Our Students are Hungry: Food Insecurity Impacts More Than We Knew, Part 2 — Structuring First-Year Retention at a Regional Public Institution: Validating and Refining Bowman’s Model

Daniel Collier, PhD
Aug 21 · 6 min read

By: Daniel A. Collier and Dan Fitzpatrick

Last week my colleague, Dan Fitzpatrick, and I wrote about the factors that correlate with being more food insecure, and about how food insecurity links both with students’ non-cognitive attributes and with first-semester performance and persistence.

This week we discuss the casual influences of food insecurity on first-year students’ incoming non-cognitive attributes and college performance and persistence. As higher education researchers are still in the early phases of understanding college student food insecurity, the field has not had many opportunities to conduct causal research on the matter — some papers exist but none that we are aware of examining persistence as the outcome. In addition to the fact that food insecurity research remains in an emergent stage, few research agendas are able (or intend) to capture the robust non-cognitive and engagement data that our project did.

This type of study is important because the causal nature of the analysis allows us to correctly understand how food insecurity influences first-year students’ non-cognitive attributes, engagement, college performance, and ultimately fall-to-fall retention. This type of research can make a stronger case regarding the importance of the influences of food insecurity — and that if food insecurity was eased, how other factors would be affected.

Before we begin to discuss our findings, we must briefly explain the differences between typical regressions and a structural equation model (SEM). Typical regressions, like those we highlighted last week, indicate whether and the degree to which direct relationships exist between the outcome examined and variables in the model. These regressions are the typical avatar of “correlation does not equal causation” and do not specify which of two correlated variables causes the other to change. On the other hand, structural equation modeling is a causal method that examines the direct and indirect influences of one variable on another. For example, last week we reported that a direct relationship exists between food insecurity and amotivation. However, we do not know in which direction the relationship flows, i.e., whether food insecurity influences amotivation or vice versa. We only know a relationship exists between the variables. SEMs can tell us how food insecurity, directly and indirectly, influences variables within a model — as well as producing a total (direct + indirect) influence statistic. These measures of influence let us better understand the importance of a given variable within a given structure.

Instead of generating a new SEM structure, we decided to take the data we collected and plug it into a model already developed and validated by Bowman, et al. (2019) — see structure below. The general premise of the Bowman model is that students’ non-cognitive attributes are more important than previously believed.

The Bowman Model

We tested two models [1] — one without food security included (Model 1) and one with food security (Model 2) [2]. Despite examining different students (at the site) and not collecting the same data [3] but testing similar non-cognitive concepts — our models confirm Bowman’s structure for how students’ non-cognitive attributes directly influence students’ first-year experience and performance. The image below represents Model 2 — these are the direct influences our model has found (solid lines denotes significance).

Given our prior findings on persistence from fall-to-spring, we hypothesized that the inclusion of food security would yield a stronger, significant direct influence from financial health to first-year fall-to-fall retention. Instead, including food security reduced the direct influence from r=.16 to r=.12 — moving the direct link out of significance.

Although the inclusion of food insecurity weakened the direct influence on persistence, we found it strengthened influences of financial health on high school GPA (from r=.09 to r=.25) and social adjustment (from r=-.05 to r=.11) — in both cases moving the influences to significance. Just as in our model without food security (Model 1), the Bowman model did not detect a significant direct influence from the financial variable to high school GPA. The strengthened correlation found in Model 2 (w/food security) better aligns with a large body of research on family income and high school performance, and with our current understanding of the influences of food insecurity on high school performance and students’ collegiate experiences.

Solid Lines = Significant Relationships, Dashed Lines = Non-Significant (p=.05)

Some may look at these findings and think that food security’s influence on fall-to-fall persistence may not be important — a fair point. However, SEMs produce indirect influences not highlighted in the visual representations provided. In this case, the indirect influences reinforce our understanding of the total influence of food insecurity.

Behind the scenes, incorporating food security increased the financial health variable’s indirect influence on the commitment to persist from r=-.01 to r=.05 — not only flipping the direction of influence but moving the total influence to significance [4]. Therefore meaning, higher food security will indirectly but positively influence students’ commitment to persist (or lower Amotivation).

Financial health’s indirect impact on college GPA exhibited the largest increase, moving from r=.29 to r=.51. Overall, the inclusion of food security to the financial health variable produced a high indirect impact on college GPA — which has the strongest direct influence on persistence (r=.46) and is tied with high school GPA for strongest overall influence. Therefore, bolstering college GPA is incredibly helpful in strengthening persistence.

For the interested, the table below provides direct, indirect, and total influences of the Bowman model and both models we generated. We highlighted the financial health differences from Model 1 (no food security) to Model 2 (including food security) — Yellow = Direct Influence, Blue = Indirect Influence, and Green = Total Influence.


So what? Our models emphasize that the impacts of food insecurity on first-year college students’ non-cognitive attributes, engagement, and performance are more complex than can be explained by descriptive studies. Although the total influence is complex, the results are easy to interpret- in that if institutions or stakeholders were to ease first-year students’ food insecurity, students would:

  1. Exhibit less incoming amotivation
  2. Demonstrate better early social adjustment (closer to faculty, staff, and peers)
  3. Earn higher first-year college GPA

Next week, we will discuss our findings on a 2-armed randomized assignment nudging intervention and wrap this arc up. Spoiler alert:

Null effects are found for everything EXCEPT for one outcome — it has something to do with food insecurity.


[1] Food Insecurity, Financial Stress, and Amotivation (Commit to Persist) were all reverse coded so that higher scores indicated increased food security, less stress, and lower amotivation. This makes the outcomes easier to read and understand.

[2] Like Bowman, in Model 1, “Financial Means” remained only as a financial distress scale. In Model 2 we turned the variable into a latent construct that included financial stress and food insecurity.1 In both models, we relabeled “Financial Means” to “Financial Health” as this was a better representation of the variable. Both models illustrate the optimal fit. The image below illustrates Model 2.

[3] Bowman, et al. captured and included the Grit scale in their SEM. They then generated non-cognitive, social adjustment, and commitment to persistence from untested scale responses. Our study used only validated survey scales — see last week’s post for links to each scale. In our paper, we made theoretical and research-based justifications for the inclusion of each scale under and how they mimic the same concepts Bowman et al. included. We also have the same temporal limitations — all student survey data were collected at the same time at the start of the fall semester (August/Sept 2018) but the college GPA and the outcome of persistence were collected later at the end of the spring semester (April 2019).

[4] Commitment to persist was measured by students’ Amotivation. Which as a reminder (footnote 1) was reverse coded so that a higher score meant lower Amotivation.

Daniel Collier, PhD

Written by

Post-Doctoral Researcher at Western Michigan University, eyeing the next step.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade