Analytics used to predict who will leave a job!

22 03 2013

HP ran a pilot program to use analytics to predict if employees might quit. Employers traditionally use exit interviews to figure out why and when employees leave for improvement (hopefully). Quantifying the sources and finding a companies risk factors before losing talent would benefit any business.


In my own research I find that most voluntarily exits by employees occur in July though September. See the chart for an example of data from the Bureau of Labor Statistics. Look for the highlighted spike in late summer. December on the other hand is one of the lowest periods of the year for turnover. Therefore January through May would be a great time to start analyzing the historic data and implementing changes to beat off those late Summer exits, or for whenever your high turnover period was. It is predictive analytics you can do without special software or a PHD in data science or statistics.

Thanks to the Operations and Supply Chain Management blog of F. Robert (Bob) Jacobs, Professor of Operations Management at the Kelley School of Business, Indiana University for the original story.

Operations and Supply Chain Management

WSJ – March 14, 2013, 3:36 PM ET

Book: HP Piloted Program to Predict Which Workers Would Quit

Hewlett Packard Co. tested a predictive scoring system that attempted to grade the likelihood that individual workers would quit the company, according to a new book.

HP piloted the scoring system in 2011 aimed at lowering attrition through a better understanding of which workers were most likely to leave, according to Predictive Analytics: The Power To Predict Who Will Click, Buy, Lie Or Die by Eric Siegel. The analytics model, Mr. Siegel says, looked at factors such as salaries, promotions and job rotations, and scored the likelihood that particular employees would leave. HP data scientists believed a companywide implementation of the system could deliver $300 million in potential savings “related to attrition replacement and productivity,” according to a November 2011 company presentation.

Data scientists made the presentation at a Predictive Analytics World…

View original post 215 more words

Create Your Own Word Clouds

5 03 2013

I decided to create word clouds based on my Linked-In profile, CV page and recent blog posts. The results are below.  Anybody can make word clouds like these, just go to Wordle, enter the text or site and start creating.

My Linked-In Profile


My Curriculum Vitae

Woofle for CV

Recent blog posts


Hat tip to the Oregon Office of Economic Analysis for the idea.

Do Job Searches and Applications Decline During the Holiday Season? (Part 3 of 3)

1 03 2013

The final post of this series documents my conclusion and analysis of the available data described in Part 1 and Part 2. As reported in my earlier post, some recruiters and organizations claim a decrease in applications per opening in December. This series examines and analyzes the available data behind this assumption.

To analyze these trends I created a database containing monthly data from January 2004 until December 2012. This series contained 108 time observations of 15 variables. This post is a little delayed as I had to wait until a very recent data release from the Department of Labor Bureau of Labor Statistics Job Openings and Labor Turnover Survey (JOLTS) and from other DOL data series.  Oregon specific data comes from the Oregon Labor Market Information System (OLMIS). The central variable I used as the dependent variable in my models is a data set I created from weekly Google Search Index of web searches in the United States for jobs under the jobs and education sub group. To be more specific this was a quantified version of weekly searches over time provided by Google that I averaged to get monthly data series. I used a specific search category to avoid cross chatter from search for topics like the late Steve Jobs of Apple. Most of the searches involved Indeed, Craigslist, Google  and Yahoo (both job posting sites and search tool sites).

I created several econometric models to investigate the effect of the holiday season on the labor market and job applications. I used the Gretl program, a cross-platform software package for econometric analysis, written in the C programming language. I used an OLS regression analysis to check correlation and connections between the data points. These correlations could then be analyzed to be good indicators of holiday hiring and job applications. All data is not seasonably adjusted to better look at the actual holiday months, rather than massaged data to account for cycles. In this study I examined the cycles themselves, thereby necessitating unadjusted data.

My first regression used the Google Search data as the dependent variable. In other words I wanted to see how it correlated with the other data. For the independent variables I used new jobs added, number of hires, the unemployment rate, job openings, layoffs and discharges, other separations and the number of voluntary exits or quits(all monthly for the United States). The data for job openings, layoffs and quits was not connected to the Google search index at any significant level. I also ran the results with Oregon specific statistics and got similar results.  In other words, the number of layoffs, job openings and quits does not seems to directly correlate with Google job searches. The variables for the unemployment rate (both locally and national) did however have a strong correlation and coefficient with the Google search index. We can interpret this simply that historically, when the unemployment rate goes up 1%, the Google search index also goes up  about 6%. The empirical observation is that a higher employment rate increase the number of job seekers thereby increasing the number of job searches online. The Search data also had a statistically significant correlation with “other separations”  and new jobs and hires. The full regression results are below for thew more wonkish audience. If you are not interested, you can skip ahead.


Some stray observations:

  • Some data is skewed by the Great Recession in the middle of the time series. See the chart below for an example.
  • Very strong correlation between the unemployment rate and the jobs search index. The coefficient is 6.2 with a very small p-value. If the model is correct, then the coefficient implies that a 1% increase in the unemployment rate would generally correlate with a 6.2% increase in web searches for jobs.  The small p-value means that it is unlikely that the data would look like this randomly.
  • I added a late variable of unemployed per opening. This is the number of job openings divided by the total unemployed population. It shows exactly how competitive the job market is.
  • Really interesting that the unemployed per hire has an inverse relationship with the Google job searches. The only way I can explain that is that as a failure of the model or that the Great Recession affected the searches in unexpected ways (see point 4 below for more details). Another explanation is that there are far fewer new jobs in December, making the ratio of new jobs to unemployed seem normal or higher in December.
  • Between 2004 and 2012, on average there are 1 million fewer unemployed people during December, compared to the average for the rest of the year. Therefore we can convulsively say there is a smaller labor pool during the holiday season.  The number of unemployed per opening in December is almost the same as the number of unemployed per job opening for the entire year (3.2).
  • Search Index on average declines 9% in December.
  • Job openings and new jobs significantly down due to hiring in ramp up to the holiday season (e.g. October and November).
  • The Unemployment rate is typically down about .1% on average in December.
  • Quits were always way down in December, about 450,000 less on average. Another strong indicator that the available labor force shrinks during December, particularly passive candidates, see #5 below.
  • For a graphical example of job creation and the following decrease in job searches, see the chart below. The outliers to the left are time series during the worst of the 2008-2009 recession. During normal economic conditions, increased numbers of jobs correlates with a decrease in the job search index.


We conclude that businesses do indeed receive fewer applications in December for a variety of reasons.

  1. Fewer job openings occur during the month of December. This is because Holiday hiring occurs in the months going into the season, not during the busy peak period.
  2. There are fewer unemployed people in December, therefore there are fewer applicants in the active labor supply pool.
  3. Currently employed individuals who look for new jobs typically slow or stop their search during the holiday season. However the volume of job searches is more closely related to the unemployment rate than the number of employees leaving their jobs.
  4. Google searches for jobs tends to follow conventional economic wisdom. Meaning that job searches increase during periods of high unemployment, however the month to month change in December is counter intuitive. The smallest decline in job searches occurred during December 2007-2010, the only way to explain that is to accept that passive candidates stuck with their jobs and avoided searching due to the terrible economic conditions.  On the bright side here, in 2012 we saw the lowest decline in holiday job searches since 2004, meaning that people feel comfortable enough to search for jobs they may have held during the recession.
  5. The unemployment rate is down and passive job seekers do not search for jobs as often in December. Therefore the labor market supply and total applications decrease during the holiday season. The data  backs up the empirical and anecdotal evidence.

The fundamental takeaway is twofold. For job applicants, December is a good time to apply for jobs due to the fact that much your competition may have taken a break for the job search. More importantly, the most qualified and experienced candidates (i.e. your strongest competition) seem to search for jobs less in December. For employers, you are likely to get fewer applicants in general and of the highly qualified sort in December. You might want to increase your December recruitment advertising budget or extend your search into January for a larger labor pool.

Do Job Searches and Applications Decline During the Holiday Season? (Part 2 of 3)

22 01 2013

As discussed in my earlier post, some recruiters and organizations claim a decrease in applications per opening during December. This series examines and analyses the data behind this assumption.

Most recruiters and hiring managers that I surveyed report a decreased amount of general applications and qualified applications during the holiday season. Of course some dissenters think this is an myth. This post follows-up and elaborates on the available data. The third and final post will be an analysis of the data and its implications. At first glance, the data seems to indicate that job web searches certainly do decline during the holidays. Searches for jobs over the past eight years for example declined by nine percent on average during December. In the charts and graphs below, you can clearly see that the biggest declines of the year nearly all come in November and December. Both domestic and international searches for careers and jobs had declines during the holidays.


1-16-2013 5-48-03 PM

1-16-2013 5-45-11 PM

My first source of interesting data is the Google Trends search index data. I review and analyzed data from searches for careers and jobs between 2004 and 2012. The graph above has the holiday season highlighted. The implication is that there is a very clear decline in job searches during the holiday season. However we also see a huge spike in searches almost like clockwork on January 1st. Most likely the post-holiday layoffs increased the number of job seekers or they are joined by a large chunk of passive job seekers who took a break. One interesting point to mention was that the biggest job search declines in December were during the months of 2007, 2008 and 2009. The only way I could reconcile that fact with economic data is that passive job searchers either gave up or “hunkered down” during the great recession during that time period. With dramatic job losses dramatically reducing job searching, the currently employed would not likely be out searching for a new job. If accurate, this could be a clear indicator that highly qualified applications do indeed decline during the holiday season.

1-17-2013 12-58-14 PM

To figure out the labor demand, you have to look at whether there are more or less jobs or jobs added in December than the rest of the year or look at unemployment. In Oregon for example, far more jobs are created during December than other months. This is one of the reasons economists seasonally adjust data. Since 1990 in Oregon for example, December had on average nearly 23,000 more jobs than the average of other months. Holiday related hiring is an obvious explanation for this. Returning to the matter at hand, this data does indicate that there is higher demand for workers in December on average. The unemployment rate is generally down in that month and job searches sharply decline. So far the data does seem to corroborate the idea that business do indeed receive fewer applications per opening in December due to a decreased supply of applicants and an increased demand from employers.

1-16-2013 6-03-04 PM

My final post on this series, will involve a regression analysis comparing the Google Search Index data with assorted economic data for a possible correlation and conclusion.

Do Job Searches and Applications Decline During the Holiday Season? (Part 1 of 3)

12 12 2012

Some recruiters and organizations report a decrease in applications per opening during December. While hiring generally increases during the Holiday season, it is believed that it can be harder to find qualified applicants. Hiring does indeed increase during December from holiday travel, shopping and shipping related purchasing. I wonder if job searching by applicants also declines during this time. I assume that employers on average receive fewer job applications per position during the holiday season due to either the decreased supply of available workers or lower demand for employment (workers not actively searching and applying as much as normal) or a combination of the two.

The anecdotal and empirical information on this can be conflicting. On one hand hiring does seem to increase  during the Holiday season.  However in my experience many businesses receive fewer applications on average per position during the Holiday season.  I assume that job applicants both active, and passive spend less time applying during the Holiday season and thereby apply to fewer positions total than during other parts of the year. For this project I surveyed a large network of recruiters, hiring managers and job applicants to compare empirical and anecdotal evidence. My findings confirmed most of what I’d already thought about hiring in December. However some of the feedback was very industry specific. A Human Resources Information Systems (HRIS) Analyst for an Oregon university reported a reduced number of applications for positions in December, with even fewer of them being qualified or not returning calls for interviews. Job application support service providers say that it is also harder to find open positions this time of year with hiring teams out of the office or not posting new positions. Another recruiter indicated that they receive nearly half the number of applications during December for the same positions posted during the summer. Most importunity it was mentioned that highly qualified applicants often do not seek new positions in December. While some hiring managers think these theories of reduced applications are a myth.

There is potential for confusion here between the ideas that there is a smaller pool of candidates in December because of increased hiring or because of decreased interest in  new jobs. “Do people with current jobs search less during the holiday season, and do unemployed do the same?” is the fundamental question I will investigate  in the next two parts of this series.

Part 2 of this series will cover the data is used to examine this idea. I plan oi examining trends from internet search data and government employment statistics. Part 3 will be my final analysis and evaluation.

Introduction to Human Resource Analytics

12 11 2012

Human Resource analytics do not require an advanced degree in mathematics, economics or statistics. Any operations, HR or business management professional can become an expert in their client or companies people based metrics. Just tracking and reviewing some of these metrics is often all the evidence you need to prove or quantify what you already knew about operations, recruitment, benefits and organizational development.

Lets take a look at a simple broad metric, Revenue Per Employee (“RPE”).  So compare two similar companies with similar workforces but one has double the revenue per employee. It paints a very different picture of productivity, profitability and compensation at these companies. Based on economic theory and my understanding of scalability, software and manufacturing companies would lead the pack here when it comes to RPE. This is likely due to scalability and the nature of their industry and products. The RPE laggards on the list should be food and professional services (low-tech). The implication is that these low RPE companies and industries often need twice or three times as many employees to produce the same revenue as the higher RPE companies.

A later post will delve deeper into an industry comparison of data I gathered from fast growing Oregon businesses across several industries.

Quality of Hire Metrics

24 04 2012

In recruitment and business statistic circles, the term quality of hire is popular as of late. Many firms are discovering new methods to track and quantify this data.

One of the most important ways for a business to gather these statistics is by tracking the source of applicants. This white paper by Indeed should be required reading for recruiters and anybody who does recruitment advertising.  The article stresses the importance of quantifying your applicants based on source.  A quick comparison of average applications from Craigslist, Monster and Career Building paints a pretty clear picture of why the later two are in decline (provided by Craigslist though). Craigslist provides far more average page views for a fraction of the cost. Specialized job boards however are doing quite well in the recruitment advertizing market. To track your hiring metrics you need a feature in your applicant tracking system. If you still get resumes via email then you can tack them via a simple spreadsheet as well. Once a company has a good grasp of its recruitment statistics, the next step is too compare them with quality of hire metrics.

Quality of hire is a broad term but it can be any number that compares the hiring source, job expectations or qualifications with eventual results. For example Company Z hired 200 employees in 2011. Of those 100 employees only 50 stayed on through a full year. In that particular snap shot, the quality of hire would be 50% for the entire recruitment process. However that is only one simple method of evaluating quality of hire. Another method might compare quality of hire of external versus internally referred applicants.  A company could even compare the quality of hire between recruitment from different colleges, staffing vendors or job posting sites.  These statistics could be used to create mathematical models predicting success of applicants based on some specific qualifier. Smaller businesses should note that a small sample size is not enough to make assumptions and that correlation does not imply causation. At the very least though, quality of hire metrics can help evaluate better spending on recruitment advertising.

Additional reading:

That Pacific Northwest Life

Having all of the fun.

Business Analytics 3.0

Data Driven Business Models

Simple Tom

Some say I was born high. Others say i'm just simple :)


Knowledge and Happiness(K&H) multiples by dividing it. More you share, higher and bigger they grow.

Where's my backpack?

Romancing the planet; a love affair with travel.

Gen Y Girl

Twentysomething. Annoyed with corporate BS. Obsessed with Gen Y. Not bratty. Just opinionated.

An International company that offers private antique art sales to clients around the globe.


all about Human Resources

Freedom (to) management

How to have an impact, gracefully, in the organizations of the future