Instructional Spending Metrics: A Multilevel Analysis Using NCES Data

Educational institutions operate within multiple hierarchical levels with each level typically influencing policies or outcomes at other levels within the hierarchy. For example, decisions made at the state level may influence decisions at the local education agency (LEA) level which may influence decisions at the school building level (as well in the other direction). Using multilevel regression models (or hierarchical linear modeling) and national data sets, this study looks at two common metrics used to describe instructional spending; per pupil dollars and percentage of total dollars allocated to instruction. When estimating multilevel regression models using these two metrics (i.e., per pupil dollars and percentage of the total) as dependent variables and using the same national data set, the findings show that when using per pupil dollars as the dependent variable, individual state characteristics appear to exhibit stronger influence over per pupil dollar allocations while regression estimates using the percentage of total current operating expenditures as the dependent variable appear to show that LEA-level administrators have more local control over the internal allocation of available fiscal resources compared to state-level influences. In other words, the chosen metric (i.e., per pupil amounts vs. percentage of total) may exhibit unintended biases depending on the unit and level of analysis when estimating linear regression models related to educational spending patterns.