How data analysts and data science specialists find jobs after courses: the experience of the Workshop

“Data Analyst” in the Workshop. Most of our students indicate successful employment in a new profession as their learning goal. Therefore, we carefully monitor which graduates managed to get the desired position, how long it took and how much effort it required. We build our product promises to students based on this data.

At the end of last year, we came up with the idea to identify the key factors influencing student employment and learn how to create more accurate and differentiated expectations. It was assumed that we would be able to create a form, by filling out the fields of which the student would receive, in addition to general statistics, a prediction specifically for himself.

What did we expect

According to the plan, this form could work like this: for example, student Anna indicates that she is 34 years old, she does not have a specialized education and relevant work experience. In response, she receives a prediction: “90% of graduates with the same profile found a job in less than six months, making 200 applications for vacancies, solving 30 test tasks and passing six interviews. The course might be more suitable for you “Data Analyst Plus” “This choice will reduce your job search time by three weeks and 46 applications.”

And Anna really thinks about whether she is ready for such work. But if you decide to try, you don’t expect a miracle and prepare to apply for 200 vacancies within six months.

And 25-year-old student Vasilisa with a specialized education and two pet projects in her portfolio receives the following prediction: “Graduates with the same profile in 90% of cases find a job in less than a month, having made 10 responses, solved two test tasks and passed two interviews.” . And Vasilisa happily begins to study.

Such promises would be more honest – and students would know in advance what to expect. And the most prepared of them would receive attractive forecasts that would motivate them to study in the Workshop.

To learn how to draw such conclusions, we conducted a study.

What data did we have?

In our work, we used two data sets.

The first set is data from the Workshop’s career tracker. This is a service where graduates and students (access is provided shortly before the end of their studies) can:

• see potentially interesting and suitable vacancies, and also respond to them,

• enter information on vacancies found in other sources,

• enter your results – responses sent, invitations received for interviews and tests completed,

• receive statistics on the progress of employment.

Graduates participating in the acceleration program are required to regularly provide job search statistics. The rest can do this if they wish – some students track their progress outside of acceleration due to the convenience of the service.

The second source of information is data on participation in projects from real companies in the Workshop Workshop. Together, the datasets helped us figure out how the presence of unique projects in the portfolio affects the job search.

Both datasets contained data on the employment of graduates and students – most of them graduated from courses of standard length (seven to eight months) in their specialties “Data Analyst” And “Data Science Specialist”. There were almost no graduates of long plus courses or short bootcamps, so we abandoned the idea of ​​​​tracking the effect of course length on the speed and complexity of employment. The sample also included several course graduates “Systems Analyst”, “Data Engineer” And “Business analyst”.

In both files, graduates' education and work experience information was stored as a categorical value. To calculate the correlation, the text description was converted into numbers.

Education categories were assigned the following values:

And the categories of experience are:

How much effort and time does the average graduate spend on finding a job?

The complexity and duration of employment in our dataset is described by the following variables:

• search duration in days;

• number of responses to vacancies,

• number of interviews completed,

• number of solved test tasks.

As can be seen from the data, the distributions of these values ​​are not normal – that is, the average values ​​are far from the median. Moreover, there are outliers in the distributions, which means that median indicators characterize them much better than simple averages.

According to the download from the database, three graduates managed to find work in three, five and six days. Obviously, you shouldn’t rely on them. As well as extremely long searches of 495 days (more than a year!) or extremely mass mailing of resumes (633 responses).

According to the median data, a graduate looks for a job for a little longer than two months, makes 57 applications, undergoes three interviews and solves three test tasks.

Approximately 40% of graduates had to put in more effort – they looked for a job for no more than six months, made no more than 203 applications, took no more than eight tests and passed no more than eight interviews. It turns out that for some, six months of searching and 200 responses is a harsh reality.

The median scores for data analysts and data scientists are almost the same. Data scientists need to make 10–11 fewer responses, and their job search is five to six days shorter.

What influences the speed of employment

Let's build a correlation matrix based on calculations of the Spearman correlation coefficient – we will get such a heat map.

The closer the number at the intersection of two different indicators is to one, the stronger the positive correlation and the more strongly as one parameter grows, the other will grow. If the correlation is close to zero, there is no connection between the parameters and they are completely independent. If the correlation is negative, then we will be interested in indicators close to −1. Such a correlation means that as one parameter increases, the other will decrease.

Let's start with the obvious conclusions. The number of responses has an average correlation with the number of interviews and the number of tests. Those who apply for jobs get to interviews and tests (and how would we know that without our research?).

The search period in days also has a moderate correlation with the number of responses. And weak – with the number of interviews and the number of tests. Anyone who has been looking for a job for a long time has a low conversion of responses into interviews and tests and therefore must make a lot of responses. It turns out that the most difficult thing in finding a job is getting an interview.

Surprisingly, the correlation of target variables with education and work experience is very close to zero. The graduate’s work experience and education prior to (!) courses do not affect the search period and the amount of effort expended. Several hypotheses can be put forward to explain this fact:

• Potential students with relevant experience and education are looking for a completely different job. Their demands for pay and other working conditions are higher. And the higher the requirements for a future employer, the longer, other things being equal, the job search will take.

• Students without relevant education and work experience understand their disadvantages and therefore make the most of the opportunities of the career track in the Workshop: they do not neglect advice on creating a resume and portfolio, and actively respond to vacancies. More prepared students prepare their resumes and portfolios worse, spend less time writing cover letters, and are less likely to respond to job openings.

Unfortunately, we cannot test these hypotheses. In any case, our general promise of average time to employment holds true for students of a wide range of educational levels and experience – but participants may differ in their approach to offer selection, degree of activity and tactics in applying for employment.

If age has a correlation with the period of job search, it is very weak. Let's look at this connection in more detail.

After 35 it's too late… or not?

Here's what the median job search time looks like depending on the candidate's age.

It can be seen that the linear trend is positive, but sharp jumps in the metric are also visible. As a result, we have a typical graph in the style of “I didn’t understand anything, but it’s very interesting.” The reason for this behavior of the metric is the very small samples of candidates at some ages. It is from such samples that we obtain abnormally large and abnormally small values.

Let's get rid of them by combining candidates of similar age into groups. Unfortunately, we will have to remove candidates under 20 and over 45 years of age, since the number of candidates in the groups 15–19, 49–54 and 54–59 years old remains extremely small even after the merger.

Great! This graph shows that “40 is the new 30.” The median duration of job search at the ages of 30 to 35 and 35 to 40 is almost identical and amounts to 2.5 months.

But the ease of finding a job at the age of 20–24 is surprising. As a rule, candidates from this group do not have completed higher education, are often students and cannot work full time. Why do candidates from this age category find jobs faster than candidates from the 25 to 30 year old category? Maybe they are sending a crazy amount of job applications?

No, this is clearly not the reason. In the figure, we see a constantly growing trend – more senior candidates are having to make more applications.

Perhaps the main reason for the length of employment is not explained by the preferences of employers, but by the flexibility of job seekers. I'll try to explain with my own example. When I “entered IT”, I was already married, I had two children and we lived in a good apartment. Almost immediately after the search began, I received an offer from a good employer, but on the other side of the city. The salary for a junior position did not justify the situation “mom leaves at dawn and returns closer to midnight.” It was also too expensive to move the whole family with a change of kindergarten, nanny, apartment and husband’s job. Therefore, I refused the employer and extended my job search for another two months. If I were 10 years younger, I would simply change rental housing and agree.

The older the applicant, the higher the likelihood of having a family and children, which, following this hypothesis, can influence the job search in two ways:

• the applicant has a reduced range of offers that he can accept,

• the applicant can afford a longer search, since there may be another working adult in the family.

How many projects should you have in your portfolio to find a job faster?

We did not have information about the number of works in the students' portfolios, but we did have information about the projects successfully completed by students in the Workshop. We had very high expectations from this study. It was assumed that we would be able to accompany the announcement of the start of a new project with a phrase like “Finish this project – and reduce your job search time by 10 days!” Well, what we got in the end can be seen in the figure.

It must be recognized that not every graduate successfully completes four or more projects in the Workshop, so the three right-hand columns of the graph visualize the median value over extremely small samples. Let’s not take them into account (and draw conclusions about the terrifying impact of the fourth project). The good side of this fact is that for most students, no more than three projects are enough for successful employment.

Perhaps working on projects in the Workshop takes up time and candidates simply don’t have time to apply for vacancies? Let's look at the median number of responses.

Trying not to look at the last three columns (extremely small samples!), we note that each project successfully completed in the Workshop increases the number of responses required for employment by 20–30 pieces.

This strange dynamic can be explained by the following factors:

• the weaker the graduate, the worse his position in the market, the more projects in the Workshop he needs to complete for successful employment,

• projects in the Workshop help you get the best offer, and the better the vacancy, the more time and effort you need to spend on getting it,

• graduates solve many projects in the Workshop when they do not feel ready for real employment.

I remember the period when many graduates of the course “Data Analyst” immediately after graduation we started studying on the course “Data Science Specialist”. To the question “why?” They answered that they had only studied for six months and did not yet feel like real analysts. Although they even had a state-issued retraining diploma.

The hypothesis “the weaker the graduate, the worse his position in the market, the more projects in the Workshop he needs to complete for successful employment” is good because you can try to refute it. To do this, we will derive a normalized distribution of graduates’ work experience levels by the number of projects completed in the Workshop.

At least our hypothesis is not refuted. We see that graduates with relevant experience either do not complete projects in the Workshop at all, or complete one, and extremely rarely, two projects. And the proportion of graduates with exclusively academic experience increases with the number of completed projects.

What month is best to start looking for a job?

We looked at many characteristics of applicants. Let's see if our data can tell us something about the labor market: for example, how the difficulty and speed of employment changes depending on the season.

We see that those who start their search in April, May and October find work the fastest. Those who started looking for work in January, July, August, November and December spend the longest time looking for work. In many ways this is expected. The duration of the search can be explained by the January holidays and summer holidays.

In October, on the contrary, all plans for the next year are more or less clear, and everyone is in a hurry to fill vacancies before the holidays. But how can we explain the low metric values ​​in April and May? Maybe candidates respond very actively to vacancies during these periods?

If the short period of job search, which began in May, can be explained by the increased activity of candidates, then those who started their search in April clearly did not overwork themselves. It turns out that the best months to start looking for a job are April and October. If you start looking at this time, you will spend less energy and time. We don’t recommend starting your job search in December.

Let's sum it up

The result of the study did not meet our expectations. We cannot adjust our product promise to students depending on their age, experience or education, since the speed and complexity of employment does not depend or almost does not depend on these parameters.

At the same time, as a result of the study, we were able to obtain several useful conclusions:

• Median course graduate “Data Analyst” And “Data Science Specialist” finds a job after two months of searching, 57 job applications, three interviews and three test assignments.

• Previous work experience and education do not influence the length and complexity of employment, but may influence the characteristics of the job found.

• You can start your journey in IT at any age. 19 is not too early, 40 is not too late. But the sooner you do this, the less time and effort you will spend on it.

• The best months to start looking for a job are April and October. If your market launch is in December, be prepared that the search may take about a month longer. Consider seasonality when planning the start of training or planning to take a break.

• For successful employment, two or three projects in your portfolio will be enough.