Is There a Deeper Meaning Behind the Words We Choose to Use?

By Jackie Lorch, Vice President, Global Knowledge Management

Three US economists recently studied how to predict the likelihood that someone would pay back a loan, based on the words they used in their application. They used data from the Prosper online lending site, where thirteen percent of borrowers default.

The researchers found that the specific words people use when applying for a loan are strong predictors of their future actions—even though everyone expressed intent to pay back the loan. This finding held true even after controlling for other factors like credit rating and income.

Researchers identified ten words and phrases people commonly use when applying for a loan. Five of them indicate a higher likelihood to repay, five indicate a lower likelihood:

  1. God
  2. promise
  3. debt-free
  4. minimum payments
  5. lower interest rate
  6. will pay
  7. graduate
  8. thank you
  9. after-tax
  10. hospital

Why do the researchers believe that these words or that certain words correlate with a higher or lower likelihood to repay a loan?

  • The words that are italicized suggest a higher level of understanding of financial matters and specific achievements, and the bold words suggest a lesser understanding.
  • The more vehemently someone promises to pay the money back, the less likely they are to do so.
  • Similarly, someone who explains why they can pay the money back is less likely to pay it back. 
  • The researchers didn’t say why they thought someone who expresses “thanks” is less likely to pay; perhaps this indicates at some level the person views the loan as a favor, even as a gift, rather than a practical business transaction.
  • The mention of “God” was one of the highest indicators of likelihood not to pay.

The researchers raised ethical questions about this. Will companies soon analyze people’s social media words when deciding whether to offer them a loan? “Do corporations have the right to judge our fitness for their services based on abstract but statistically predictive criteria not directly related to those services?,” they ask.

It may already be too late to stop the practice. Algorithms currently in use are becoming ever more sophisticated. How much more will we soon learn about people from the words they use in open end research question responses?

It’s right to consider the ethical questions posed here, because, depending on how we use these techniques, we in research could risk alienating respondents, making them less likely to give us the candid responses we want — and it may even cause them to turn their backs on research altogether.