Exploring New Hampshire Statewide Assessment Results
Source:vignettes/newhampshire-assessment.Rmd
newhampshire-assessment.RmdNew Hampshire’s statewide assessment results reach back to 2009. The data spans two very different testing regimes: the New England Common Assessment Program (NECAP, 2009-2014) and the Smarter Balanced / NH SAS assessments (2015-2025). 2020 is absent statewide because the federal COVID assessment waiver cancelled spring testing.
A few things to keep in mind throughout:
-
NH DOE never publishes a raw student count. Every
count is reported as a 5-wide band (e.g.
"35 - 40"), son_testedis alwaysNAand the bounds live inn_tested_lower/n_tested_upper. We never invent a point value. -
Suppressed cells are
NA, never 0. A school with fewer than 11 tested students is masked, and that mask is preserved. -
NECAP and Smarter Balanced are not comparable.
Different tests, scales, and cut scores. The
assessment_eracolumn flags the 2014/2015 boundary, and Story 3 below shows exactly why you must not read across it.
# Bundled NH DOE assessment files: 2009-2019 and 2021-2025 (no 2020).
asmt <- fetch_assessment_multi(c(2009:2019, 2021:2025), use_cache = TRUE)
asmt |>
count(assessment_era, type) |>
arrange(assessment_era, type)
#> assessment_era type n
#> 1 NECAP School 27631
#> 2 SBAC_NHSAS District 218964
#> 3 SBAC_NHSAS School 342716
#> 4 SBAC_NHSAS State 22811. New Hampshire math never recovered from COVID
State-level results are published from 2018 onward (NECAP and the early Smarter Balanced years are reported at the school level only). Across that window, math tells a stark story: statewide proficiency sat at 48% in 2018 and 2019, collapsed to 38% in the 2021 return-to-testing year, and five years later is still stuck at 42%. Reading dipped too, but has nearly clawed its way back.
state_trend <- asmt |>
filter(
is_state, grade == "TOTAL", subgroup == "total_enrollment",
subject %in% c("math", "reading")
) |>
select(end_year, subject, pct_proficient) |>
arrange(subject, end_year)
stopifnot(nrow(state_trend) > 0)
state_trend
#> end_year subject pct_proficient
#> 1 2018 math 48
#> 2 2019 math 48
#> 3 2021 math 38
#> 4 2022 math 40
#> 5 2023 math 42
#> 6 2024 math 41
#> 7 2025 math 42
#> 8 2018 reading 58
#> 9 2019 reading 56
#> 10 2021 reading 52
#> 11 2022 reading 51
#> 12 2023 reading 52
#> 13 2024 reading 53
#> 14 2025 reading 55
ggplot(state_trend, aes(end_year, pct_proficient, color = subject)) +
geom_line(linewidth = 1.1) +
geom_point(size = 2) +
geom_vline(xintercept = 2020, linetype = "dotted", color = "grey50") +
annotate("text", x = 2020, y = 36, label = "2020: no test\n(COVID waiver)",
size = 3, color = "grey40", hjust = -0.05) +
scale_color_manual(values = c(math = "#c0392b", reading = "#2c7fb8")) +
scale_y_continuous(limits = c(30, 65)) +
labs(
title = "NH statewide proficiency, all students (SBAC / NH SAS era)",
subtitle = "Math fell to 38% in 2021 and has not recovered; reading nearly has",
x = "School year (end)", y = "% proficient (levels 3 & 4)", color = "Subject"
) +
theme_minimal()
2. A 20-point economic divide in math, baked in before COVID
Economically disadvantaged students (the econ_disadv
subgroup) trail the statewide average by roughly 20 percentage points in
math, every single year. The gap was 21 points in 2018 and was still 19
points in 2025. COVID lowered both lines without meaningfully closing
the distance between them.
ses_gap <- asmt |>
filter(
is_state, grade == "TOTAL", subject == "math",
subgroup %in% c("total_enrollment", "econ_disadv")
) |>
select(end_year, subgroup, pct_proficient) |>
pivot_wider(names_from = subgroup, values_from = pct_proficient) |>
mutate(gap = total_enrollment - econ_disadv) |>
arrange(end_year)
stopifnot(nrow(ses_gap) > 0)
ses_gap
#> # A tibble: 7 × 4
#> end_year total_enrollment econ_disadv gap
#> <int> <dbl> <dbl> <dbl>
#> 1 2018 48 27 21
#> 2 2019 48 27 21
#> 3 2021 38 18 20
#> 4 2022 40 21 19
#> 5 2023 42 22 20
#> 6 2024 41 22 19
#> 7 2025 42 23 19
ses_long <- ses_gap |>
select(end_year, total_enrollment, econ_disadv) |>
pivot_longer(-end_year, names_to = "group", values_to = "pct")
ggplot(ses_long, aes(end_year, pct, color = group)) +
geom_line(linewidth = 1.1) +
geom_point(size = 2) +
scale_color_manual(
values = c(total_enrollment = "#2c3e50", econ_disadv = "#e67e22"),
labels = c(total_enrollment = "All students", econ_disadv = "Econ. disadvantaged")
) +
scale_y_continuous(limits = c(0, 55)) +
labs(
title = "NH math proficiency: all students vs. economically disadvantaged",
subtitle = "A persistent ~20-point gap that COVID did not close",
x = "School year (end)", y = "% proficient (levels 3 & 4)", color = NULL
) +
theme_minimal()
3. The 2015 “cliff” is a test change, not a collapse in learning
In 2015 New Hampshire retired NECAP and adopted the Smarter Balanced
assessment, which uses a substantially higher proficiency bar. Looking
at the median school’s grade-3 math proficiency, the rate appears to
fall off a cliff, from about 71-75% under NECAP (2012-2014) to roughly
53-58% under Smarter Balanced (2015 onward). This drop is an
artifact of the cut-score change, not real learning loss. It is
exactly the kind of comparison the assessment_era flag
exists to prevent. (The later 2021 dip, by contrast, sits entirely
inside the Smarter Balanced era and is comparable, which is why
it reads as a real COVID effect.)
median_g3_math <- asmt |>
filter(
type == "School", subject == "math", grade == "03",
subgroup == "total_enrollment", !is.na(pct_proficient)
) |>
group_by(end_year, assessment_era) |>
summarise(median_prof = median(pct_proficient), n_schools = n(), .groups = "drop") |>
arrange(end_year)
stopifnot(nrow(median_g3_math) > 0)
median_g3_math
#> # A tibble: 16 × 4
#> end_year assessment_era median_prof n_schools
#> <int> <chr> <dbl> <int>
#> 1 2009 NECAP 72 218
#> 2 2010 NECAP 75 200
#> 3 2011 NECAP 77 209
#> 4 2012 NECAP 74 200
#> 5 2013 NECAP 75 211
#> 6 2014 NECAP 71 220
#> 7 2015 SBAC_NHSAS 53 235
#> 8 2016 SBAC_NHSAS 58 232
#> 9 2017 SBAC_NHSAS 55 229
#> 10 2018 SBAC_NHSAS 57.5 234
#> 11 2019 SBAC_NHSAS 59 228
#> 12 2021 SBAC_NHSAS 45 215
#> 13 2022 SBAC_NHSAS 53 229
#> 14 2023 SBAC_NHSAS 53 223
#> 15 2024 SBAC_NHSAS 52 224
#> 16 2025 SBAC_NHSAS 54 225
ggplot(median_g3_math, aes(end_year, median_prof, color = assessment_era)) +
geom_line(aes(group = 1), linewidth = 0.6, color = "grey70") +
geom_point(size = 3) +
geom_vline(xintercept = 2014.5, linetype = "dashed", color = "grey40") +
annotate("text", x = 2014.5, y = 40, label = "NECAP -> Smarter Balanced\n(not comparable)",
size = 3, color = "grey30", hjust = 1.02) +
scale_color_manual(values = c(NECAP = "#16a085", SBAC_NHSAS = "#8e44ad")) +
labs(
title = "Median school-level grade-3 math proficiency, by year",
subtitle = "The 2015 drop reflects a harder test, not falling achievement",
x = "School year (end)", y = "Median school % proficient", color = "Test regime"
) +
theme_minimal()
Working with the data
The tidy schema is long: one row per entity x subject x grade x
subgroup. Entity flags (is_state, is_district,
is_school) make filtering simple, and every percentage
column is on a 0-100 scale.
asmt |>
filter(is_school, subject == "reading", grade == "TOTAL",
subgroup == "total_enrollment", end_year == 2025) |>
select(district_name, school_name, n_tested_lower, n_tested_upper,
pct_proficient, pct_participate, avg_score) |>
arrange(desc(pct_proficient)) |>
head(10)
#> district_name school_name n_tested_lower
#> 1 Portsmouth New Franklin School 115
#> 2 Hollis-Brookline Cooperative Hollis-Brookline High School 140
#> 3 Bedford Bedford High School 285
#> 4 Grantham Grantham Village School 145
#> 5 Jackson Jackson Grammar School 20
#> 6 Lyme Lyme Elementary School 125
#> 7 Oyster River Coop Oyster River High School 175
#> 8 Dresden Frances C. Richmond School 345
#> 9 Lafayette Regional Lafayette Regional School 55
#> 10 Mason Mason Elementary School 50
#> n_tested_upper pct_proficient pct_participate avg_score
#> 1 120 88 98 NA
#> 2 145 87 88 NA
#> 3 290 86 90 NA
#> 4 150 86 98 NA
#> 5 25 86 100 NA
#> 6 130 86 98 NA
#> 7 180 86 93 NA
#> 8 350 85 96 NA
#> 9 60 84 98 NA
#> 10 55 84 98 NAAvailable subgroups grow over time. The NECAP era (2009-2014) reports all students only; the Smarter Balanced / NH SAS era adds race/ethnicity, gender, economic status, special education, English learners, and additional special populations.
asmt |>
filter(end_year == 2025) |>
distinct(subgroup) |>
arrange(subgroup)
#> subgroup
#> 1 asian_pacific_islander
#> 2 black
#> 3 econ_disadv
#> 4 female
#> 5 foster_care
#> 6 hispanic
#> 7 homeless
#> 8 lep
#> 9 male
#> 10 military_family
#> 11 multiracial
#> 12 native_american
#> 13 not_econ_disadv
#> 14 not_special_ed
#> 15 special_ed
#> 16 total_enrollment
#> 17 white
sessionInfo()
#> R version 4.5.0 (2025-04-11)
#> Platform: aarch64-apple-darwin22.6.0
#> Running under: macOS 26.1
#>
#> Matrix products: default
#> BLAS: /opt/homebrew/Cellar/openblas/0.3.30/lib/libopenblasp-r0.3.30.dylib
#> LAPACK: /opt/homebrew/Cellar/r/4.5.0/lib/R/lib/libRlapack.dylib; LAPACK version 3.12.1
#>
#> locale:
#> [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
#>
#> time zone: America/New_York
#> tzcode source: internal
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets methods base
#>
#> other attached packages:
#> [1] testthat_3.3.1 tidyr_1.3.2 ggplot2_4.0.1 dplyr_1.2.1
#> [5] nhschooldata_0.3.4
#>
#> loaded via a namespace (and not attached):
#> [1] gtable_0.3.6 jsonlite_2.0.0 compiler_4.5.0 brio_1.1.5
#> [5] tidyselect_1.2.1 jquerylib_0.1.4 systemfonts_1.3.1 scales_1.4.0
#> [9] textshaping_1.0.4 yaml_2.3.12 fastmap_1.2.0 R6_2.6.1
#> [13] labeling_0.4.3 generics_0.1.4 knitr_1.51 htmlwidgets_1.6.4
#> [17] tibble_3.3.1 desc_1.4.3 bslib_0.9.0 pillar_1.11.1
#> [21] RColorBrewer_1.1-3 rlang_1.3.0 utf8_1.2.6 cachem_1.1.0
#> [25] xfun_0.55 S7_0.2.1 fs_2.1.0 sass_0.4.10
#> [29] otel_0.2.0 cli_3.6.6 withr_3.0.3 pkgdown_2.2.0
#> [33] magrittr_2.0.5 digest_0.6.39 grid_4.5.0 rappdirs_0.3.4
#> [37] lifecycle_1.0.5 vctrs_0.7.3 evaluate_1.0.5 glue_1.8.1
#> [41] farver_2.1.2 codetools_0.2-20 ragg_1.5.0 purrr_1.2.2
#> [45] rmarkdown_2.30 tools_4.5.0 pkgconfig_2.0.3 htmltools_0.5.9