4. Human Science and Selection Decisions

Why this matters is because any hiring or promotion decision based on factors other than who is the best candidate for the job will ultimately lead to suboptimal performance outcomes. What can advanced analytics do to assist with eliminating biases? A lot. Biased decision making is discrimination, and using advanced analytics to assist in the decision-making process will help eliminate biases. What difference does eliminating bias make? Also, a lot. Getting the right person in the right job drastically improves the probability of success.

I, however, have a bias (although it is not actually a bias because that suggests it is unfounded) when it comes to selection decisions. Time spent making as robust a decision as possible is time very well spent. It is in everyone’s self-interest to be working in a meritocracy (in effect, where people are doing what they are best suited to be doing). In this type of organization, it is not the loudest, the prettiest, or the boss’s nephew who gets the job or the promotion. Instead, it is the person most likely to perform in the position. Any and all extraneous factors are eliminated from the selection decision, including the obvious demographic qualities (age, race, gender, disability, and so on). Also eliminated are factors like school attended, how well the interview went, how good someone’s golf game is, and various other factors that decision makers may have a bias about but that have no bearing on job performance.

This chain would be very wise to do everything in its power to retain and motivate this individual and, frankly, clone him. Much of what this chapter covers suggests how the use of tools from advanced analytics could allow this chain to do the next best thing to replicating him.

You may have wondered how people like Bill Gates, Steve Jobs, Larry Ellison, or Mark Zuckerberg (or David Karp, founder of Tumblr and who did not finish high school) could be so successful without finishing college. (Alternatively, you might wonder how so many who graduated from college are successful.) According to economic theory, one advantage to getting an education is that it sends a signal, a signal that someone is high ability. It signals that they can stick with something to its completion and have the ability necessary to see that it is accomplished. As we become better at identifying attributes associated with success, increasingly, these traditional signals have less meaning. Each of the company founders listed has great vision and ability, and many of the normal signals (educational certification, GPA) were just not necessary.

The fact is, if you have the ability and the skills and have acquired the necessary knowledge, the formal degree and all the traditional check lists are becoming less important. This does not mean that I am disparaging a college education. For me, college was transformational; new worlds (countries anyway) were opened up to me. The point is that the better we get at measuring ability, aptitude, and potential, proxies become less relevant.

In 2009, Gary Moore and Dan Selic established the nonPareil Institute.³ The charter of the institute is to provide those on the autism spectrum with training and technical job skills. This spectrum includes those at the high-functioning end who may have Asperger syndrome. This syndrome is characterized by difficulty with social interaction but also being very logical and very focused (traits important for computer programmers).

Some computer programmer will tell you that communication skills are very important when it comes to being a successful programmer. For some coding projects, this is certainly true. Others will tell you that it depends on the type of programming that you are doing. For example, if you are managing projects and coding, then social skills are critical. However, if your job mostly focuses on the actual coding associated with the project, and even if the project requires interaction between programmers, as long as the communication is focused on the task, those on the autism spectrum may perform exceptionally.

We all have our preconceived notions of what autism is, and so we might consciously or unconsciously exclude those with autism from consideration for a lot of things, including jobs. If we were to instead focus on the specific job, and the characteristics that makes for an excellent programmer in a specific setting, someone on the autistic spectrum could (and has been shown to) make a superb employee.

This is what advanced analytical tools enable us to do. We can use them to establish ever-closer approximation of fit between the task or job and those who function in those jobs. This is what advanced analytical tools enable us to do. We can use them to establish ever-closer approximations of fit between the task or job and those who function in those jobs.

According to an article in Fox Business, the CEO of Yahoo!, Marissa Mayer, insists on approving every hire.⁷ In this case, tools from human science could help Mayer substantially with her decision-making process. Absent the use of sophisticated analytics, there are potential downsides to Ms. Mayer’s approach. The obvious is the potential for slowing the hiring process down to a crawl. In addition, there is considerable potential for biases to be present in her selection decisions. The potential for bias is actually exasperated because she tends to prefer to hire those who went to top-tier schools and prefers computer science majors over electrical engineers.⁸ Tools such as those being developed by Gild and Evolv could assist Yahoo’s CEO (along with virtually everyone else making section decisions) to make accurate, unbiased, and timely decisions.

A multitude of online courses are available to anyone anywhere. It would not be difficult to develop a curriculum using free online course from places like MIT and Stanford for one of the hottest emerging jobs: Data Scientist. If you are sufficiently motivated you can become proficient in one of the most marketable jobs today. However, this is only part of the equation, the supply side. The other side, of course, is the demand side: organizations making it clear what specific skill set they are looking for. This is the role of workforce planning.

Workforce planning is a critical function for the success of the organization, but it is too often not done as well as it could be. Organizations may be too quick (or slow) to hire during an upturn and too quick to lay off upon a downturn. One of the biggest challenges I have run into time and again in my own professional life is having all the other resources readily available (such as financial and technical) but lacking the human capital that is ready and able to execute the plan. At this point in our very slow recovery, many organizations are seriously pondering the decision to hire or not to hire. They are loath to miss opportunities, but overstaffing is in no one’s interest. Advanced analytics can assist with making a more accurate workforce headcount projection.

It is possible to get an accurate read on the current situation of the firm and to determine the future needs of the organization. This requires a mechanism for effectively scanning the environment, in order to get an idea on the state of the economy, the workforce, and your organization’s challenges and opportunities.

Substantial research provides evidence that getting the right employee is associated with higher productivity, greater profitability, lower employee turnover, and generally much better organizational outcomes. Arguably, who you decide to bring in to your company is an important decision (perhaps the most important). If you bring in the wrong people, not much else of what you do matters. Granted, depending on the country in which you reside, you may have a fair degree of flexibility to sever the employment relationship. Wherever you are, though, hiring and training employees takes much time and energy. So, the better your recruitment and selection decisions, the better performance outcomes and cost savings.

Traditionally, the selection decision has been heavily weighted toward the interview. However, this has been shown to be a poor predictor of future job performance. A number of other instruments are considerably more useful when making selection decisions, and this section focuses on where these instruments and current and emerging analytical tools complement one another. For instance, one tool increasingly used today and a good fit for developing and emerging technologies is the use of biographical data (or Bio data) for selection decisions.

The recruitment and selection process is one rife with potential biased decisions making. Two researchers recently found that when pictures are included with a curriculum vitae, as they are often in Europe and Asia, they found that attractive men were more likely to be invited in for an interview. The inverse applied for attractive women. The less attractive, the more likely you are to be invited. Bradley Ruffle and Ze’ev Shtudiner determined that this was not because attractive women were considered less intelligent, but rather because recruiters (who are generally female) are attempting to limit the competition.⁹

Workforce planning, in particular, has a big impact on whether an organization has the bandwidth to respond to ever-changing business challenges and opportunities. Workforce planning, like all human capital management (HCM) decisions, has been viewed as part art and part science. I am clearly attempting to put as much science as possible into the decision-making process.

One of the more important and upfront decisions that has to be taken is how many employees do we need? Generally, organizations are notoriously bad at getting this number right; and it matters a lot. Having the right talent in the right place at the right time makes all the difference. Firms miss significant opportunities when they do not have the right employee in the right place at the right time.

Workforce planning consists of the following steps:

• Scanning the macro-environment

• Organizational strategic objectives

• Current workforce situation

• Projected workforce needs

• Analysis

• Action plan to fill gaps

Workforce planning may be much more helpful and necessary (and difficult) in substantially dynamic environments. If the organization is experiencing predictable growth, and business as usual is expected to continue with fairly static employee turnover, a fairly uncomplicated workforce planning system is needed.¹⁰ If the situation is a dynamic work environment in which there is uncertainty in the environment, there is a need for more sophisticated workforce planning methods.

Kahneman saw this as one of the more formative events in his professional life. It had a substantial impact on his view of forecasting, which he and Amos Tversky labeled the inside view and the outside view.¹² The inside view is the one we initially assume when evaluating a potential outcome. For instance, until he took the time to reflect, the dean thought it would take only up to 2.5 years to complete the book. It was not until after reflecting on the other curriculum planning groups that a more realistic view emerged. Forecasting based on previous data or information from similar tasks and related outcomes can be considered the outside view.

The COWI consulting group and the academic Bent Flyvbjerg have taken this process and applied it to a variety of processes that involve forecasting (mostly estimates associated with costing projects). The process is called reference class forecasting, and it applies mostly to transportation policy and planning.¹³ The process consists of the following steps:

(1) Identification of a relevant reference class of past, similar projects. The class must be broad enough to be statistically meaningful but narrow enough to be truly comparable with the specific project.

(2) Establishing a probability distribution for the selected reference class. This requires access to credible, empirical data for a sufficient number of projects within the reference class to make statistically meaningful conclusions.

(3) Comparing the specific project with the reference class distribution, to establish the most likely outcome for the specific project.¹⁴ This might sound complicated, but it can be applied to a number of different forecasting issues that arise within HCM. In essence, this entails looking back at previous workforce planning numbers and evaluating to determine how close they are to actual.

What’s more, experimentation can be undertaken. Seeing what happened during the last recession allows an organization to do a much better job of predicting the likely outcome associated with the next downturn, and how and when to start getting back in the game again. The same applies when things are going along well; this will provide you the information necessary so that you can know best when to start throttling back on hires.

It is true that when it comes to headcount and labor costs, the emphasis is on maximum flexibility. Two problems generally emerge when attempting to forecast future outcomes. One is optimism bias, and the second strategic misrepresentation. This means that the starting point for this question (optimal workforce planning) is getting the model right. What factors go in to predicting the kind of knowledge skills and abilities we will need to carry out the strategic objectives of the organization?

For examples of these and other tools, please go to: Decision AnalyticsInc.com.

Social analytics are already being used extensively within the marketing function, with much of this evolving around the use of sentiment analysis to examine and determine how consumers or potential consumers feel about a specific service or product. Organizations are also increasingly using social media to assist in recruitment efforts and to gauge the morale of the organization.

Other organizations are also providing very valuable data that can be utilized to make more accurate headcount and selection decisions. For instance, Glassdoor is a company that provides good background information into what is being said about other companies.¹⁵ Founded in 2007 by Richard Barton, Robert Hohman, and Tim Besse, Glassdoor provides information on job postings for more than 150,000 companies across 100 countries. They provide salary information, CEO ratings, and impressions of the work environment from current and former employees.

There are challenges associated with the use of social analytics. The issue of privacy is a serious issue. Data privacy laws are in place in Europe, and the issue can be contentious in the United States. It is becoming increasingly common for interviewers to ask for passwords to gain access to social media content.¹⁶ This has become pervasive enough that state legislatures are proposing bills to prevent employers from discriminating against employees who refuse to give access to their social media information.¹⁷

One of the more promising tools to assist with selection and promotions decisions is the use of biographical data (Bio data). This information is based largely on identifying specific characteristics that ultimately predict job success. Google has been using this technique to assist with hiring decisions for some years.¹⁸ There is evidence that Bio data surveys are better predictor of future performance than, say, the job interview.

You’ll recall my earlier example of the excellent check-out person I encountered. In the case of that grocery store chain, they would administer a survey to him that would identify characteristics and attributes, then administer a survey to prospective candidates evaluating for the same qualities. This can all be done online and very cost effectively.

The movie Moneyball has done much to bring analytics into popular consciousness. While there is no question that the use of analytics has helped to put predictive analytics on the radar for many, much of what is being accomplished in the professional sports arena is an attempt to do exactly what firms are spending a tremendous amount of time doing. They are attempting to predict who will be first-rate employees. The crux of the matter is determining which characteristics are actually associated with superior performance.

As those who saw Moneyball may remember, traditional analytics did not predict success. (In the case of the Oakland A’s, that would be winning games.) This is an important lesson that can be learned from the use of analytics: The factors that are actually associated with predicting success are usually much broader than traditionally thought.

The same hard work to establish what is actually associated with success in your particular organization is just as critical. For example, you often hear the phrase “we only hire the best people.” It should read like this instead: We only hire the best person for our specific organization and situation. An example is that many organizations now recognize the value of hiring people who are also willing and able to work collaboratively.

Many of the advances associated with increased accuracy of recruitment and selection are due to more accurate models of what ultimately impacts enterprise success. Therefore, it is critical that firms are clear on what is associated with success in their specific organization and follow this up with an appropriate recruitment, selection, incentive, and performance and talent management programs.

We have a lot to learn from professional sports teams. Keep in mind a couple of critical factors here. Most sports, except golf and singles tennis, epitomize a team production function. Of course, there can be a star center or a great shortstop or quarterback who makes a big difference; however, it is impossible to win alone. The same obviously applies when attempting to achieve organizational goals.

The selection decision is one of the most critical aspects of the employment decision and also is subject to considerable downside risk. Traditional hiring practices come with a number of potential downsides. There is the halo effect and the similar-to-me bias, neither of which can influence an algorithm. To see the impact of technological advances in HCM decision making, you just have a look at Google patents.²¹ Doing an open search on “employee selection” returns 17,300 sites. The use of technology to assist with many of these appears to be focused on the use of technology to enhance selection decisions. Many companies are starting to use advanced analytics for the selection and recruitment decisions. This is one function where there is quite a lot of scope for the utilization of these technologies to make better decisions.

One of the worst predictors of employment success is the job interview, and job testing is not much better. You might be able to nail the GRE, but you might lack the proverbial “fire in the belly.” Aaron Rogers, the quarterback of the Green Bay Packers (Super Bowl winners in 2010), started off in a community college, and he could not get a scholarship. He played and he played well; he had something to prove, and he constantly reminds himself of those days.

The use of bio data is a straightforward: Determine successful characteristics of current job holders and determine the relationship of these characteristics to job performance. Prospective job candidates are then screened for these characteristics.

The use of the bio data it allows for the inclusion of candidates who may have been screened out using traditional measures.²⁷ An algorithm allows for an evaluation of a much broader set of characteristics, and they also avoid the potential downsides human decision makers often exhibit.

In the first instance, start by identifying those behaviors, characteristics, and activities that lead to success on the job. Ultimately, this also needs to be mapped to how job success leads to organizational success. At this point, there is also a role for Kahnman’s “thinking fast.” Of course, I say this with my standard warning against all the biases associated with a decision like this. However, a manager or executive with long experience and expertise may well have a good intuition about someone.

Advanced analytics in the form of predictive modeling is ideally suited for the use of bio data. Analytics can go considerably beyond the straightforward who-to-hire question. We can use these techniques to help reduce or eliminate discrimination and even wage inequality.

These are risks with these two issues and another example of where seasoned human expertise plays a critical role. If the recommended candidates are trending toward too little diversity, adjustments can be made. In addition, the tool should be carefully validated so that it accurately reflects actual job tasks and also attributes that accurately predict success at those tasks.

As discussed, any kind of bias is suboptimal when making selection decisions, and the stronger the connection between attributes or characteristics and performance outcomes, the greater the validity of the selection instrument. A well designed bio data instrument should provide a strong connection between attributes and performance, eliminating all other factors from consideration. Effort should go into verifying the measures of these characteristics and attributes and their relationship to performance. Purely technically, bio data should provide a direct data-based connection. However, selection decisions will ultimately be made by someone or ones. Those ultimate decision makers need to be aware of their potential for bias and guard against it. If in doubt, like flight instructors tell pilots working toward obtaining an Instrument Rating (flying by instruments alone), trust what the instruments are telling you. If you have put the time in to make a robust validated instrument, the same applies here.

• Multiple Regression Techniques: Multiple regression techniques can be used to assess impact. It can be used to determine the impact on performance associated with the introduction of a new policy or program.

• Decision Trees: Essentially a graph or model depicting steps to a decision. This can be used to provide evidence-based recommendations on the type of developmental experience an potential executive should obtain prior to promotion.

• Monte Carlo Simulation: This consists of using computation algorithms to arrive at probability distributions, allowing determination of the likelihood that a particular intervention (such as putting in a child care facility) will have on employee turnover.