Proposition NN and Initiative 195: Unlikely to Solve Colorado’s Education Challenges

Colorado taxpayers could soon be paying more for K-12 education, with no guarantee that the money will reach students. Voters will likely face two major ballot measures in the November election: Proposition NN (created by Senate Bill 135), which is confirmed for the ballot, and Initiative 195, which is awaiting confirmation by the Secretary of State.  

Proposition NN asks voters to give up their tax refunds of overcollected tax revenue under the state’s Taxpayer’s Bill of Rights (TABOR). TABOR puts a constitutional limit on state revenue collection, and Proposition NN would permanently raise that limit by an amount equal to the state’s annual K-12 spending, currently about $4.6 billion. The measure also mandates up to 2% annual increases in K-12 funding over the next decade, with the limit expanding each year. According to the measure’s own fiscal note, this would reduce TABOR refunds by $329.9 million in FY2026-27 and $521 million in FY2027-28, or about $111 and $175 per tax filer, based on Department of Revenue Statistics of Income. 

Even though the TABOR limit increase is permanent, the retained revenue would only be required to go toward education for the first 10 years. After that, the government will have nearly unrestricted authority over excess revenue. 

Initiative 195 takes a different approach, replacing Colorado’s flat 4.4% income tax with a “progressive” or graduated tax, with rates as high as 8.4% on income over $1 million. The measure’s own ballot language estimates it would increase state income tax revenue by $2.7 billion. 

Together, the two measures would redirect well over $3 billion a year in new revenue and forgone tax refunds. While the measures sound appealing to some on paper, what exactly would these expensive measures accomplish? 

Recent Education Trends 

Data from the Colorado Department of Education highlights the trajectory of Colorado test scores. Consistent with national trends, Colorado test scores dropped after the COVID-19 pandemic, as measured by CMAS—Colorado’s standardized educational assessment. To address the trends, Colorado received $1.6 billion in federal COVID-19 relief funding (ESSER), which was distributed to districts at their discretion. K-12 spending in Colorado increased from an average of $17,400 per student in 2019 to $19,757 per student in 2024, after adjusting for inflation and weighting enrollment.  

Part of the increase in per-student spending reflects Colorado’s declining public school enrollment, which decreased from 913,223 in 2019 to 881,065 in 2024. 

Figure 1

Despite this increase in spending, 59% of Colorado districts saw Math scores decline, and 71% saw English Language Arts scores decline. Figure 1 shows these declines and that districts’ scores decreased regardless of how much they increased spending.

Critically, scores in high-revenue districts declined as much as in low-revenue ones. Districts that increased funding by $8,000 or more per student fared no better than those that increased by $1,000. If funding were the issue, higher-revenue districts would be expected to recover faster, but that is not the case. Math declined in 53.8% of the highest-revenue districts compared to 50% in the lowest-revenue districts. COVID-19 was a severe interruption for educational and social development—an issue that increased spending did not fix.

Pre-COVID Education Trends 

The disconnect between spending and outcomes did not begin with the pandemic. In the years leading up to COVID, school districts were generally increasing spending and improving outcomes, as 85% of districts improved in Math and 87% in ELA. However, achievement varied during this period regardless of revenue or enrollment. 

Figure 2

Colorado students were progressing well in the pre-COVID era for both ELA and Math. Nearly all districts showed improvement in test scores, as shown in Figure 2. However, on average, the biggest improvements were not associated with the biggest increases in per-pupil spending. Most districts saw gains in that period regardless of how much their spending changed, suggesting other factors drove the improved results. 

Both the pre-COVID and post-COVID comparisons show that academic achievement moved independently of spending, whether scores were declining or improving. However, each figure only covers a single period. To test whether these trends hold across all years, a two-way fixed-effects model was estimated on the full 2015-2024 panel. 

Fixed-Effects Model 

This model gives each district its own typical test score level—a baseline—and checks whether scores moved up or down relative to the baseline when its spending or Free and Reduced Lunch (FRL) rates changed over time. FRL rates were used as a proxy for income level in districts. The model is not comparing Aspen to Pueblo; it is comparing Aspen in 2019 to 2017, and separately for Pueblo, and averaging the within-district comparisons. 

The “two-way” factor (district + year fixed effects) removes anything that affects every district each year alike (test redesign, funding formula changes, the COVID disruption). This approach shows association within districts over time, not causation. 

This model was chosen because Colorado’s districts differ enormously. Comparing across districts (richer vs. poorer) would conflate spending with other factors (community wealth, local labor markets, historical achievement patterns). However, the model cannot explain why, for example, Aspen and Pueblo differ in the first place. It only speaks to within-district movement.  

The model was run twice for each subject, once across all districts and once restricted to districts with 500 or more students. This was to confirm the results were not being driven by very small, data-sparse districts. 

Table 1 

 
These results in Table 1 reinforce the pattern shown in Figures 1 and 2, now tested across the full panel rather than a single period. A $1,000 increase in per-pupil revenue is associated with a decrease of between 0.015 and 0.021 percentage points in Math proficiency (depending on the sample), which is not statistically distinguishable from a score increase of zero. Additionally, a $1,000 increase is associated with a 0.005-0.012 increase in ELA proficiency, also not statistically distinguishable from zero. 

The near-zero R² further suggests that within-district changes in spending and poverty explain almost none of the year-to-year movement in test scores. This finding held consistently across several robustness checks, detailed in the appendix. 

According to the model, the Free and Reduced Lunch estimates indicate that lower test scores are slightly correlated with higher FRL rates within a district. However, these estimates are also not significant. 

Spending Accountability 

Beyond the outcomes measure, Proposition NN and Initiative 195 lack the fiscal accountability to ensure that the new revenue will be tracked or reach classrooms at all.  

SB26-135’s own fiscal note “assumes that school districts are not required to report specifically how the funds are used to CDE.” Essentially, it directs districts to allocate funds for classrooms and teachers but does not ensure accountability within districts.  

While Initiative 195 requires the spending to be audited, it only confirms that the revenue is directed toward one of three broad categories: Colorado Public School Education, health care, or early child care and education. There is no formula deciding how much money is split between those three areas, or how much specifically reaches classrooms. 

This structural gap has real consequences. Colorado has a track record of education spending flowing away from instruction. 

One study from the Reason Foundation shows that from 2002 to 2023, Colorado’s spending on employee benefits in education increased by 155.8%—the 6th-largest increase in the nation—while non-teaching staff grew by 48.2% —the 7th-largest increase in the nation. These increases have some justification, but their pace surpasses spending that reaches students directly. A separate Reason Foundation analysis of national K-12 staffing trends found that these increases in non-instructional roles are not associated with gains in student achievement. Colorado’s pattern of increased education spending without verification of how the funds are used could explain why additional revenue did not translate into improved outcomes. Proposition NN and Initiative 195 are no different. 

Without fiscal accountability, Colorado risks misallocating education spending, as its history illustrates. Even accounting for the state’s history of underfunding, the composition of education spending raises questions about whether new revenue would reach students. 

Conclusion 

COVID caused real damage to education, but allocating more revenue to school districts has not proven to resolve the shortfalls in test scores. These gaps are not a funding problem, but a developmental one stemming from a loss of foundational learning. Proponents of increased spending appear to assume that funding alone can offset the learning lost due to disrupted instruction. 

If voters are expected to forgo the taxpayer protections in exchange for more school funding, they deserve a guarantee that the $3 billion will reliably improve student outcomes. The evidence does not confirm that it does. Voters should think deeply about whether surrendering their refunds or taxing the rich for K-12 spending is the right path to fix the developmental challenges facing Colorado students  

Appendix 

Test Score Data 

Colorado Measures of Academic Success (CMAS) testing data were obtained through the Colorado Department of Education. The data combined Grades 3-8, used only English Language Arts and Math, and reported the percentage of students who met or exceeded expectations for each school district. 2020 and 2021 were excluded from the data due to the absence or disruption of statewide testing.  

CMAS proficiency percentages reflect only students who participated in testing. Districts or years with higher opt-out rates may yield different results than full-participation years, and this analysis does not adjust for participation rates. 

Revenue Data  

Revenue data were obtained through the Colorado Department of Education per-pupil revenue files, using operating expenditure only. This excludes capital outlay, so one-time spending on items such as building construction, major renovations, or land purchases is not included. All revenue figures were adjusted to real 2024 dollars calculated from the Consumer Price Index for All Urban Consumers. Statewide spending figures are enrollment-weighted rather than computed as a simple mean across districts, so that per-pupil averagesreflect the experience of the average student. District renames were matched by district code to ensure continuity across the panel. 

Free and Reduced Lunch (FRL) Data 

FRL eligibility data was obtained through CDE, covering Fall counts of free/reduced lunch eligibility by district. Due to district name changes throughout the period, districts were merged by their 4-digit code. 

A small number of districts show suppressed FRL rates, due to small district enrollment or missed reporting. 
While FRL remains the best available proxy for income-level, figures are likely biased due to the Community Eligibility Provision (CEP) and Healthy School Meals for All (HMSA) initiatives. CEP is a federal program that allows schools with enough students already directly certified for benefits (such as SNAP, TNAF, foster care, etc.) to offer free meals to all students without collecting household applications. HMSA guarantees free meals to every student statewide. Both of these measures reduce families’ incentives to complete and submit the FRL paperwork. FRL rates are therefore likely to be underreported. 

Scatterplot Sample 

For the scatterplot analysis, districts with at least 500 students enrolled were included. Six districts were excluded as structurally non-comparable to geographic districts: Charter School Institute, CO Digital BOCES, Expeditionary BOCES, San Juan BOCES, CO School for the Deaf and Blind, and Revere SD. After the exclusions, there were 87–88 districts in the final sample, depending on subject and period. This covers 97.7% of Colorado’s enrolled students (73 districts were excluded, representing approximately 2.3% of students).  

Fixed-Effects Regression Appendix 

Model Specification 
𝑆𝑐𝑜𝑟𝑒𝑖𝑡=𝛽1×𝑅𝑒𝑣𝑒𝑛𝑢𝑒𝑖𝑡+𝛽2×𝐹𝑅𝐿𝑖𝑡+𝛼𝑖+𝛾𝑡+𝜀𝑖𝑡Scoreit=𝛽1×Revenueit+𝛽2×FRLit+𝛼i+𝛾t+𝜀it
 

• 𝑆𝑐𝑜𝑟𝑒𝑖𝑡Scoreit = CMAS proficiency percentage for district i in year t 

• 𝛽1,𝛽2𝛽1,𝛽2 = coefficients – how much Score moves per unit change in Revenue and FRL, within a district, net of year effects 

• 𝑅𝑒𝑣𝑒𝑛𝑢𝑒𝑖𝑡Revenueit = real per-pupil revenue (in thousands of 2024 dollars) for district i in year t 

• 𝐹𝑅𝐿𝑖𝑡FRLit = free/reduced lunch rate for district i in year t 

• 𝛼𝑖𝛼i = district fixed effect – a separate intercept for each district, absorbing everything permanent/time-invariant about that district 

• 𝛾𝑡𝛾t = year fixed effect – a separate intercept for each year, absorbing anything that affected all districts alike in that year 

• 𝜀𝑖𝑡𝜀it = error term – everything else (noise, unmeasured factors that vary both within-district and over time) 

A two-way fixed-effects model was estimated using the plm package in R, regressing CMAS proficiency (Math and English Language Arts) on per-pupil revenue (in thousands of 2024 dollars) and Free and Reduced Lunch rates. The specification is contemporaneous, meaning fiscal year t is paired with spring t test scores. Standard errors are clustered by district. Clustering is necessary because a Wooldridge test finds significant serial correlation in the panel (p<0.001) for both subjects. This is evidence that a district’s year-to-year outcomes are correlated over time, which uncorrected standard errors would not account for. 

The model uses Math and ELA outcomes separately, across Colorado school districts from 2015-2024 (excluding 2020-2021). The model was estimated once across all districts and once restricted to districts with 500 or more enrolled students to confirm results were not dependent on small, data-sparse districts. 

Districts with suppressed FRL or CMAS data in a given year were dropped from that year’s estimation and are assumed to be unrelated to the error term conditional on fixed effects. 

Conditions 
The fixed-effects estimator requires that revenue and the FRL rate be uncorrelated with unobserved factors driving test scores, once district and year effects are removed. This is supported by Colorado’s school finance formula, where per-pupil allocations are determined by enrollment, at-risk pupil counts, and cost factors. Granger-style tests find no evidence that prior scores predict future revenue or vice versa. Local revenue from mill levy overrides and bonds, which may correlate with community characteristics, is not tested. 

Districts with little year-to-year change in spending or FRL contribute limited information to the estimates, which is consistent with the analysis’ low R-Squared. 

Robustness Checks 
A quadratic relationship between revenue and achievement initially appeared significant for ELA but did not hold under district-clustered standard errors and was driven by a small number of high-revenue districts. A Pesaran CD test finds weak evidence of cross-sectional dependence (p=0.058-0.071), though this test’s power is limited by the panel’s unbalanced structure following the 2020-2021 testing gap. 

*Logan McCahill studies economics and statistics at the University of Georgia. He is a fiscal policy research intern at Independence Institute, a free market think tank in Denver.