Thinking Like a Data Scientist Part 3: The Role of Scores

In New Zealand, they are taking a "Moneyball" approach to optimizing social worker spending and focus attention can be most effective. A recent article in BusinessWeek "A Moneyball Approach To Helping Troubled Kids" (May 11, 2015) highlights the role that "scores" can play in identifying and prioritizing problem areas, and deciding what corrective actions to take.

Using data from welfare, education, employment, and the housing agencies and the courts, the government identified the most expensive welfare beneficiaries - kids who have at least one close adult relative who's previously been reported to child safety authorities, been to prison, and spent substantial time on welfare. "There are million-dollar [cost] kids in those families," Minister of Finance Bill English says. "By the time they are 10, their likelihood of incarceration is 70 percent. You've got to do something about that."

...one idea is to rate families, giving them a number [score] that could be used to identify who's most at risk in the same way that lenders rely on credit scores to determine creditworthiness. "The way we may use it, it's going to be like it's a FICO score," says Jennie Feria, Head of Los Angeles' Department of Children and Family Service. The information, she says, could be used both to prioritize cases and to figure out who needs extra services.

In continuing my "Thinking Like a Data Scientist" blog series, we're going to focus on how "scores" can play a critical role in supporting an organization's key business decisions. The power of a score is that it is relatively easy to understand from a business user perspective, and it focuses the data science efforts on identifying and exploring new variables, metrics and relationships that might be better predictors of performance.

Definition of a Score
Let's start by understanding what a score is:

• A score is a dynamic rating or grade standardized to aid in comparisons, performance tracking and decision-making; scores can help to predict likelihood of certain actions or outcomes
• Scores are actionable, analytic-based measures that support the decisions your organization is trying to make, and guide the outcomes the organization is trying to predict

A common example of a score is the intelligence quotient or IQ score. An IQ score is derived from several standardized tests in order to create a single number that assesses an individual's "intelligence." The IQ score is standardized at 100 with a standard deviation of 15, which means that 68% of the population is within one standard deviation of the 100 standard (between 85 to 115). This standardization makes the IQ score easier to compare different candidates or applicants, and support key business decisions.

The true beauty of a "score" is its ability to convert a wide range of variables and metrics, all weighted, valued and correlated differently depending upon what's being measured, into a single number that can be used to guide decision-making. And the true power of the "score" is the ability to start small with some simple analytics, and then constantly fine-tune and expand the score with new metrics, variables and the relationships that might yield better predictors of performance.

FICO Score Example
FICO may be the best example of a business score that is used to predict certain behaviors, in this case, the likelihood of a borrower to repay a loan. Fair, Isaac, and Company first introduced the FICO score in 1989. The FICO model uses a wide range of consumer data to create and update these scores.

A person's FICO score can range between 300 and 850. A FICO score above 650 indicates that the individual has a very good credit history while people with scores below 620 will often find it substantially more difficult to obtain financing at a favorable rate (see Figure 1).

Figure 1: http://tightwadtravelers.com/check-fico-credit-score-free/

The FICO score considers a wide range of consumer data to generate the single score for every individual. The data elements that are used in the calculation of an individual's FICO score include[1]:

Payment History: 35 percent of the FICO credit score is based on a borrower's payment history, making the repayment of past debt the most important factor in calculating credit scores. According to FICO, past long-term behavior is used to forecast future long-term behavior. This is a measure of how do you handle

• credit; think credit "behavioral analytics." This particular category encompasses the following metrics and variables:

• Payment information on various types of accounts, including credit cards, retail accounts, installment loans and mortgages
• The appearance of any adverse public records, such as bankruptcies, judgments, suits and liens, as well as collection items and delinquencies
• Length of time for any delinquent payments
• Amount of money still owed on delinquent accounts or collection items
• Length of time since any delinquencies, adverse public records or collection items
• Number of past-due items listed on a credit report
• Number of accounts being paid as agreed

Credit Utilization: 30 percent of the FICO credit score is based on a borrower's credit utilization; that is, the percentage of available credit that has been borrowed by that individual. The Credit Utilization calculation is comprised of six variables:

• The amount of debt still owed to lenders
• The number of accounts with debt outstanding
• The amount of debt owed on individual accounts
• The types of loan
• The percentage of credit lines in use on revolving accounts, like credit cards
• The percentage of debt still owed on installment loans, like mortgages

Length of credit history: 15 percent of the FICO credit score is based on the length of time each account has been open and the length of time since the account's most recent activity. FICO breaks down "length of credit history" into three variables:

• Length of time the accounts have been open
• Length of time specific account types have been open
• Length of time since those accounts were used

New credit applications: 10 percent of the FICO credit score is based upon borrowers' new credit applications. Within the new credit application category, FICO considers the following variables:

• Number of accounts have been opened in the past six to 12 months, as well as the proportion of accounts that are new, by account type
• Number of recent credit inquiries
• Length of time since the opening of any new accounts, by account type

• Length of time since any credit inquiries
• The re-appearance on a credit report of positive credit information for an account that had earlier payment problems

Credit Mix: 10 percent of the FICO credit score is based upon repaying the variety of debt, which is a measure of the borrower's ability to handle a wide range of credit including:

• Installment loans, including auto loans, student loans and furniture purchases
• Mortgage loans
• Bank credit cards
• Retail credit cards
• Gas station credit cards
• Unpaid loans taken on by collection agencies or debt buyers
• Rental data

The point of showing all of this FICO calculation detail is to reinforce the basic concept (and power) of a score - that a score can take into consideration a wide range of variables, metrics and relationships to create a single, standardized number that be used to support an organization's key decisions, or in the case of the FICO score, used by lenders to predict a particular loan applicant's ability to repay a loan. That's a very powerful concept. Scores are a critical concept in getting your business stakeholders to contemplate how they might want to integrate different variables and measures to create scores for the key business decisions that they need to make.

Other Industry Score Examples
Scores can be created to support business stakeholder decision-making across a number of different industries. Let's brainstorm just a few, and as my MBA students are going to find out this fall, there are many, many more waiting to be discovered!!

Financial Services

• Retirement Readiness Score. This would be a score that measures how ready each client or investor is for retirement. This score could include variables such as age, current annual income, current annual expenses, net worth, value of primary home, value of secondary homes, desired retirement age, desired retirement location (Iowa is a lot cheaper than Palo Alto!!), number of dependent children, number of dependent parents, desired retirement lifestyle, etc.
• Job Security Score. This score would measure the security of each individual's job. This score could include variables such as industry, job type, employer(s), job level/title, job experience, age, education level, skill sets, industry publications and presentations, Klout scores, etc.
• Home Value Stability Score. This score would measure the stability of the value of a particular house. This score could consider variables such as current value, turnover and house sales history, value of house compared to comparable houses, whether it's a primary residence or rental residence, local price-to-rent ratio, local housing trends (maybe pulled from Zillow), etc.

[1] FICO's 5 factors: The components of a FICO credit score (http://www.creditcards.com/credit-card-news/help/5-parts-components-fico...)

Additional Scores for different industries can be seen in Table 3 below.

Table 3: Potential Scores by Industry

Summary
Scores are a very important and actionable concept for business stakeholders who are trying to envision where and how data science can improve their decision-making in support of their key business initiatives. As we saw from the FICO example, scores aid in performance tracking and decision-making by predicting likelihood of certain actions or outcomes (e.g., likelihood to repay a loan, in the case of the FICO score).

The beauty of a "score" is its ability to integrate a wide range of variables and metrics into a single number, and the power of the "score" is the ability to start small and then constantly looking for new metrics and variables that might yield better predictors of performance.

Simple but powerful, exactly what big data and data science should strive to be.

• Thinking Like a Data Scientist Part I: Understanding Where To Start
• Thinking Like a Data Scientist Part II: Predicting Business Performance