Covid Socio-Economic Inquiry
Unmet Mental Health Care During COVID-19
This exploratory analysis tracks changes in unmet demand for counseling or therapy across U.S. states from August 19, 2020, to June 23, 2021. It also tests whether state poverty, inequality, and urbanization measures help explain the differences.
Source Code
The source code is on GitHub.
Mental Health Inquiry
The analysis focuses on the period after the pandemic's initial shock and compares each state's beginning and ending observations.
Preliminary Notes
The mental health data comes from Data.gov. State-level Gini coefficients, poverty rates, and urbanization measures come from the linked reference tables below. The script downloads and cleans the data, caches the result in HDF5 format under root/data/, and generates the charts used here.
Summary
The outcome variable is "Needed Counseling or Therapy But Did Not Get It, Last 4 Weeks," renamed unmet_mental_health in the analysis. The comparison runs from August 19, 2020, through June 23, 2021.
Figures 1 and 2 compare the ending observation with the confidence interval from the starting date. States whose ending value falls outside that interval are marked as having a statistically significant change over the period.
__
Fig.1: U.S. states with improvement in mental health
Only three states showed a statistically significant reduction in unmet mental health care over the period.
__
Fig.2: U.S. states with a deterioration in mental health
Many more states showed a statistically significant increase in unmet mental health care. This comparison identifies where the measure changed, but it does not explain why.
__
Fig.3: Correlation Matrix
The correlation matrix shows little linear association between state income inequality and unmet mental health care. Poverty and urbanization have positive correlations with the outcome, although correlation alone does not establish an effect.
__
Fig. 4: Combined Linear Model
The combined model contains statistically significant coefficients, but the design matrix is ill-conditioned because of the urb_2010 variable. I therefore fit reduced specifications before interpreting the result.
__
Fig. 5: Reduced Linear Models
None of the explanatory variables is statistically significant at the 5% level in the reduced models. Urbanization is significant at the 10% level in one specification, but that is not enough to support a causal claim. The result is a prompt for further study, not evidence that poverty, inequality, or urbanization caused the observed changes.