In answer to our first question, we found that greater school disadvantage was significantly predictive of increased chronic PM2.5
exposure as well as the frequency of peak PM2.5
exposures at schools. School disadvantage was more strongly related to peak exposures than chronic exposures (e.g., the ratio of the coefficient to the standard error was 8.10 in the model predicting the daily annual average and 11.10 in the model predicting the number of peak days). Our findings associating racial/ethnic minority composition and economic deprivation (combined into one index) with the chronic PM2.5
metric (i.e., annual daily average) aligns with other studies in Michigan [6
], and California [14
]. While our peak PM2.5
findings align with our findings from the model predicting chronic PM2.5
, they illustrate another dimension of environmental injustice that is less often investigated. How those peak exposures might translate into negative effects for children is still an open question. In terms of how they affect academic proficiency at the school level, we address that with our second question.
In answer to our second question, chronic annual average exposure was positively related to the percentage of students with low proficiency in math and ELA initially, but those significant effects were explained away by the inclusion of social disadvantage. Unlike chronic exposures, the frequency of peak exposures was positively associated with the percentage of students with low proficiency in math and ELA, even after accounting for school disadvantage. The effect size of the acute exposure coefficient was similar between the math and ELA models. This has been found in other studies [7
], although some have found that math [10
] or verbal abilities [45
] are more affected by air pollution.
Why peak exposures were more closely related to low academic proficiency than chronic exposures is an open question. Certainly, in Salt Lake County, those two metrics are related to each other (Pearson’s correlation = 0.550). It is possible that repeated peak exposures have more severe impacts on the brain than chronic exposures. As air pollutants enter the body, they induce neuroinflammation as a result of activated microglia, the immune cells of the brain that regulate neuroinflammation [72
]. Neuroinflammation contributes to cell loss within the central nervous system, which is believed to be linked to cognitive deficits [5
]. It may be that neuroinflammation triggered by frequent peak exposures is more damaging to children’s brains than the effects caused by a lower concentration of daily exposures. Relatedly, research on peak exposures in cars has shown that in-vehicle PM2.5
contains a high level of chemicals that causes oxidative stress. The body responds similarly to these chemicals and the PM2.5
to cope with the reactive oxygen species contained in both. The dual exposure leads the body to overreact, which may be destructive to DNA [73
]. A similar overreaction could be behind the damaging effects of peak exposures at school. However, these hypotheses are tentative.
These results contribute to a growing body of literature suggesting that PM2.5
standards are not low enough [74
]. In this case, we defined peak exposures as the 95th percentile, which was just under 23.00 μm3
; the federal 24 h standard is 35.00 μm3
. While not focused on lowering the PM2.5
standard, the State of Utah has taken steps to reduce peak exposures to PM2.5
. Recognizing that mass transit is key to reducing traffic pollution, the state passed House Bill 0353 in 2019, which provides free public transit fees on “red air” days to encourage ridership and reduce the use of personal vehicles [77
]. The State is also trying to better quantify the environmental impacts of new developments and address those impacts on proximate communities, as evidenced by the passage of Sen. Escamilla’s Senate Bill 0112 in 2020 [78
]. We concur with the assertion of Pastor et al. [16
] that, “in some sense, schools are a barometer for society as a whole: Improving air quality with kids in mind can improve air quality for everyone” (p. 356).
School disadvantage was a strong and robust predictor of having lower proficiency in math and ELA, controlling for the other variables in the model (including pollution). Our results align with extant knowledge about how social factors are critical in predicting academic proficiency. The problem of poverty (measured by the proportion of students on free and reduced-price meals and Title 1 status in our index) often shares a dynamic relationship with poor health outcomes [23
]. Poor students may lack access to proper nutrition, healthcare and medication, which all may negatively impact academic outcomes [16
]. Poor parents often have lower educational attainment than more affluent parents, and parental educational attainment is an important social factor shaping academic proficiency [16
]. Racial/ethnic minority students may have an unequal educational experience compared to White students as a result of differential treatment and stigmatization. For example, racially biased interactions with teachers and peers negatively impact racial/ethnic minority students’ academic experiences [79
]. Differential teacher treatment is witnessed through teachers having higher expectations of White and Asian students and lower expectations of Black and Hispanic students as well as teachers providing more positive feedback to White students than racial/ethnic minority students [80
]. Steele and Aronson [81
] found that Black students were subject to a stereotype threat effect that caused them to underperform on tests in comparison to White students. An awareness of racial stereotypes about Black students’ intellectual ability and feelings of membership in a stigmatized group contributed to reduced test performance [81
], and a recent study testing Black third graders had similar findings [82
]. Racial/ethnic minority students are also sometimes immigrants or children of immigrants and English language learners. This is an added educational challenge [79
], especially when standardized tests are administered in English [16
Results addressing our second question, in terms of the insignificant finding for chronic pollution as well as the significant findings for peak exposures and school disadvantage, shed light on the social and environmental structure of academic disparities in this particular context and might be relevant for the design of interventions. In order to improve math and ELA proficiency, it is imperative to reduce barriers that accompany low incomes and racial/ethnic minority status. This insignificant finding does not necessarily indicate that chronic environmental exposures have no actual influence on academic proficiency. Several prior individual-level studies have shown that they do [7
]. What our findings show is that social disadvantage fully encompasses the effect of chronic exposures on academic proficiency in Salt Lake County, but that is not the case with peak exposures. When accounting for environmental and social inequalities—rather than demonstrating that one factor is important and the others are not—results from such analyses highlight the multiple forms of jeopardy that affect children [83
In answer to the third question, we found synergies between the peak PM2.5 exposures and school disadvantage for math, but not for ELA. Advantaged schools’ math proficiency was significantly and negatively impacted by higher frequency peak PM2.5 exposures, but the frequency of peak exposures did not have a significant impact on math proficiency in disadvantaged schools. While we found that more socially advantaged schools were relatively more affected by additional days of peak pollution, it is important to note that, in absolute, unadjusted terms, higher proportions of students were below proficient in socially disadvantaged schools. At socially disadvantaged schools, 45.56% and 45.82% tested below proficient in math and ELA, respectively, while those corresponding percentages were 15.40% and 13.78% at socially advantaged schools. It is also the case that advantaged schools (i.e., those ≤ one standard deviation below the mean) averaged 16 peak days in 2016 while the disadvantaged schools (i.e., those ≥ one standard deviation above the mean) averaged 20 days; this difference was statistically significant (p < 0.001) as per an independent samples t-test (table not shown).
This interaction effect suggests that the effect of peak PM2.5
exposures on decreased math proficiency in the base model was driven by the stronger association in more advantaged schools. This is presumably because in more disadvantaged schools there are more influential social determinants of low academic proficiency, closely related to economic deprivation and minority race/ethnicity, e.g., food and housing insecurity, reduced access to quality physical and mental health care, experiences with discrimination, and financial stress [79
]. Those factors are unmeasured in our models, but they likely combine as a constellation of risks that influence children’s academic proficiency in disadvantaged schools to a greater degree than air pollution, even though PM2.5
exposures are higher. In contrast, in more socially advantaged schools, that constellation of risk factors is much less present, enabling variation in air pollution to exert a stronger association with academic proficiency.
Pearce et al.’s [36
] findings suggest that environmental predictors mattered more in wealthy areas than in poor areas because health needs (e.g., access to health care, medications, housing, and nutrition) were already being met. Grineski et al. [83
] posed the following hypotheses, after finding that social factors explained away the effect of air pollution on asthma hospitalization rates in El Paso Texas: “if all members of a population had equal access to needed health resources (e.g., medical care, healthy homes), outdoor environmental exposures would play a more important role in explaining geographic disparities in health than they do in socially unequal communities” (p. 43). In this case, it seems like our interaction effects findings for math provide some support for that hypothesis. It is worth noting that while the interaction between school disadvantage and peak pollution was not statistically significant in the ELA model, the directionality of the findings aligns with the math results.
Why the interaction was significant when predicting math but not ELA is unclear. A recent national Chinese study found that verbal abilities were more affected by air pollution exposure—measured as a city-level index of sulfur dioxide, nitrogen dioxide, and particulate matter equal to or smaller than 10.00 μm (PM10
)—than were math abilities across two waves of data collection, i.e., 2010 and 2014, in an ‘all ages’ sample [45
]. It is possible that ELA abilities are more directly affected by pollution such that school disadvantage is less of a modifier of that associations. However, other studies have found children’s math abilities to be more affected by PM2.5
]. Clearly, additional research on how pollution affects student performance in different subject matter areas is needed.
Limitations and Future Directions
Our study has some limitations. We only examined outdoor air pollution at schools. However, it is established that personal exposure to air pollution is driven primarily by outdoor pollution levels [88
], and that indoor and outdoor air pollution are highly correlated [89
]. Future research should seek to incorporate indoor and outdoor measures of pollutants at schools.
There are also limitations with the SAGE testing data. Test dates are not available and so we are unsure exactly when the tests were conducted beyond the “end of the school year”. To provide an estimation of pollution exposures that might be affecting the children, we used the year before the test, which is imprecise yet captures the general levels of PM2.5
air pollution at each school. We were not able to include any measures of air pollution before 2016, even though it is well established that early exposures are harmful [28
], as we were not able to track individual children as they progressed through primary school. Using panel data on individual children is an important next step to address many of these limitations.