Background/Context: Previous studies have consistently found that increases in school spending predict positive educational outcomes (Jackson & Mackevicius, 2021). More recent work estimated that ESSER spending overall helped support recovery (Dewey et al., 2024; Goldhaber & Falken, 2024), yet little research has examined which types of spending were related to higher test scores. In prior work (Barragan Torres et al., 2024), we documented how Illinois districts varied in how they spent ESSER funds by district characteristics. Purpose/Research Questions: The current study seeks to understand whether and how ESSER expenditure was related to learning recovery in Illinois. We answer the following research questions: (1) How did a district's "amount" of ESSER spending per pupil relate to their average change in test scores from SY19 to SY23?; and (2) How did a district's types of ESSER expenditures relate to their average change in test scores from SY19 to SY23, controlling for student, school, and district characteristics? Setting: We use state-wide data for all students in Illinois districts. Population/Participants: We drew on three unique data sources. (1) ISBE's ESSER Spending Dashboard provided publicly available data on how each district in Illinois has allocated and spent their ESSER funds; (2) The Illinois Report Card provided publicly available data on district and school characteristics; and (3) A student-level file containing test score, demographic, program, and attendance data for all students who were in grades 3-8 and 11--all testing grades--for each year between SY19 and SY23. Research Design: Our sample included cross-sections of students in grades 3-8 and 11 from SY19 to SY23. Each cross-section included all students who had non-missing test score data and attended a traditional district. For our regression models, we further limited the sample to students who had non-missing data on each of our student-level controls and who attended schools and districts with non-missing data on school- and district-level controls. Our dependent variable was student achievement as measured by Illinois Assessment of Readiness (IAR) scores for grades 3-8 and SAT scores for grade 11. We used two different sets of ESSER spending variables as our primary independent variables, controlling for student, school, and district characteristics. These measures are described below. "Proportion of ESSER Spending: Function-by-Object Codes." Our first set of spending variables were based on the top 20 most commonly used function-by-object codes used by districts over all rounds of ESSER. We created variables to represent a district's proportion of expenditure under each code out of their total ESSER expenditure as of November 2023. "Proportion of ESSER Spending: Aggregating Budget Codes into 5 Categories." We developed five broader categories that grouped similar types of expenditures reported in function-by-object codes. The methods we used to aggregate, as well as the codes contained in each category, are described in our previous study (Authors et al., 2024, p. 7, 13-14). We calculated each district's proportion of expenditure in each category out of their total ESSER expenditure as of November 2023. To prevent spurious associations, we removed outlier districts from each expenditure category, which we identified in the following way: We first took the natural log of each category, which created normal distributions. We classified outliers as observations greater than 1.96 standard deviations above the mean in these natural log distributions. We ran models with and without inclusion of outliers. Analysis: We first described trends in achievement across districts that varied in amount and type of ESSER spending. We grouped districts into quartiles based on their overall amount of per pupil ESSER expenditure as well as their proportion of expenditure in each of our 16 function-by-object codes and our 5 aggregated spending categories. Across quartiles, we show district average scaled scores in SY23 compared to SY19 and SY21. We then analyzed relationships between districts' types of ESSER spending and their achievement using three-level Hierarchical Linear Models (HLM). We estimated how outcomes of students in SY23 compared to outcomes of same-grade students in SY19 and how district-level ESSER spending patterns were related to these between-cohort differences, all else equal. At level 1 (students), we controlled for student demographic characteristics and the proportion of SY21 each student had spent in remote instruction and absent, each of which were significantly related to achievement outcomes. At level 2 (schools), we controlled for a range of Illinois Report Card measures of school characteristics in SY19, using t ratios to determine which controls to keep in our final models. At level 3 (districts), we controlled for several district characteristics in SY19 as well as overall ESSER expenditure per pupil. Our predictors of interest were ESSER spending patterns, which we modeled as predictors of the SY23 slope. The first set of models operationalized ESSER spending using district proportion of spending on the 20 function-by-object codes that were most commonly used statewide as predictors (with salaries and benefits object codes combined, for a total of 16 predictor variables). The second set of models used district proportion of spending on the five aggregated spending categories. Findings: Districts that spent the most ESSER funding per pupil had the lowest pre-pandemic (SY19) achievement on the IAR and SAT. As shown in Figure 1, in elementary and middle grades (grades 3-8), districts in the highest quartile of ESSER expenditure had experienced the greatest achievement declines at the outset of the pandemic, but they recovered slightly more quickly than districts with lower ESSER funding. In high school (grade 11), districts in higher quartiles of expenditure experienced achievement declines at rates similar to districts in quartiles of lower expenditure (see Table 1). Conclusion: Descriptively, patterns between districts' proportion of expenditure on the 16 function-by-object codes and achievement and the other five aggregated spending categories and achievement were unclear and inconsistent. Our HLM models--both with and without inclusion of outliers--showed no statistically significant relationships between achievement and expenditure type. Findings from models excluding outliers are shown in Tables 2 and 3. Possible explanations for our null results include: (1) imprecise and broad measures of district spending; (2) unobserved confounders; (3) insufficient timespan to observe significant relationships.