There's been a lot of buzz around job postings, or "real-time LMI" lately, and EMSI has been hard at work crafting new data to meet our clients' needs. This presentation describes the background and sources behind EMSI's new Job Postings Analytics.
3. GOALS
- Provide the background and philosophy
- Review the data
- Help you walk away with an understanding
of the behind the scenes “WHY” and “HOW”
of our data process.
- Prep you for looking at the data release on
July 10
4. INTRODUCTION
- Since 2000 – working with structural LMI
• Collected and published by gvt agencies
- Data to help understand nature of people,
economies and work
- Strategic, long-term, big picture analysis
- New need to help “drill down” into more
tactical decisions (Job posting data or “real-
time LMI”)
7. INTRODUCTION
“Measuring the economy in real time is hard.
Ignoring the jobs report is foolish, but
overreacting is equally foolish.”
-Matt Yglesias, nnn
8. INTRODUCTION
On Canada’s misuse of job postings data:
“With these sites removed from the source data, the
government’s latest labour market report points to
a job vacancy rate of 1.5%, which is dramatically
less than the 4% vacancy rate Finance Canada
warned of on budget day.”
9. INTRODUCTION
Big Data
– We sometimes throw caution to the wind
– Huge pressure to understand “demand”
We need to know:
– Where does it come from?
– What can it tell us?
– How should we use it / interpret it?
10. INTRODUCTION
Basic Rules
–Data is a tool—we need to use it properly
–Data, like any tool, has limits
–Data is exciting because it illuminates things
we have never been able to look at before:
MICROSCOPE
–Data must talk to other data. Cannot be an
island.
11. The New Guys
• Job Postings Analytics (JPA)
• Hires
• Separations
• Job Churn
12. The New Guys
• Job Postings Analytics (JPA)
• Hires
• Separations
• Job Churn
13. JOB POSTING
ANALYTICS
Analyzing job ads has gotten a lot of press lately.
We’ve looked into it, found some good things to pull
out, and brought you the cream of the crop.
14. HIRES, JOB POSTINGS, AND
JOB GROWTH FOR AN
AVERAGE MONTH
(Based on data from
2013 and 2014)
15. HIRES, JOB POSTINGS, AND
JOB GROWTH FOR AN
AVERAGE MONTH
(Based on data from
2013 and 2014)
16. HIRES & JOB
CHURNSeparations
– Separations account for workers leaving a
business for any reason:
•Quitting, layoffs, firing, retiring
-It won’t be in the upcoming Analyst release
• Where to put it
• How to use it
• We know it’s helpful and want to present it
correctly
17. HIRES & JOB
CHURN
Hires helps us bridge the gap between structural LMI
and JPA.
A hire is counted whenever someone shows up on a
company’s payroll when they didn’t show up in the
previous quarter. A hire is a hire.
18. HIRES & JOB
CHURNHires vs. Annual Openings:
– They both measure demand however…
•Annual Openings is a modeled number, based off of
national projections
•Hires is informed by Quarterly Workforce Indicators (QWI),
an actual headcount of hirings that have taken place
19. HIRES & JOB
CHURN
Hires & Separations Sources:
JOLTS
+ Timely
- Less accurate
QWI
-Slower
+ More accurate
20. HIRES & JOB
CHURNChurn
Churn (hires and separations) is the bridge
between Job Postings and traditional LMI,
and why we spent so much time finding a
good source for it in QWI.
24. JOB POSTINGS
What are Job Postings good for?
-Who’s hiring?
• Company-specific info
-What positions are they looking for?
• Detailed job titles
25. JOB POSTINGS
What are Job Postings good for?
-Who’s hiring?
• Company-specific info
-What positions are they looking for?
• Detailed job titles
-What skills do those people need?
• Skills data
26. JOB POSTINGS
What are Job Postings good for?
-Who’s hiring?
• Company-specific info
-What positions are they looking for?
• Detailed job titles
-What skills do those people need?
• Skills data
-How badly do they want to hire?
• Posting Intensity
27. JOB POSTINGS
Posting Intensity
– Let’s say a trucking company needed to hire 15 drivers
– They start off posting the jobs on their site, but don’t get
them filled
– A recruiter advises them to put postings on the local WIB
site, community boards, CareerBuilder.com, Monster.com,
and Indeed.com
– Depending on how you counted them, all of a sudden it
looks like there are 1,000 open truck driver jobs
– The number of total postings put out compared to the
number of unique postings meant that trucking company
really, really wanted to find people for those jobs
28. JOB POSTINGS
Limitations of Job Postings:
– Voluntary
•Jobs are posted at the will of individual companies, no
standardization on descriptions or timeframes
•Some sectors see steady hiring with NO job postings
•High-tech, computer related occupations usually have many
more postings than available jobs
29. JOB POSTINGS
Limitations of Job Postings:
– A posting doesn’t always equal a job
•A posted job isn’t necessarily a hire waiting to happen
•Postings are collected and sorted automatically, so
reliability is somewhat dependent on the spidering
technology used
•Even when the technology is great, under and
overrepresentation of different sectors will still produce
inaccurate numbers
30. JOB POSTINGS
What’s EMSI doing about it?
“Measuring the economy in real time is hard.
Ignoring the jobs report is foolish, but
overreacting is equally foolish.”
-Matt Yglesias, nnn
31. JOB POSTINGS
What’s EMSI doing about it?
– Providing the context
•EMSI’s job postings are presented alongside our trusted LMI
data, with our new hires figures.
•Remember that these are actual jobs that have been
counted through mandatory government collection.
•LMI helps “reality check” job postings figures
32. JOB POSTINGS
EMSI’s Job Postings Come From:
– CareerBuilder’s job postings database
– Scraping the web for individual postings
– Third-party aggregators
34. JOB POSTINGS
Job Posting De-duplication
– By City
– By O*NET (job description)
– By Company
•If we find multiple postings for the same job title, in the
same city, for the same company, we reduce all of those
postings to “1” posting
39. SUMMARY
• Hires, Separations & Job Churn
– Better measure of job demand
– Helps JP’s better inform LMI
– EMSI data, new sources from QWI
– Churn is the lens that brings JP’s into focus with LMI
• JPA
– Good for finding who’s hiring, what skills/positions they want,
and how badly they want to hire
– Need context to be helpful, which we’re providing in Analyst
– Comes from CareerBuilder, web scraping, and other aggregators
Hi everyone. Thanks for joining us today. My name is Rob Sentz -- I head up the marketing efforts here at EMSI. I’m joined by Matt Gaither, who directs our training and certification programs and Deacon James, who manages all of EMSI’s data, which informs our products. Today we want to talk to you about some new data we have been working on. This will serve as a precursor for the update we are going to make to Analyst in July. Before we make that update we think it is important to give you a behind the scenes look at our thought process and some of the data we are implementing. We want to tell you why we are doing what we are doing, how we are doing it, and what the data is.
Here is a quick outline for the call. After the introduction we will talk about Hires and Job churn, followed by job postings. Once we are done with the data discussion, we will walk through a quick illustration of the data and present a summary at the end.
So, if you walk away with anything today we want you to, first and foremost, understand the background and general philosophy of why we are doing what we are doing. We think that if you have that background, you will be a better user of the data.
We also want to prep you for what you will start to see in July. This is also a good way for us to field some questions you have about the data in preparation for that release.
So, just as a bit of background here. We have been working with structural labor market information for the past 14 years. Structural labor market data or information (LMI) helps us understand the total employment picture across all industries and occupations. It is data on people who are currently employed in specific industries and occupations, and it comes from government sources such as the Bureau of Labor Statistics and Census Bureau. These sources collect data via administrative records from employers (the BLS’s QCEW, for example, includes all workers covered by unemployment insurance), information from tax returns, or surveys. Most of this data is required to be reported to the government but also can lag six months to a year. The data is standardized across the entire nation, allowing researchers to perform in-depth analysis of our huge economy.
The data is inherently related to the more strategic, big picture analysis and perspectives. This is because labor market information doesn’t shift and change day-to-day and represents millions and millions of jobs.
We also want to prep you for what you will start to see in July. This is also a good way for us to field some questions you have about the data in preparation for that release.
Here is a quick look at the US labor market.
The blue represents how many jobs are in our economy. 140 M (the green) of those jobs are held by “employees” or people who work for employees and are therefore covered by unemployment insurance. The orange represents the 10M self-employed jobs in the US. These are jobs where the individual primarily works for themselves and does not employ anyone else.
As we look at other data sources we always want to look at how it relates to this structured labor market.
We also want to prep you for what you will start to see in July. This is also a good way for us to field some questions you have about the data in preparation for that release.
Here is a look at the 20 sectors that make up our labor market. It is grouped from least to greatest. Utilities is the smallest industry sector, followed by mining and agricultre. The biggest sector by far is government.
So, as we march closer to reviewing the data on hires and job postings we want to frame it in a specific way. For the better part of two years we have been working with new datasets that are much harder to deal with than the structural LMI. A lot of this has to do with the fact that the data is harder to filter, refine, and interpret because it is flying at your much faster and it doesn’t have the same structure. Matt Yglesias recently made an insightful comment that we feel really captures the issue at hand : “Measuring the economy in real time is hard. Ignoring the jobs report is foolish, but overreacting is equally foolish.”
We couldn’t agree more. What many refer to as “real-time” data can be quite difficult to work with AND to interpret in the correct ways. We don’t want to simply ignore the data but we don’t want to misinterpret or overreact. This is where we are at with much of the data being collected on job postings.
Case in point: In May, The Globe and Mail and other news outlets in Canada reported on a very real world example of this. The Canadian government had used what turned out to be flawed job posting data that inflated the country’s job vacancy rate and led to large-scale policy changes.
“The Conservative government has quietly adjusted its labour data to ignore job postings from Kijiji and similar websites, a change that essentially erases the dire warnings of labour shortages that Ottawa has used as justification for expanding the controversial temporary foreign worker program. With these sites removed from the source data, the government’s latest labour market report points to a job vacancy rate of 1.5 per cent, which is dramatically less than the 4 per cent vacancy rate Finance Canada warned of on budget day.”
http://www.theglobeandmail.com/report-on-business/economy/jobs/ottawa-adjust-labour-data-raising-questions-about-national-skills-shortage/article18457198/
This story makes it clear: data certainly has an ability to tell a story. In this case, data was used to tell a false story that quickly led to some poor decisions. The article points out a crucial fact about job posting information: Lots and lots of companies might be advertising for jobs, which is a great indication of “intention to hire,” but this is very different from the number of people they actually hire. If you do not understand that distinction, and are primarily looking at the number of postings being pushed across hundreds of job websites for your employment figures, you could be making some very bad conclusions from that data if it isn’t nuanced with other perspectives. Given this, what exactly is data’s legitimate role? And how do we know we are working with reliable numbers?
For professionals working in higher education, public workforce investment, economic development, and strategic workforce planning, this is an all-to-familiar issue. We are tasked with basing many (if not all of) our decisions on data. This creates a tendency to rush in too quickly—to just get one’s hands on the data without asking where the data is coming from and ideally what it is telling us. Now more than ever we must know how to balance or approach and have access to the right data for our decisions.
Over the past year, we have spent a good deal of time talking (LINK) about what defines a job and the differences between jobs from a labor market perspective and jobs from a hiring/recruitment perspective. The two perspectives are bolstered by two primary and very different datasets. So as we review that data we want to know where it comes from, what it can tell us and how we should use it
As we do this – here are some good guidelines.
First, data is, after all, just a tool that helps accomplish some sort of goal. To achieve the goal, we have to use the right tool in the right way for the right job.
Second, data is just data and it has it’s limits. When we know those limits and boundaries, then we are ultimately going to be better users.
Third, the world of data can be quite exciting (yes, that sounds weird) because it is very potent and powerful and because we can look at stuff we’ve never really been able to see before. For this reason, data has been likened to the invention of the microscope. We can look at processes, trends, and movements which were previously only guesses.
Fourth, we must allow data to talk to other data. If a set of data is just out on an island and it cannot help us interpret other sets, it’s usefulness is suspect. At EMSI we always try to crosswalk datasets to each other because it makes for much better analysis. Our goal is the same with job postings.
We’ve been working on a lot of new stuff lately. Some of it you’re going to see in the new Analyst, and other elements are more behind the scenes but just as important. Job Postings Analytics (or JPA, because at EMSI we’re hopelessly addicted to acronyms), Hires, Separations, and Job Churn.
And just to set the stage, JPA and Hires are making their way into Analyst in the next month. Separations and Churn are necessary building blocks for getting good analysis of job postings and determining hires, but they won’t be “in the tool” at our big release.
Let’s start off with the one on a lot of people’s minds: Job Postings. We’ve spent a lot of time digging through these numbers to find out where they’re strong and where they have limitations, and how we can bring them to our clients in the most useful way.
Here’s a big picture illustrating one of the issues we needed to deal with when approaching job postings. From our background in dealing with LMI, we knew that job postings aren’t representative of the entire labor market, but we weren’t sure how far off they might be. Here you can see the size of the average number of hirings in a month (the blue), the average number of monthly postings (after being de-duplicated, in green), and the average monthly job growth (you can’t see it yet….[CLICK]
There it is! So understanding and creating a good hires number gave us the context we needed to compare LMI to job postings. Now you might be wondering how these numbers add up, with 8 million hires and only 152 thousand jobs in growth.
That’s where separations comes in. If you had 1000 hires and 1000 separations, there’d be no growth, and 100% churn.
Ok, with that image of hires, job postings, and growth in mind, let’s talk some more about Hires and Job Churn. [Read slide]
We recently posted an article about ways to measure job demand, and how we’ve used the BLS’s openings projections in the past. True hires is a much more accurate picture of what industries are hiring or churning.
To get hires we researched two main sources: the BLS’s Job Openings and Labor Turnover Survey (JOLTS), and the Census’s Quarterly Workforce Indicators (QWI)
Initial findings have shown that JOLTS provides timely estimates, but ultimately falls far short of actual hirings and separations that take place in the economy. We know that by comparing to QWI, which records hirings and separations by connecting SS numbers to mandatory employment records.
QWI takes some time to come out since it’s so thorough, but is the closest thing to a complete headcount of job movement we’ve found
EMSI uses QWI to inform our final hires number, but there are still holes that we fill in
Self-employed
Suppressions
If you’re interested in more about how we compared these two, check out our in-depth article on the EMSI blog, “Measuring Demand: A Comparison of JOLTS and QWI”
Churn looks at both hirings and separations to give us a better picture of ALL movement in the economy.
With all the JP’s out there, we needed a better way to tie them back to reality. Regular, annualized LMI data doesn’t do that very well. But a true “hires” number is a much better benchmark to compare JP’s to.
So that’s the background on WHY we’ve gotten into all these new elements, and how they help us understand job postings. Here are the treasures we found within job postings, these are the questions we think job postings can answer:
You’ll notice that these are all elements typically absent from traditional LMI. This is one example of what we mean when we’ve said that JPA can supplement LMI but not replace it.
Posting intensity needs a little unpacking. We find the number of unique postings for an occupation and compare to the total number of postings for that occ. This gives us an idea of how intense the recruiting process is for those jobs in a given region. [Go through slide] This would show up as an occupation or job posting with high posting intensity.
We don’t want to be TOO down on something we’ve spent a lot of time on and are making a big deal about adding to Analyst, but it’s always a good idea to keep these limitations of job postings in mind.
Some postings appear because a company wants to hire that person now. Others because the company wants to hire in 6 months. Some postings appear for companies to build up a list of possible recruits to pull from sometime in the future.
In the spirit of Matt Yglesias’s quote at the beginning, we’re not shying away from something that’s hard. We’ve spent years getting the world’s best LMI in place that folks have been using for years to make strategic planning decisions. The next challenge is how to make that data more current and relevant. That’s where job postings in the context of traditional LMI comes in.
Here’s a little background on where they come from:
Something that’s tricky for us here is that we’re so used to talking about all the sources we pull from to create EMSI’s core data. In the case of LMI, more sources are nearly always better than less. We combine sets that describe different parts of the economy, like industries, occupations, and demographics, and use other sets that give more detail or better timeliness. All said and done, our data is better with the 90 sources we have than if we only used 50 of them.
But postings don’t necessarily work the same way. We could claim 15 million postings a month if we were less rigorous with de-duplication, but that wouldn’t mean that the data was better. If we added 30 more “sources” that are just picking up the same postings we already have, that isn’t clearly improving the value either. So more isn’t always better in this world.
Our de-duplication process is done in two steps, with the first round looking for postings that are identical in city, ONET, and company. The second sweep looks more closely at these same elements to find other cases of duplication.
Ok, that’s the philosophy behind and origin of the new stuff that we’re up to. Let’s look at some of it in action.
The big picture: There are roughly 465,000 app developers in the top 50 MSAs in the country. (counting employees and self-employed workers), a 14% increase since 2010 and a 24% increase since 2004. Nearly three-quarters of all app developers work in the 50 largest metros—and nearly a quarter can be found in four metros: New York, Seattle, D.C., and San Jose.
Those are some of the numbers we know by tapping into EMSI’s structural labor market data. With job posting analytics…
…we can also see that companies are very active in advertising jobs for app developers. The 50 largest metros had over 30,000 unique postings per month in the first part of 2014, actually down 14% from last year. And hires, while not keeping up with postings, have been robust, too: around 20,000 per month.
Now, here is what is tricky about job posting data. We don’t know if 20 of these software developers got hired off of one posting or if a company made 30 postings and only hired five people. We also don’t know how many companies hired software developers without actually making a posting. The relationship here isn’t 1:1.
In addition, if we looked solely at postings, we’d see there’s a lot of activity in the nation’s biggest job markets. But we’d also see declining job posting activity from last year—which goes against the growth trend for app developers and almost every other computer occupation. Without the LMI context we wouldn’t get a sense for the sheer number of developers in the workforce today and how they’ve expanded since the recession.
Let’s look at a different occupation, one that’s also growing with a good number of jobs. [Read slide] When we bring in the postings data…
We see a very different picture in the graph. Compared to app developers, welders are far less represented in job postings, but still show steady hiring.
This means that companies in the 50 most populous metros are hiring seven times more welders than they’re posting for. Which makes intuitive sense if you consider that manufacturing and construction firms looking for welders aren’t as likely to post job listings online as, say, a tech company in San Jose looking for an app developer.
Without the context that Hires provides, it would be really difficult to tell if 1,000 postings per month was close to actual demand, or way off. We wouldn’t know by looking at just postings.
EMSI’s goal is to give our customers the best, most up-to-date and comprehensive information possible. This includes the most robust structural labor market data available, as well as our new job posting analytics data (see our announcement) that’s paired with job churn numbers.