Work Text:
Chapter 5: Results
Chapter Text
Descriptive Statistics
The sample comprised 4,595 households from the 2022 SCF, with a mean age of 54.47 years (SD = 16.19). Socioeconomic characteristics revealed a predominantly White (62.05%), married (63.22%), and educated sample (49.48% held at least a bachelor’s degree). Notably, 29.10% reported a COVID-19 diagnosis, 6.90% experienced pandemic-related medical financial hardship, and 12.14% received government assistance. Financial knowledge scores averaged 1.89 (SD = 0.92) on a 0– 3 scale. See Table 1 below for the descriptive statistics for the study sample.
Table 1: Descriptive Statistics of Study Sample (N = 4,595) (see the original PDF).
Pearson correlations (see Table 2) showed no coefficients exceeding 0.50, and variance inflation factors (VIFs) were all below 5 (mean VIF = 2.16), indicating no multicollinearity concerns. Expected patterns emerged, such as positive correlations between income, networth, and education (all p < .05).
Table 2: Pearson Correlation Matrix of Study Variables (see the original PDF).
Hierarchical Analysis
The results of the hierarchical logistic regression analysis examining factors associated with precautionary saving behavior show three sequential models, with each model building upon the previous one to examine the effects of demographic factors, Health Belief Model variables, and financial knowledge as a potential moderator (See Table 3).
Our initial analysis of demographic factors alone revealed several important patterns in precautionary saving behavior. Age was negatively associated with precautionary saving (β = - 0.015, **p < .001), indicating that for each additional year of age, the odds of engaging in precautionary saving decreased slightly (OR = 0.985). Race/ethnicity was a significant predictor. Compared to White households, Black households were significantly more likely to report precautionary saving motives ( β = 0.505, **p < .001), corresponding to 65.6% higher odds (OR = 1.656). Higher educational attainment was also positively associated with precautionary saving. Compared to those with less than a high school education, households where the respondent had completed some college (β = 0.371, *p < .01; OR = 1.449) or held a college degree or higher (β = 0.431, **p < .001; OR = 1.539) showed significantly greater odds of precautionary saving behavior. Among employment categories, compared to those who were unemployed or laid off, retired individuals ( β = 0.668, *p < .01; OR = 1.951) and disabled individuals ( β = 0.564, *p < .01; OR = 1.757) were significantly more likely to report precautionary saving.
Model 2 incorporated Health Belief Model factors while controlling for demographic characteristics (Table 2). The significant relationships observed in Model 1 for age, Non- Hispanic Black race, education level, and retired/disabled status remained significant predictors after adding the HBM variables. Among the Health Belief Model factors added in Model 2, facing financial hardship (β = 0.047, *p < .01; OR = 1.048) and receiving government assistance (such as unemployment insurance, stimulus payments, or PPP loans) during the pandemic (β = 0.347, **p < .001) were significant positive predictors of precautionary saving behavior.
Households that received government assistance had 41.4% higher odds of engaging in precautionary saving compared to those who did not receive such assistance. Being diagnosed with COVID-19 (representing perceived susceptibility in this model) did not show a significant direct relationship (p > .05) with precautionary saving behavior in this step of the analysis.
Model 3 examined whether financial knowledge moderated the relationship between financial hardship and precautionary saving. When including the financial knowledge score and its interaction with financial hardship, the previously established relationships remained consistent. Age, Non- Hispanic Black race, education level, retired/disabled status, financial hardship and government assistance maintained their significant associations with precautionary saving. However, financial knowledge was not significant as a main effect (β = - 0.014, p = .838) but the interaction between financial hardship and financial knowledge ( β = 0.103, p = .055) was marginally significant. This suggests that while financial knowledge itself does not directly predict precautionary saving behavior in this sample, it marginally moderated the relationship between financial hardship and precautionary saving. It is noteworthy that the coefficient for financial hardship changed from a small positive value ( β = 0.047) in Model 2 to a larger negative value (β = -0.170) in Model 3 when the interaction term was added.
Overall, these findings highlight the importance of demographic factors—particularly age, race, education, and retirement status—as well as government assistance in predicting precautionary saving behavior. The Health Belief Model provides support through the significance of perceived severity (financial hardship), cues to action (government assistance), and the moderation effect by financial knowledge.
Table 3: Hierarchical Logistic Regression Analysis for Variables Predicting Precautionary Saving Behavior (see the original PDF).
Model fit statistics
To assess whether the addition of variables in each model significantly improved model fit, Wald tests were conducted to evaluate the joint significance of added variables across successive models. The initial model (Model 1) with sociodemographic factors was statistically significant (F(17, ∞) = 3.20, p < .001), indicating that demographic factors collectively helped predict precautionary saving behavior.
The addition of Health Belief Model factors in Model 2 significantly improved model fit, as indicated by a significant Wald test for the joint effect of COVID-19 diagnosis, financial hardship, and government assistance (F(3, ∞) = 4.14, p = .006). This provides evidence that the Health Belief Model factors, when considered together, made a significant contribution to explaining variation in precautionary saving behavior beyond demographic factors alone. The overall model fit for Model 2 was also significant (F(20, ∞) = 3.38, p < .001).
In contrast, the addition of financial knowledge as a moderator in Model 3 did not significantly improve model fit. The Wald test for the joint contribution of financial knowledge score and its interaction with financial hardship was marginally significant (F (2, ∞) = 0.65, p = .055). This finding indicates that the addition of financial knowledge as a potential moderator did not significantly improve model fit beyond what was already explained by Model 2, suggesting that these variables did not contribute significantly to the prediction of precautionary saving behavior beyond what was already explained by the variables in Model 2. Nevertheless, the overall model remained significant (F (22, ∞) = 3.12, p < .001). The Area Under the Curve (AUC = 0.62) for Model 2 also indicated moderate discriminative ability, acceptable for behavioral prediction models in complex social contexts. Based on these model comparison tests and applying the principle of parsimony, Model 2 represents the most efficient model for explaining precautionary saving behavior in this sample. The analysis revealed three key findings in the study. First, perceived susceptibility (operationalized in the model as having a COVID diagnosis), showed no significant influence on precautionary savings. Therefore, H1, was not supported. Second, experiencing perceived severity, measured as households who faced financial hardship, significantly predicted precautionary savings ( β = 0.047, *p < .01), suggesting that economic vulnerability motivates financial preparedness. This finding provides clear support for H2. Third, reflecting cues to action within the Health Belief Model, receiving government assistance (e.g., UI, stimulus, PPP) also showed a statistically significant positive association with precautionary savings ( β = 0.347, **p < .001), corresponding to higher odds of saving among recipients compared to non-recipients.
In addition, predicted probability graphs were created based on the regression model to visualize key relationships and perform robustness check.
Figure 2: Predicted Probability of Precautionary Saving by Race/Ethnicity (see the original PDF).
Substantial differences emerge in the predicted probability of precautionary saving across racial and ethnic groups, even after accounting for socioeconomic covariates. Black households exhibit a notably higher predicted probability (~56%) relative to White households (~41%), consistent with prior research indicating stronger precautionary saving motives among Black households, and potentially reflecting adaptive strategies to perceived economic vulnerability or long-standing structural barriers to wealth accumulation. Hispanic households display a slightly higher probability (~44%) than White households; however, this difference is not statistically significant. Asian and other racial/ethnic groups show similar probabilities to White households (~39%). These persistent differences, independent of socioeconomic controls, may reflect differential perceptions of financial risk or varied historical experiences with income volatility. The findings underscore the importance of culturally informed financial policies aimed at enhancing economic resilience in diverse communities.
Figure 3: Predicted Probability of Precautionary Saving by Government Assistance (see the original PDF).
Receipt of government assistance during the COVID-19 pandemic—such as unemployment insurance, stimulus checks, or Paycheck Protection Program (PPP) funds —is significantly associated with a higher predicted probability of precautionary saving, net of other factors.
Households receiving assistance demonstrated a predicted probability of ~46%, compared to ~37% among non-recipients, representing a statistically significant 9 percentage- point difference (β = 0.347, p < .001). These findings align with the Health Belief Model’s concept of “cues to action,” suggesting that receipt of financial assistance served as a prompt for savings behavior. This supports existing literature indicating that portions of pandemic-related stimulus payments were saved rather than spent. Potential mechanisms include increased liquidity for financially constrained households, heightened awareness of economic fragility, or differential treatment of lump-sum (“windfall”) income. Contrary to traditional economic models positing that public transfers may crowd out private saving, these results suggest that, under conditions of heightened uncertainty, government aid can complement and even encourage precautionary saving. This underscores the potential for well-designed assistance programs to promote prudent financial behavior during economic crises.
Figure 4: Predicted Probability of Precautionary Saving by Employment Status (see the original PDF).
Precautionary saving behavior also varies significantly by employment status after adjusting for covariates. Retired individuals (~52%) exhibit the highest predicted probabilities of saving, significantly exceeding those of unemployed or laid-off individuals ( ~35%). Employed individuals fall in the middle range, while those categorized as “Other/Out of Labor Force”
(e.g., students, homemakers) have the lowest predicted probabilities. These results support existing research suggesting that labor market status meaningfully shapes saving motives. The elevated likelihood of saving among retired and disabled individuals may reflect concerns over healthcare costs, fixed income dependency, or broader reassessments of financial security.
Conversely, the lower likelihood among individuals outside the labor force may be attributable to limited resources or alternative financial priorities. These findings suggest that financial education and policy interventions should be tailored to employment contexts, particularly to support savings behavior among groups facing heightened economic vulnerability.