just highlight how many questions still exist about tech and talent.
Women did not reject a short-term computer programming job at a higher rate than men.
For the incentivized portions of the experiment, women had the same reservation wage to program. Women also seemed equally confident in their ability after a belief elicitation.
The main gender-related outcomes were, surprisingly, null results. I ran the experiment three times with slightly different subject pools.
However, I did find that women might be less likely to pursue programming outside of the experiment based on their self-reported survey answers. Women are more likely to say they are “not confident” and more likely to say that they expect harassment in a tech career.
In all three experiments, the attribute that best predicted whether someone would program is if they say they enjoy programming. This subjective attitude appears more important even than having taken classes previously.
Along with “enjoy programming” or “like math”, subjects who have a high opportunity cost of time were less willing to return to the experiment to do programming at a given wage level.
I wrote this paper partly written to understand why more people are not attracted to the tech sector where wages are high. This recent tweet indicates that, although perhaps more young people are training for tech than ever before, the market price for labor is still quite high.
The neat thing about controlled experiments is that you can randomly assign treatment conditions to subjects. This post is about what happened after adding either extra information or providing encouragement to some subjects.
Informed by reading the policy literature, I assumed that a lack of confidence was a barrier to pursuing tech. A large study done by Google in 2013 suggested that women who major in computer science were influenced by encouragement.
I provided an encouraging message to two treatment groups. The long version of this encouraging message was:
If you have never done computer programming before, don’t worry. Other students with no experience have been able to complete the training and pass the quiz.
Not only did this not have a significant positive effect on willingness to program, but there is some indication that it made subjects less confident and less willing to program. For example, in the “High Stakes” experiment, the reservation wage for subjects who had seen the encouraging message was $13 more than for the control subjects.
My experiment does not prove that encouragement never matters, of course. Most people think that a certain type of encouragement nudges behavior. My results could serve as a cautionary tale for policy makers who would like to scale up encouragement. John List’s latest book The Voltage Effect discusses the difficulty of delivering effective interventions at scale.
The other randomly assigned intervention was extra information, called INFO. Subjects in the INFO treatment saw a sample programming quiz question. Instead of just knowing that they would be doing “computer programming,” they saw some chunks of R code with an explanation. In theory, someone who is not familiar with computer programming could be reassured by this excerpt. My results show that INFO did not affect behavior. Today, most people know what programming is already. About half of subjects said that they had already taken a class that taught programming. Perhaps, if there are opportunities for educating young adults, it would be in career paths rather than just the technical basics.
Since the differences between treatments turned out to be negligible, I pooled all of my data (686 subjects total) for certain types of analysis. In the graph below, I group every subject as either someone who accepted the programming follow-up job or as someone who refused to return to program at any wage. Recall that the highest wage level I offered was considerably higher on a per-hour basis than what I expect their outside earning option to be.
I’ll discuss the three features in this graph in what appear to be the order of importance for predicting whether someone wants to program. There was an enormous difference in the percent of people who were willing to return for an easy tedious task that I call Counting. By inviting all of these subjects to return to count at the same hourly rate as the programming job, I got a rough measure of their opportunity cost of time. Someone with a high opportunity cost of time is less likely to take me up on the programming job. This might seem very predictable, but this is a large part of the reason why more Americans are not going into tech.
Considering the first batch of 310 subjects, I have a very clean comparison between the programming reservation wage and the reservation wage for counting. People who do not enjoy programming require a higher payment to program than they do to return for the counting job. Self-reported enjoyment is a very significant factor. The orange bar in the graph shows that the majority of people who accepted the programming job say that they enjoy programming.
Lastly, the blue bar shows the percent of female subjects in each group. The gender split is nearly the same. As I show several ways in the paper, there is a surprising lack of a gender gap for incentivized decisions.
I hope that my experiment will inspire more work in this area. Experiments are neat because this is something that someone could try to replicate with a different group of subjects or with a change to the design. Interesting gaps could open up between subject types under new circumstances.
The topic of skill problems in the US represents something reasonably new for labor market and public policy discussions. It is difficult to think of a labor market issue where academic research or even research using standard academic techniques has played such a small role, where parties with a material interest in the outcomes have so dominated the discussion, where the quality of evidence and discussion has been so poor, and where the stakes are potentially so large.
Cappelli, PH, 2015. Skill gaps, skill shortages, and skill mismatches: evidence and arguments for the United States. ILR Rev. 68 (2), 251–290.
I am pleased to announce that my paper “Willingness to be Paid: Who Trains for Tech Jobs?” has been accepted at Labour Economics.
Having a larger high-skill workforce increases productivity, so it is useful to understand how workers self-select into high-paying technology (tech) jobs. This study examines how workers decide whether or not to pursue tech, through an experiment in which subjects are offered a short programming job. I will highlight some results on gender and preferences in this post.
Most of the subjects in the experiment are college students. They started by filling out a survey that took less than 15 minutes. They could indicate whether or not they would like an invitation for returning again to do computer programming.
Subjects indicate whether they would like an invitation to return to do a one-hour computer programming job for $15, $25, $35, …, or $85.This is presented as 9 discrete options, such as:
“I would like an invitation to do the programming task if I will be paid $15, $25, $35, $45, $55, $65, $75 or $85.”,
“I would like an invitation to do the programming task if I will be paid $85. If I draw a $15, $25, $35, $45, $55, $65 or $75 then I will not receive an invitation.”,
and the last choice is
“I would not like to receive an invitation for the programming task.”
Ex-ante, would you expect a gender gap in the results? In 2021, there was only 1 female employee working in a tech role at Google for every 3 male tech employees. Many technical or IT roles exhibit a gender gap.
To find a gender gap in this experiment would mean female subjects reject the programming follow-up job or at least they would have a different reservation wage. In economics, the reservation wage is the lowest wage an employee would accept to continue doing their job. I might have observed that women were willing to program but would reject the low wage levels. If that had occurred, then the implication would be that there are more men available to do the programming job for any given wage level.
However, the male and female participants behaved in very similar ways. There was no significant difference in reservation wages or in the choice to reject the follow-up invitation to program. The average reservation wage for the initial experiment was very close to $25 for both males and females. A small number of male subjects said they did not want to be invited back at even the highest wage level. In the initial experiment, 5% of males and 6% of females refused the programming job.
The experiment was run in 3 different ways, partly to test the robustness of this (lack of) gender effect. About 100 more subjects were recruited online through Prolific to observe a non-traditional subject pool. Details are in the paper.
Ex-ante, given the obvious gender gap in tech companies, there were several reasons to expect a gender gap in the experiment, even on a college campus. Ex-post, readers might decide that I left something out of the design that would have generated a gender gap. This experiment involves a short-term individual task. Maybe the team culture or the length of the commitment is what deters women from tech jobs. I hope that my experiment is a template that researchers can build on. Maybe even a small change in the format would cause us to observe a gender gap. If that can be established, then that would be a major contribution to an important puzzle.
For the decisions that involved financial incentives, I observed no significant gender gaps in the study. However, subjects answered other questions and there are gender gaps for some of the self-reported answers. It was much more likely that women would answer “Yes” to the question
If you were to take a job in a tech field, do you expect that you would face discrimination or harassment?
I observed that women said they were less confident if you just asked them if they are “confident”. However, when I did an incentivized belief elicitation about performance on a programming quiz, women appear quite similar to men.
Since wages are high for tech jobs, why aren’t more people pursing them? The answer to that question is complex. It does not all boil down to subjective preferences for technical tasks, however in my results enjoyment is one of the few variables that was significant.
People who say they enjoy programming are significantly more likely to do it at any given wage level, in this experiment.
Figure 3 from the paper shows the reservation wage of participates from all three waves. Subjects who say that they enjoy programming usually pick a reservation wage at or near the lowest possible level. This pattern is quite similar whether you are considering males or females.
Interestingly, enjoyment mattered more than some of the other factors that I though would predict willingness to participate. About half of subjects said they had taken a class that taught them some coding, but that factor did not predict their behavior in the experiment. Enjoyment or subjective preferences seemed to matter more than training. To my knowledge, policy makers talk a lot about training and very little about these subjective factors. I hope my experiment helps us understand what is happening when people self-select into tech. Later, I will write another blog about the treatment manipulation and results, and perhaps I will have the official link to the article by then.
Buchanan, Joy. “Willingness to be Paid: Who Trains for Tech Jobs.” Labour Economics.
 We use a quasi-BDM to obtain a view of the labor supply curve at many different wages. The data is not as granulated as that which a traditional Becker-DeGroot-Marschak (BDM) mechanism obtains, but it is easy for subjects to understand. The BDM, while being theoretically appropriate for this purpose, has come under suspicion for being difficult for inexperienced subjects to understand (Cason and Plott, 2014). We follow Bartling et al. (2015) and use a discrete version.
Last year, our economics department launched a data analytics minor program. The first class is a simple 2 credit course called Foundations of Data Analytics. Originally, the idea was that liberal arts majors would take it and that this class would be a soft, non-technical intro of terminology and history.
However, it turned out that liberal arts majors didn’t take the class and that the most popular feedback was that the class lacked technical challenge. I’m prepping to teach the class and it will have two components. A Python training component where students simply learn Python. We won’t do super complicated things, but they will use Python extensively in future classes. The 2nd component is still in the vein of the old version of the course.
I’ll have the students read and discuss “Big DataDemystified” by David Stephenson. He spends 12 brief chapters introducing the reader to the importance of modern big data management, analytics, and how it fits into an organization’s key performance indicators. It reads like it’s for business majors, but any type of medium-to-large organization would find it useful.
Davidson starts with some flashy stories that illustrate the potential of data-driven business strategies. For example, Target corporation used predictive analytics to advertise baby and pregnancy products to mothers who didn’t even know that they were pregnant yet. He wets the appetite of the reader by noting that the supercomputers that could play Chess or Go relied on fundamentally different technologies.
The first several chapters of the book excite the reader with thoughts of unexploited potentialities. This is what I want to impress upon the students. I want them to know the difference between artificial intelligence (AI) and machine learning (ML). I want them to recognize which tool is better for the challenges that they might face and to see clear applications (and limitations).
AI uses brute force, iterating through possible next steps. There are multiple online tic-tac-toe AI that keep track records. If a student can play the optimal set of strategies 8 games in a row, then they can get the general idea behind testing a large variety of statistical models and explanatory variables, then choosing the best.
But ML is responsive to new data, according to what worked best on previous training data. There are multiple YouTubers out there who have used ML to beat Super Mario Brothers. Programmers identify an objective function and the ML program is off to the races. It tries a few things on a level, and then uses the training rounds to perform quite well on new levels that it has never encountered before.
There are a couple of chapters in the middle of the book that didn’t appeal to me. They discuss the question of how big data should inform a firm’s strategy and how data projects should be implemented. These chapters read like they are written for MBAs or for management. They were boring for me. But that’s ok, given that Stephenson is trying to appeal to a broad audience.
The final chapters are great. They describe the limitations of big data endeavors. Big data is not a panacea and projects can fail for a variety of what are very human reasons.
Stephenson emphasizes the importance of transaction costs (though he doesn’t say it that way). Medium sized companies should outsource to experts who can achieve (or fail) quickly such that big capital investments or labor costs can be avoided. Or, if internals will be hired instead, he discusses the trade-offs between using open source software, getting locked in, and reinventing the wheel. These are a great few chapters that remind the reader that data scientists and analysts are not magicians. They are people who specialize and can waste their time just as well as anyone else.
Overall, I strongly recommend this book. I kinda sorta knew what machine learning and artificial intelligence were prior to reading, but this book provides a very accessible introduction to big data environments, their possible uses, and organizational features that matter for success. Mid and upper level managers should read this book so that they can interact with these ideas prudentially. Those with a passing interest in programming should read it for greater clarity and to get a better handle on the various sub-fields. Hopefully, my students will read it and feel inspired to be on one side or the other of the manager- data analyst divide with greater confidence, understanding, and a little less hubris.
Courtesy of the St. Louis Fed, you can download a report published in 1958 titled “Automation and Employment Opportunities for Office-Workers: A Report on the Effect of Electronic Computers on Employment of Clerical Workers, with a Special Report on Programmers.”
I teach students about data and software to prepare them to enter the hot field of business analytics. It has been a growing field for a few years, especially since the advent of “Big Data”. Something I explain in class is how brand-new technology has changed business.
Reading this report forced me to re-think just how new data analytics is. The authors saw machines in use for data processing and correctly predicted that this would be a dynamic source of new jobs.
The introduction states that millions of “clerical workers” were employed in the United States. That fact would have been obvious at the time, but today we might not realize just how many humans would be needed to store and fetch the files we access regularly on our computers. The creation of clerical jobs was especially important for women.
In view of the volume of work that needed to be done, installing new computers was economical. “A computer system can automatically do such jobs as prepare payrolls for thousands of employees, control inventory on a multitude of items…”
“Although computers are often described as machines that can “think,” that is, of course, not so. Like other machines, they must be operated or controlled by people… The people who prepare the instructions are called programmers.”
“Electronic computers were developed during World War II as an aid in solving intricate scientific and engineering problems such as gunfire control, but their application to the processing of office data is more recent. The Federal Government lead the way in 1951, when an electronic computer was installed by the Bureau of the Census…”
The authors see the primary role of computers in business as a way to automate the routine work that could be performed by clerks. Secondly, they state that computers can by used for solving complex math problems “such as those related to launching and tracking earth satellites.”
The report was created for young people who are considering their own choices for education and careers. The authors describe the programming but also various machine support roles. For example, the Coding Clerk’s job is to convert the programmers’ instructions into “machine language”.
The authors recognize that computers will replace some of the traditional clerk roles. “These developments will not only increase the output of clerical workers and slow down growth in clerical employment, but will also change the character of many jobs… Many of the new jobs … will generally pay better and require higher levels of skill and training than most other clerical jobs.” The next sentence is where the authors fail to predict PCs and the internet: “Moreover, a continued increase is expected in the number of officeworkers in jobs not greatly affected by office automation – for example, secretary, stenographer, messenger, receptionist, and others involving contacts with customers and the public.”
The discussion of women in the workplace is clinical in tone. Turnover is high in the clerical fields because many young women stop working when they get married or have children.
There is a special report on “programmers”, one of the newest occupations in the country. Programmers specialize in either of the following: 1) “processing the great masses of data which have to be handled in large business and government offices” 2) “solving scientific and engineering problems”.
The authors describe typical training and career paths. At the time, a college student could not major in computer science. Companies were filling most positions by selecting employees familiar with the subject matter and giving them training in programming. A few colleges purchased computers and provided some training opportunities.
The culture was different back then. “Although many employers recognize the ability of women to do programming, they are reluctant to pay for their training in view of the large proportion of women who stop working…” The authors tip off their female readers that they are more likely to get training in government than industry, if they aspire to be programmers in the 1950’s. Today, the risk and cost of training has largely shifted from the employer to the worker. If you are interested in the topic of bootcamps and STEM pipelines, read the document for their discussion of education.
These authors made a good long-term prediction because they anticipated the business analytics boom. “Continued expansion in employment of programmers is expected over the long run… In offices where the volume of recordkeeping is great, there will continue to be need to reduce the cost of processing tremendous amounts of data and to produce more timely reports on which management decision can be based.” After explaining salary, they talk about perks: “Programmers usually work in well lighted, air-conditioned, modern offices. Employers make special efforts to provide better than average surroundings for programmers, so that they may concentrate to achieve the extreme accuracy necessary for programming.” The nap pods of Silicon Valley have a long history that can be traced back to the Census Bureau.
Andrew Weaver is doing interesting work on “the skills gap.” One of his key methods is to create new data by interviewing firms. As someone who has looked hard for good data on the skills gap, I can say that we need more work like his.
Weaver’s 2017 paper with Paul Osterman is about data for U.S. manufacturing firms. These findings may or may not generalize perfectly outside of manufacturing, but I think this was a great place to start. There is plenty of talk about the decline of U.S. manufacturing and at least some of the talk was about a lack of skilled Americans to meet the great demand for high-tech doings. For this survey, they only ask about “core workers” who are doing the specialized roles of making widgets.
Here are two important empirical questions: a.) do American manufacturing firms want high-skill workers? b.) do they have trouble finding them? The authors answer, “not as much as you might think from policy discussions.”
There are lots of details in the paper that I don’t have time to cover. In table 2, they go over the determinants of a firm facing long-term vacancies. What is common among the (minority of) firms that report having long-term vacancies? Advanced computer proficiency is not associated with difficulty of filling jobs. The implication is that most manufacturing companies around 2017 were able to find workers who had the computer-related skills needed to do the core production tasks. What seemed to be a limiting factor was not computer skills but advanced reading skills. Half of the establishments surveyed said that they require workers with extended reading skills. That could mean, for example, reading a 10-page technical article in a trade journal.