Linguistic analysis of empathy in medical school admission essays

Mary Yaden, David B. Yaden, Anneke Buffone, Johannes C. Eichstaedt, Patrick Crutchley, Laura Smith, Jonathan L. Cass, Clara A. Callahan, Susan Rosenthal, Lyle Ungar, Andrew Schwartz, Mohammadreza HojatInternational Journal of Medical Education 2020 · DOI

Abstract

This study demonstrates that physician empathy can be predicted from medical school admission essays. The implications of this methodological capability, i.e. to quantitatively associate linguistic features or words with psychometric outcomes, bears on the future of medical education research and admissions. In particular, these findings suggest that those responsible for medical school admissions could identify more empathetic applicants based on the language of their application essays.

Introduction

Empathy is far more than a "bedside manner" luxury; it is a fundamental clinical necessity. Previous research has established that empathetic physicians achieve superior patient outcomes, such as better glycemic control in diabetic patients and higher rates of treatment compliance. Despite its importance, identifying empathetic candidates during the medical school admissions process remains a significant challenge. Standardized tests often fail to capture this trait, and self-reported empathy scales are frequently skewed by "social desirability bias," where applicants provide answers they believe recruiters want to hear.

This research explores an unobtrusive, data-driven approach: analyzing the natural language in admission essays to bypass these biases and identify truly patient-centered candidates.

Methodology

The researchers utilized Differential Language Analysis (DLA), a computational technique designed to link linguistic patterns to specific psychometric outcomes. The study analyzed personal statements from 1,805 matriculants at Sidney Kimmel Medical College between 2008 and 2015.

The workflow consisted of three primary stages:

  1. Feature Extraction: Using the Differential Language Analysis ToolKit (DLATK), essays were tokenized into individual words.
  2. Topic Modeling: To capture thematic context beyond simple word counts, the team employed Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF). These algorithms grouped frequently co-occurring words into 20 distinct "topics."
  3. Correlation Analysis: Multivariate linear regression was used to determine which topics correlated with scores from the Jefferson Scale of Empathy (JSE), a validated 20-item instrument. The analysis controlled for gender, as women typically score higher on empathy scales.

Results

The analysis revealed a statistically significant correlation between specific linguistic patterns and JSE scores ($r = .074, p < .05$). While the effect size is small, it provides a crucial "proof of concept" for automated linguistic screening.

Word cloud representing linguistic features positively correlated with empathy scores.

High-empathy scores were consistently associated with a specific cluster of words centered on the patient experience and human connection. Key terms included compassion, empathy, trust, understanding, human, and care.

Discussion: The "Patient-First" Signal

The most significant finding was the emergence of a "Patient-First" signal. The words predicting high empathy were not related to "science," "technology," or "prestige," but rather to the subject of care—the patient. This suggests that empathetic candidates are linguistically focused on the human element of medicine rather than the technical mechanics of the profession.

Furthermore, the linguistic topics identified (e.g., "understanding," "perspective," and "trust") align with the theoretical definition of physician empathy as a cognitive attribute—the ability to understand a patient's experience—rather than a purely affective "feeling." This indicates that even in highly curated and performative admission essays, subtle linguistic signals regarding a candidate's psychological profile remain detectable.

Conclusion and Future Directions

While the correlation found is modest, the authors attribute this to a "ceiling effect." Since the dataset consisted entirely of accepted students, the variance in empathy scores and essay quality was likely restricted.

Future research should address the risk of "gaming" the system; if these linguistic markers become widely known, applicants might artificially insert "empathy words" into their essays. To counter this, more sophisticated models focusing on narrative structure and context are needed. Additionally, applying these techniques to "spontaneously generated" language, such as interview transcripts, may provide a clearer, less-curated signal of a candidate's true empathic orientation.

Ultimately, this study demonstrates that automated linguistic modeling could eventually allow admissions committees to screen thousands of applications for empathy more objectively and efficiently than human review alone.

Summary updated

Abstract

This study demonstrates that physician empathy can be predicted from medical school admission essays. The implications of this methodological capability, i.e. to quantitatively associate linguistic features or words with psychometric outcomes, bears on the future of medical education research and admissions. In particular, these findings suggest that those responsible for medical school admissions could identify more empathetic applicants based on the language of their application essays.
BibTeX
@article{yaden2020linguistic,
  author = {Mary Yaden and David B. Yaden and Anneke Buffone and Johannes C. Eichstaedt and Patrick Crutchley and Laura Smith and Jonathan L. Cass and Clara A. Callahan and Susan Rosenthal and Lyle Ungar and Andrew Schwartz and Mohammadreza Hojat},
  title = {{Linguistic analysis of empathy in medical school admission essays}},
  journal = {International Journal of Medical Education},
  year = {2020},
  doi = {10.5116/ijme.5f2d.0359},
}
Figure