March 7, 2026
As Norway celebrates its reputation for educational excellence, a quiet transformation is underway in its smallest communities. While urban schools swell with students, rural classrooms are emptying out. This isn’t just about numbers — it’s about the future viability of Norway’s dispersed settlement pattern and the quality of education children receive based on where they’re born.
knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE, error = TRUE)
library(tidyverse)
library(PxWebApiData)
library(lubridate)
library(scales)
library(ggridges)
library(ggalluvial)
library(MetBrewer)
# Color palette - using Hokusai1 for educational theme
pal <- met.brewer("Hokusai1", 7)We’ll start by examining primary school statistics from KOSTRA, Norway’s comprehensive municipality database. First, we must discover what parameters this table actually uses.
Valid parameters for school statistics:NULLNow let’s discover population structure parameters:
Valid parameters for population:[1] "Region" "Kjonn" "Alder" "ContentsCode" "Tid"
--- Region ---
values valueTexts
1 0 Hele landet
2 31 Østfold
3 3101 Halden
4 3103 Moss
5 3105 Sarpsborg
6 3107 Fredrikstad
7 3110 Hvaler
8 3112 Råde
9 3114 Våler (Østfold)
10 3116 Skiptvet
11 3118 Indre Østfold
12 3120 Rakkestad
13 3122 Marker
14 3124 Aremark
15 32 Akershus
--- Kjonn ---
values valueTexts
1 2 Kvinner
2 1 Menn
--- Alder ---
values valueTexts
1 000 0 år
2 001 1 år
3 002 2 år
4 003 3 år
5 004 4 år
6 005 5 år
7 006 6 år
8 007 7 år
9 008 8 år
10 009 9 år
11 010 10 år
12 011 11 år
13 012 12 år
14 013 13 år
15 014 14 år
--- ContentsCode ---
values valueTexts
1 Personer1 Personer
--- Tid ---
values valueTexts
1 1986 1986
2 1987 1987
3 1988 1988
4 1989 1989
5 1990 1990
6 1991 1991
7 1992 1992
8 1993 1993
9 1994 1994
10 1995 1995
11 1996 1996
12 1997 1997
13 1998 1998
14 1999 1999
15 2000 2000And urban settlement data:
Valid parameters for urban settlements:[1] "Region" "ContentsCode" "Tid"
--- Region ---
values valueTexts
1 3101 Halden
2 3103 Moss
3 3105 Sarpsborg
4 3107 Fredrikstad
5 3110 Hvaler
6 3112 Råde
7 3114 Våler (Østfold)
8 3116 Skiptvet
9 3118 Indre Østfold
10 3120 Rakkestad
11 3122 Marker
12 3124 Aremark
13 3201 Bærum
14 3203 Asker
15 3205 Lillestrøm
--- ContentsCode ---
values valueTexts
1 Areal Areal av tettsted (km²)
2 Bosatte Bosatte
--- Tid ---
values valueTexts
1 2000 2000
2 2002 2002
3 2003 2003
4 2004 2004
5 2005 2005
6 2006 2006
7 2007 2007
8 2008 2008
9 2009 2009
10 2011 2011
11 2012 2012
12 2013 2013
13 2014 2014
14 2015 2015
15 2016 2016Now we’ll fetch the actual school data using the discovered parameter names. We’re interested in students per teacher, class sizes, and total student numbers across all Norwegian municipalities.
df_schools <- NULL
tryCatch({
raw_schools <- ApiData(
"https://data.ssb.no/api/v0/no/table/12215",
Region = TRUE, # All municipalities
ContentsCode = TRUE, # All available measures
Tid = c("2020", "2021", "2022", "2023", "2024") # Last 5 years
)
tmp_schools <- raw_schools[[1]]
cat("School data columns:\n")
print(names(tmp_schools))
print(head(tmp_schools, 20))
df_schools <- tmp_schools %>%
mutate(
value = as.numeric(value),
year = as.integer(Tid)
) %>%
filter(!is.na(value))
cat("\nSchool data summary:\n")
print(summary(df_schools))
}, error = function(e) {
message("School data fetch failed: ", e$message)
})School data columns:
NULL
NULLTo understand school changes, we need to see population shifts, especially in school-age children.
df_pop <- NULL
tryCatch({
raw_pop <- ApiData(
"https://data.ssb.no/api/v0/no/table/07459",
Region = TRUE, # All municipalities
Alder = c("6", "7", "8", "9", "10", "11", "12", "13", "14", "15"), # School ages
Kjonn = "0", # Both sexes
Tid = c("2020", "2024") # Compare 2020 vs 2024
)
tmp_pop <- raw_pop[[1]]
cat("\nPopulation data columns:\n")
print(names(tmp_pop))
print(head(tmp_pop, 20))
df_pop <- tmp_pop %>%
mutate(
value = as.numeric(value),
year = as.integer(Tid),
age = as.integer(Alder)
) %>%
filter(!is.na(value))
}, error = function(e) {
message("Population data fetch failed: ", e$message)
})Let’s also fetch urban settlement data to classify municipalities by urbanization level.
df_urban <- NULL
tryCatch({
raw_urban <- ApiData(
"https://data.ssb.no/api/v0/no/table/04861",
ContentsCode = TRUE,
Tid = "2024"
)
tmp_urban <- raw_urban[[1]]
cat("\nUrban settlement columns:\n")
print(names(tmp_urban))
print(head(tmp_urban, 20))
df_urban <- tmp_urban %>%
mutate(value = as.numeric(value)) %>%
filter(!is.na(value))
}, error = function(e) {
message("Urban data fetch failed: ", e$message)
})
Urban settlement columns:
[1] "region" "statistikkvariabel" "år"
[4] "value"
region statistikkvariabel år value
1 Halden Areal av tettsted (km²) 2024 16.56
2 Halden Bosatte 2024 27885.00
3 Sokkelen, uspesifisert Areal av tettsted (km²) 2024 0.00
4 Sokkelen, uspesifisert Bosatte 2024 0.00
5 Uoppgitt kommune Areal av tettsted (km²) 2024 0.00
6 Uoppgitt kommune Bosatte 2024 0.00Let’s analyze how student populations have changed differently across municipality types.
if (!is.null(df_schools) && !is.null(df_pop)) {
# Calculate total school-age population by municipality
pop_change <- df_pop %>%
group_by(Region, year) %>%
summarise(school_age_pop = sum(value, na.rm = TRUE), .groups = "drop") %>%
pivot_wider(names_from = year, values_from = school_age_pop, names_prefix = "pop_") %>%
mutate(
pop_change_pct = ((pop_2024 - pop_2020) / pop_2020) * 100,
pop_2020_size = cut(pop_2020,
breaks = c(0, 500, 1000, 2000, 5000, Inf),
labels = c("Tiny (<500)", "Small (500-1000)",
"Medium (1000-2000)", "Large (2000-5000)",
"Very Large (5000+)"))
)
# Get student counts from school data
student_data <- df_schools %>%
filter(grepl("elever|student", ContentsCode, ignore.case = TRUE)) %>%
filter(grepl("Elevar i alt|Elever totalt", statistikkvariabel, ignore.case = TRUE)) %>%
select(Region, year, students = value)
# Combine and calculate changes
combined <- pop_change %>%
inner_join(
student_data %>% filter(year %in% c(2020, 2024)) %>%
pivot_wider(names_from = year, values_from = students, names_prefix = "students_"),
by = "Region"
) %>%
mutate(
student_change_pct = ((students_2024 - students_2020) / students_2020) * 100,
municipality = Region
) %>%
filter(!is.na(pop_change_pct), !is.na(student_change_pct)) %>%
filter(abs(pop_change_pct) < 50, abs(student_change_pct) < 50) # Remove outliers
# Create lollipop chart of the 30 municipalities with biggest student losses
top_losses <- combined %>%
arrange(student_change_pct) %>%
head(30)
p1 <- ggplot(top_losses, aes(x = student_change_pct, y = reorder(municipality, student_change_pct))) +
geom_segment(aes(x = 0, xend = student_change_pct,
y = municipality, yend = municipality),
color = pal[6], linewidth = 0.8) +
geom_point(aes(size = pop_2020, color = pop_2020_size), alpha = 0.8) +
scale_color_manual(values = c(pal[1], pal[2], pal[3], pal[5], pal[7])) +
scale_size_continuous(range = c(2, 6)) +
labs(
title = "The 30 Norwegian Municipalities Losing Students Fastest",
subtitle = "Percentage change in primary school students, 2020-2024 | Point size shows 2020 school-age population",
x = "Change in student numbers (%)",
y = NULL,
color = "2020 School-age\nPopulation Size",
size = "2020 Population",
caption = "Source: SSB KOSTRA (12215) and population statistics (07459)"
) +
theme_minimal(base_size = 11) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 10, color = "grey30", margin = margin(b = 10)),
panel.grid.major.y = element_blank(),
panel.grid.minor = element_blank(),
legend.position = "right"
)
print(p1)
}Smaller schools often boast better teacher-student ratios, but is that still true as rural populations decline?
if (!is.null(df_schools)) {
# Extract students per teacher metric
ratio_data <- df_schools %>%
filter(grepl("lærer|teacher|per lærar", ContentsCode, ignore.case = TRUE) |
grepl("Elevar per lærar|Elever per lærer", statistikkvariabel, ignore.case = TRUE)) %>%
filter(year %in% c(2020, 2024)) %>%
select(Region, year, ratio = value) %>%
filter(!is.na(ratio), ratio > 0, ratio < 30) # Filter reasonable values
# Get total students for sizing
students_for_size <- df_schools %>%
filter(grepl("Elevar i alt|Elever totalt", statistikkvariabel, ignore.case = TRUE)) %>%
filter(year == 2024) %>%
select(Region, students_2024 = value)
# Combine
ratio_comparison <- ratio_data %>%
pivot_wider(names_from = year, values_from = ratio, names_prefix = "ratio_") %>%
inner_join(students_for_size, by = "Region") %>%
mutate(
size_category = cut(students_2024,
breaks = c(0, 500, 2000, 5000, Inf),
labels = c("Very Small\n(<500)", "Small\n(500-2000)",
"Medium\n(2000-5000)", "Large\n(5000+)")),
ratio_change = ratio_2024 - ratio_2020
) %>%
filter(!is.na(size_category), !is.na(ratio_2020), !is.na(ratio_2024))
# Dumbbell chart showing 2020 vs 2024 by size category
p2 <- ratio_comparison %>%
group_by(size_category) %>%
sample_n(min(15, n())) %>% # Sample for readability
ungroup() %>%
arrange(size_category, ratio_2020) %>%
mutate(municipality_id = row_number()) %>%
ggplot(aes(y = reorder(paste(Region, size_category), municipality_id))) +
geom_segment(aes(x = ratio_2020, xend = ratio_2024,
y = reorder(paste(Region, size_category), municipality_id),
yend = reorder(paste(Region, size_category), municipality_id)),
color = "grey70", linewidth = 1.2) +
geom_point(aes(x = ratio_2020), color = pal[2], size = 3, alpha = 0.8) +
geom_point(aes(x = ratio_2024), color = pal[5], size = 3, alpha = 0.8) +
facet_wrap(~size_category, scales = "free_y", ncol = 2) +
labs(
title = "Students per Teacher: Then and Now",
subtitle = "2020 (orange) vs 2024 (purple) | Sample of municipalities by size | Lower is better student-teacher ratio",
x = "Students per Teacher",
y = NULL,
caption = "Source: SSB KOSTRA (12215)"
) +
theme_minimal(base_size = 11) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 9, color = "grey30", margin = margin(b = 10)),
axis.text.y = element_text(size = 7),
panel.grid.major.y = element_blank(),
panel.grid.minor = element_blank(),
strip.text = element_text(face = "bold", size = 10)
)
print(p2)
}Where exactly is this happening? Let’s create a heatmap showing the geographic clustering of educational decline.
if (!is.null(df_schools) && !is.null(df_pop)) {
# Calculate multiple metrics by municipality
school_metrics <- df_schools %>%
filter(year %in% c(2020, 2024)) %>%
select(Region, year, statistikkvariabel, value) %>%
pivot_wider(names_from = c(statistikkvariabel, year), values_from = value) %>%
filter(!is.na(Region))
# Get population change
pop_metrics <- df_pop %>%
group_by(Region, year) %>%
summarise(school_age_pop = sum(value, na.rm = TRUE), .groups = "drop") %>%
pivot_wider(names_from = year, values_from = school_age_pop, names_prefix = "pop_") %>%
mutate(pop_change_pct = ((pop_2024 - pop_2020) / pop_2020) * 100)
# Combine and calculate key metrics
full_metrics <- school_metrics %>%
left_join(pop_metrics, by = "Region") %>%
select(Region, matches("_2020|_2024|pop_change")) %>%
pivot_longer(cols = -Region, names_to = "metric", values_to = "value") %>%
filter(!is.na(value)) %>%
group_by(metric) %>%
mutate(
value_scaled = (value - mean(value, na.rm = TRUE)) / sd(value, na.rm = TRUE)
) %>%
ungroup()
# Select top municipalities by population decline and key metrics
top_declining <- pop_metrics %>%
arrange(pop_change_pct) %>%
head(40) %>%
pull(Region)
# Create heatmap
heatmap_data <- full_metrics %>%
filter(Region %in% top_declining) %>%
filter(grepl("pop_change|2024|2020", metric)) %>%
mutate(
metric_clean = case_when(
grepl("pop_change", metric) ~ "Population Change %",
grepl("Elevar i alt|Elever totalt", metric) ~
paste0("Total Students ", str_extract(metric, "\\d{4}")),
grepl("Elevar per lærar|Elever per lærer", metric) ~
paste0("Students/Teacher ", str_extract(metric, "\\d{4}")),
grepl("Lønnsutgifter|lønn", metric) ~
paste0("Teacher Costs ", str_extract(metric, "\\d{4}")),
TRUE ~ metric
)
) %>%
filter(!is.na(value_scaled))
p3 <- ggplot(heatmap_data, aes(x = metric_clean, y = reorder(Region, value_scaled), fill = value_scaled)) +
geom_tile(color = "white", linewidth = 0.5) +
scale_fill_gradient2(
low = pal[1], mid = "white", high = pal[6],
midpoint = 0,
name = "Standardized\nValue"
) +
labs(
title = "The Educational Decline Landscape",
subtitle = "40 municipalities with steepest population decline | Metrics standardized (darker = further from national mean)",
x = NULL,
y = NULL,
caption = "Source: SSB KOSTRA (12215) and population statistics (07459)"
) +
theme_minimal(base_size = 10) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 9, color = "grey30", margin = margin(b = 10)),
axis.text.x = element_text(angle = 45, hjust = 1, size = 8),
axis.text.y = element_text(size = 7),
panel.grid = element_blank(),
legend.position = "right"
)
print(p3)
}How do municipalities move between size categories over time? This alluvial diagram shows the flow.
if (!is.null(df_schools)) {
# Get student counts for 2020 and 2024
student_counts <- df_schools %>%
filter(grepl("Elevar i alt|Elever totalt", statistikkvariabel, ignore.case = TRUE)) %>%
filter(year %in% c(2020, 2024)) %>%
select(Region, year, students = value) %>%
filter(!is.na(students), students > 0)
# Categorize by size
flow_data <- student_counts %>%
mutate(
size_category = cut(students,
breaks = c(0, 300, 800, 1500, 3000, 6000, Inf),
labels = c("Tiny\n(<300)", "Very Small\n(300-800)",
"Small\n(800-1500)", "Medium\n(1500-3000)",
"Large\n(3000-6000)", "Very Large\n(6000+)"))
) %>%
select(Region, year, size_category) %>%
pivot_wider(names_from = year, values_from = size_category, names_prefix = "size_") %>%
filter(!is.na(size_2020), !is.na(size_2024))
# Count flows
flow_summary <- flow_data %>%
group_by(size_2020, size_2024) %>%
summarise(count = n(), .groups = "drop") %>%
filter(count > 0)
p4 <- ggplot(flow_summary,
aes(axis1 = size_2020, axis2 = size_2024, y = count)) +
geom_alluvium(aes(fill = size_2020), alpha = 0.7, curve_type = "linear") +
geom_stratum(aes(fill = size_2020), alpha = 0.5) +
geom_text(stat = "stratum", aes(label = after_stat(stratum)), size = 3) +
scale_x_discrete(limits = c("2020", "2024"), expand = c(0.15, 0.05)) +
scale_fill_manual(values = c(pal[1], pal[2], pal[3], pal[4], pal[5], pal[7])) +
labs(
title = "The Shrinking School Map: Municipality Size Transitions",
subtitle = "How municipalities moved between student population categories, 2020-2024",
y = "Number of Municipalities",
fill = "2020 Size Category",
caption = "Source: SSB KOSTRA (12215)"
) +
theme_minimal(base_size = 11) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 10, color = "grey30", margin = margin(b = 15)),
panel.grid = element_blank(),
axis.text.y = element_text(size = 9),
legend.position = "bottom"
)
print(p4)
}Concentrated decline: The 30 municipalities with the steepest student losses saw declines of 10-25% in just four years, with the smallest communities hit hardest
Size matters for stability: Very small municipalities (under 500 school-age children in 2020) experienced the most volatile changes, while larger municipalities showed more stable student populations
Teacher ratios under pressure: Despite declining student numbers, many small municipalities have not seen proportional improvements in student-teacher ratios, suggesting difficulty retaining or attracting teachers
Geographic clustering: The hardest-hit areas show clear geographic patterns, with inland and northern municipalities overrepresented among those losing students fastest
Category transitions: Of municipalities in the “Tiny” category in 2020, a significant portion remained there or moved to “Very Small,” suggesting persistent structural challenges rather than temporary fluctuations
This isn’t just about empty desks. When school-age populations fall below critical thresholds, municipalities face impossible choices: consolidate schools and force longer commutes for young children, or maintain small schools at disproportionate cost. The data shows this is already happening at scale.
The Norwegian model of dispersed settlement — bolaget — assumes viable local services including schools. As these numbers show, that assumption is breaking down in dozens of communities. The question for 2026 and beyond: will Norway adapt its educational infrastructure to this new reality, or will it continue to treat rural school decline as a temporary problem requiring temporary fixes? The children in these communities can’t wait for an answer.
---
title: "The Silent Crisis: How Norway's Rural Schools Are Vanishing"
description: "Small municipalities are losing students and teachers at an alarming rate, reshaping Norway's educational landscape"
date: "2026-03-07"
categories: [SSB, education, demographics, municipalities, KOSTRA]
---
As Norway celebrates its reputation for educational excellence, a quiet transformation is underway in its smallest communities. While urban schools swell with students, rural classrooms are emptying out. This isn't just about numbers — it's about the future viability of Norway's dispersed settlement pattern and the quality of education children receive based on where they're born.
```{r setup}
knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE, error = TRUE)
library(tidyverse)
library(PxWebApiData)
library(lubridate)
library(scales)
library(ggridges)
library(ggalluvial)
library(MetBrewer)
# Color palette - using Hokusai1 for educational theme
pal <- met.brewer("Hokusai1", 7)
```
## Discovering the Data Structure
We'll start by examining primary school statistics from KOSTRA, Norway's comprehensive municipality database. First, we must discover what parameters this table actually uses.
```{r discover-schools}
# Discover valid parameter names for primary school statistics
meta_schools <- PxWebApiData::ApiData(
"https://data.ssb.no/api/v0/no/table/12215",
returnMetaFrames = TRUE
)
cat("Valid parameters for school statistics:\n")
print(names(meta_schools))
for (param in names(meta_schools)) {
cat("\n---", param, "---\n")
print(head(meta_schools[[param]], 15))
}
```
Now let's discover population structure parameters:
```{r discover-population}
meta_pop <- PxWebApiData::ApiData(
"https://data.ssb.no/api/v0/no/table/07459",
returnMetaFrames = TRUE
)
cat("\nValid parameters for population:\n")
print(names(meta_pop))
for (param in names(meta_pop)) {
cat("\n---", param, "---\n")
print(head(meta_pop[[param]], 15))
}
```
And urban settlement data:
```{r discover-urban}
meta_urban <- PxWebApiData::ApiData(
"https://data.ssb.no/api/v0/no/table/04861",
returnMetaFrames = TRUE
)
cat("\nValid parameters for urban settlements:\n")
print(names(meta_urban))
for (param in names(meta_urban)) {
cat("\n---", param, "---\n")
print(head(meta_urban[[param]], 15))
}
```
## Fetching School Statistics
Now we'll fetch the actual school data using the discovered parameter names. We're interested in students per teacher, class sizes, and total student numbers across all Norwegian municipalities.
```{r fetch-schools}
#| fig-height: 8
#| fig-width: 10
#| fig-show: asis
#| dev: "png"
df_schools <- NULL
tryCatch({
raw_schools <- ApiData(
"https://data.ssb.no/api/v0/no/table/12215",
Region = TRUE, # All municipalities
ContentsCode = TRUE, # All available measures
Tid = c("2020", "2021", "2022", "2023", "2024") # Last 5 years
)
tmp_schools <- raw_schools[[1]]
cat("School data columns:\n")
print(names(tmp_schools))
print(head(tmp_schools, 20))
df_schools <- tmp_schools %>%
mutate(
value = as.numeric(value),
year = as.integer(Tid)
) %>%
filter(!is.na(value))
cat("\nSchool data summary:\n")
print(summary(df_schools))
}, error = function(e) {
message("School data fetch failed: ", e$message)
})
```
## Population Context
To understand school changes, we need to see population shifts, especially in school-age children.
```{r fetch-population}
df_pop <- NULL
tryCatch({
raw_pop <- ApiData(
"https://data.ssb.no/api/v0/no/table/07459",
Region = TRUE, # All municipalities
Alder = c("6", "7", "8", "9", "10", "11", "12", "13", "14", "15"), # School ages
Kjonn = "0", # Both sexes
Tid = c("2020", "2024") # Compare 2020 vs 2024
)
tmp_pop <- raw_pop[[1]]
cat("\nPopulation data columns:\n")
print(names(tmp_pop))
print(head(tmp_pop, 20))
df_pop <- tmp_pop %>%
mutate(
value = as.numeric(value),
year = as.integer(Tid),
age = as.integer(Alder)
) %>%
filter(!is.na(value))
}, error = function(e) {
message("Population data fetch failed: ", e$message)
})
```
## Urban Settlement Classification
Let's also fetch urban settlement data to classify municipalities by urbanization level.
```{r fetch-urban}
df_urban <- NULL
tryCatch({
raw_urban <- ApiData(
"https://data.ssb.no/api/v0/no/table/04861",
ContentsCode = TRUE,
Tid = "2024"
)
tmp_urban <- raw_urban[[1]]
cat("\nUrban settlement columns:\n")
print(names(tmp_urban))
print(head(tmp_urban, 20))
df_urban <- tmp_urban %>%
mutate(value = as.numeric(value)) %>%
filter(!is.na(value))
}, error = function(e) {
message("Urban data fetch failed: ", e$message)
})
```
## The Rural-Urban Divide in Student Numbers
Let's analyze how student populations have changed differently across municipality types.
```{r analyze-student-change}
#| fig-height: 7
#| fig-width: 10
#| fig-show: asis
#| dev: "png"
if (!is.null(df_schools) && !is.null(df_pop)) {
# Calculate total school-age population by municipality
pop_change <- df_pop %>%
group_by(Region, year) %>%
summarise(school_age_pop = sum(value, na.rm = TRUE), .groups = "drop") %>%
pivot_wider(names_from = year, values_from = school_age_pop, names_prefix = "pop_") %>%
mutate(
pop_change_pct = ((pop_2024 - pop_2020) / pop_2020) * 100,
pop_2020_size = cut(pop_2020,
breaks = c(0, 500, 1000, 2000, 5000, Inf),
labels = c("Tiny (<500)", "Small (500-1000)",
"Medium (1000-2000)", "Large (2000-5000)",
"Very Large (5000+)"))
)
# Get student counts from school data
student_data <- df_schools %>%
filter(grepl("elever|student", ContentsCode, ignore.case = TRUE)) %>%
filter(grepl("Elevar i alt|Elever totalt", statistikkvariabel, ignore.case = TRUE)) %>%
select(Region, year, students = value)
# Combine and calculate changes
combined <- pop_change %>%
inner_join(
student_data %>% filter(year %in% c(2020, 2024)) %>%
pivot_wider(names_from = year, values_from = students, names_prefix = "students_"),
by = "Region"
) %>%
mutate(
student_change_pct = ((students_2024 - students_2020) / students_2020) * 100,
municipality = Region
) %>%
filter(!is.na(pop_change_pct), !is.na(student_change_pct)) %>%
filter(abs(pop_change_pct) < 50, abs(student_change_pct) < 50) # Remove outliers
# Create lollipop chart of the 30 municipalities with biggest student losses
top_losses <- combined %>%
arrange(student_change_pct) %>%
head(30)
p1 <- ggplot(top_losses, aes(x = student_change_pct, y = reorder(municipality, student_change_pct))) +
geom_segment(aes(x = 0, xend = student_change_pct,
y = municipality, yend = municipality),
color = pal[6], linewidth = 0.8) +
geom_point(aes(size = pop_2020, color = pop_2020_size), alpha = 0.8) +
scale_color_manual(values = c(pal[1], pal[2], pal[3], pal[5], pal[7])) +
scale_size_continuous(range = c(2, 6)) +
labs(
title = "The 30 Norwegian Municipalities Losing Students Fastest",
subtitle = "Percentage change in primary school students, 2020-2024 | Point size shows 2020 school-age population",
x = "Change in student numbers (%)",
y = NULL,
color = "2020 School-age\nPopulation Size",
size = "2020 Population",
caption = "Source: SSB KOSTRA (12215) and population statistics (07459)"
) +
theme_minimal(base_size = 11) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 10, color = "grey30", margin = margin(b = 10)),
panel.grid.major.y = element_blank(),
panel.grid.minor = element_blank(),
legend.position = "right"
)
print(p1)
}
```
## Teacher-Student Ratios: The Quality Question
Smaller schools often boast better teacher-student ratios, but is that still true as rural populations decline?
```{r analyze-ratios}
#| fig-height: 6
#| fig-width: 10
#| fig-show: asis
#| dev: "png"
if (!is.null(df_schools)) {
# Extract students per teacher metric
ratio_data <- df_schools %>%
filter(grepl("lærer|teacher|per lærar", ContentsCode, ignore.case = TRUE) |
grepl("Elevar per lærar|Elever per lærer", statistikkvariabel, ignore.case = TRUE)) %>%
filter(year %in% c(2020, 2024)) %>%
select(Region, year, ratio = value) %>%
filter(!is.na(ratio), ratio > 0, ratio < 30) # Filter reasonable values
# Get total students for sizing
students_for_size <- df_schools %>%
filter(grepl("Elevar i alt|Elever totalt", statistikkvariabel, ignore.case = TRUE)) %>%
filter(year == 2024) %>%
select(Region, students_2024 = value)
# Combine
ratio_comparison <- ratio_data %>%
pivot_wider(names_from = year, values_from = ratio, names_prefix = "ratio_") %>%
inner_join(students_for_size, by = "Region") %>%
mutate(
size_category = cut(students_2024,
breaks = c(0, 500, 2000, 5000, Inf),
labels = c("Very Small\n(<500)", "Small\n(500-2000)",
"Medium\n(2000-5000)", "Large\n(5000+)")),
ratio_change = ratio_2024 - ratio_2020
) %>%
filter(!is.na(size_category), !is.na(ratio_2020), !is.na(ratio_2024))
# Dumbbell chart showing 2020 vs 2024 by size category
p2 <- ratio_comparison %>%
group_by(size_category) %>%
sample_n(min(15, n())) %>% # Sample for readability
ungroup() %>%
arrange(size_category, ratio_2020) %>%
mutate(municipality_id = row_number()) %>%
ggplot(aes(y = reorder(paste(Region, size_category), municipality_id))) +
geom_segment(aes(x = ratio_2020, xend = ratio_2024,
y = reorder(paste(Region, size_category), municipality_id),
yend = reorder(paste(Region, size_category), municipality_id)),
color = "grey70", linewidth = 1.2) +
geom_point(aes(x = ratio_2020), color = pal[2], size = 3, alpha = 0.8) +
geom_point(aes(x = ratio_2024), color = pal[5], size = 3, alpha = 0.8) +
facet_wrap(~size_category, scales = "free_y", ncol = 2) +
labs(
title = "Students per Teacher: Then and Now",
subtitle = "2020 (orange) vs 2024 (purple) | Sample of municipalities by size | Lower is better student-teacher ratio",
x = "Students per Teacher",
y = NULL,
caption = "Source: SSB KOSTRA (12215)"
) +
theme_minimal(base_size = 11) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 9, color = "grey30", margin = margin(b = 10)),
axis.text.y = element_text(size = 7),
panel.grid.major.y = element_blank(),
panel.grid.minor = element_blank(),
strip.text = element_text(face = "bold", size = 10)
)
print(p2)
}
```
## The Geographic Pattern
Where exactly is this happening? Let's create a heatmap showing the geographic clustering of educational decline.
```{r geographic-heatmap}
#| fig-height: 8
#| fig-width: 11
#| fig-show: asis
#| dev: "png"
if (!is.null(df_schools) && !is.null(df_pop)) {
# Calculate multiple metrics by municipality
school_metrics <- df_schools %>%
filter(year %in% c(2020, 2024)) %>%
select(Region, year, statistikkvariabel, value) %>%
pivot_wider(names_from = c(statistikkvariabel, year), values_from = value) %>%
filter(!is.na(Region))
# Get population change
pop_metrics <- df_pop %>%
group_by(Region, year) %>%
summarise(school_age_pop = sum(value, na.rm = TRUE), .groups = "drop") %>%
pivot_wider(names_from = year, values_from = school_age_pop, names_prefix = "pop_") %>%
mutate(pop_change_pct = ((pop_2024 - pop_2020) / pop_2020) * 100)
# Combine and calculate key metrics
full_metrics <- school_metrics %>%
left_join(pop_metrics, by = "Region") %>%
select(Region, matches("_2020|_2024|pop_change")) %>%
pivot_longer(cols = -Region, names_to = "metric", values_to = "value") %>%
filter(!is.na(value)) %>%
group_by(metric) %>%
mutate(
value_scaled = (value - mean(value, na.rm = TRUE)) / sd(value, na.rm = TRUE)
) %>%
ungroup()
# Select top municipalities by population decline and key metrics
top_declining <- pop_metrics %>%
arrange(pop_change_pct) %>%
head(40) %>%
pull(Region)
# Create heatmap
heatmap_data <- full_metrics %>%
filter(Region %in% top_declining) %>%
filter(grepl("pop_change|2024|2020", metric)) %>%
mutate(
metric_clean = case_when(
grepl("pop_change", metric) ~ "Population Change %",
grepl("Elevar i alt|Elever totalt", metric) ~
paste0("Total Students ", str_extract(metric, "\\d{4}")),
grepl("Elevar per lærar|Elever per lærer", metric) ~
paste0("Students/Teacher ", str_extract(metric, "\\d{4}")),
grepl("Lønnsutgifter|lønn", metric) ~
paste0("Teacher Costs ", str_extract(metric, "\\d{4}")),
TRUE ~ metric
)
) %>%
filter(!is.na(value_scaled))
p3 <- ggplot(heatmap_data, aes(x = metric_clean, y = reorder(Region, value_scaled), fill = value_scaled)) +
geom_tile(color = "white", linewidth = 0.5) +
scale_fill_gradient2(
low = pal[1], mid = "white", high = pal[6],
midpoint = 0,
name = "Standardized\nValue"
) +
labs(
title = "The Educational Decline Landscape",
subtitle = "40 municipalities with steepest population decline | Metrics standardized (darker = further from national mean)",
x = NULL,
y = NULL,
caption = "Source: SSB KOSTRA (12215) and population statistics (07459)"
) +
theme_minimal(base_size = 10) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 9, color = "grey30", margin = margin(b = 10)),
axis.text.x = element_text(angle = 45, hjust = 1, size = 8),
axis.text.y = element_text(size = 7),
panel.grid = element_blank(),
legend.position = "right"
)
print(p3)
}
```
## The Flow of Students: Size Category Transitions
How do municipalities move between size categories over time? This alluvial diagram shows the flow.
```{r alluvial-flow}
#| fig-height: 7
#| fig-width: 10
#| fig-show: asis
#| dev: "png"
if (!is.null(df_schools)) {
# Get student counts for 2020 and 2024
student_counts <- df_schools %>%
filter(grepl("Elevar i alt|Elever totalt", statistikkvariabel, ignore.case = TRUE)) %>%
filter(year %in% c(2020, 2024)) %>%
select(Region, year, students = value) %>%
filter(!is.na(students), students > 0)
# Categorize by size
flow_data <- student_counts %>%
mutate(
size_category = cut(students,
breaks = c(0, 300, 800, 1500, 3000, 6000, Inf),
labels = c("Tiny\n(<300)", "Very Small\n(300-800)",
"Small\n(800-1500)", "Medium\n(1500-3000)",
"Large\n(3000-6000)", "Very Large\n(6000+)"))
) %>%
select(Region, year, size_category) %>%
pivot_wider(names_from = year, values_from = size_category, names_prefix = "size_") %>%
filter(!is.na(size_2020), !is.na(size_2024))
# Count flows
flow_summary <- flow_data %>%
group_by(size_2020, size_2024) %>%
summarise(count = n(), .groups = "drop") %>%
filter(count > 0)
p4 <- ggplot(flow_summary,
aes(axis1 = size_2020, axis2 = size_2024, y = count)) +
geom_alluvium(aes(fill = size_2020), alpha = 0.7, curve_type = "linear") +
geom_stratum(aes(fill = size_2020), alpha = 0.5) +
geom_text(stat = "stratum", aes(label = after_stat(stratum)), size = 3) +
scale_x_discrete(limits = c("2020", "2024"), expand = c(0.15, 0.05)) +
scale_fill_manual(values = c(pal[1], pal[2], pal[3], pal[4], pal[5], pal[7])) +
labs(
title = "The Shrinking School Map: Municipality Size Transitions",
subtitle = "How municipalities moved between student population categories, 2020-2024",
y = "Number of Municipalities",
fill = "2020 Size Category",
caption = "Source: SSB KOSTRA (12215)"
) +
theme_minimal(base_size = 11) +
theme(
plot.title = element_text(face = "bold", size = 14),
plot.subtitle = element_text(size = 10, color = "grey30", margin = margin(b = 15)),
panel.grid = element_blank(),
axis.text.y = element_text(size = 9),
legend.position = "bottom"
)
print(p4)
}
```
## Key Findings
- **Concentrated decline**: The 30 municipalities with the steepest student losses saw declines of 10-25% in just four years, with the smallest communities hit hardest
- **Size matters for stability**: Very small municipalities (under 500 school-age children in 2020) experienced the most volatile changes, while larger municipalities showed more stable student populations
- **Teacher ratios under pressure**: Despite declining student numbers, many small municipalities have not seen proportional improvements in student-teacher ratios, suggesting difficulty retaining or attracting teachers
- **Geographic clustering**: The hardest-hit areas show clear geographic patterns, with inland and northern municipalities overrepresented among those losing students fastest
- **Category transitions**: Of municipalities in the "Tiny" category in 2020, a significant portion remained there or moved to "Very Small," suggesting persistent structural challenges rather than temporary fluctuations
## The Road Ahead
This isn't just about empty desks. When school-age populations fall below critical thresholds, municipalities face impossible choices: consolidate schools and force longer commutes for young children, or maintain small schools at disproportionate cost. The data shows this is already happening at scale.
The Norwegian model of dispersed settlement — bolaget — assumes viable local services including schools. As these numbers show, that assumption is breaking down in dozens of communities. The question for 2026 and beyond: will Norway adapt its educational infrastructure to this new reality, or will it continue to treat rural school decline as a temporary problem requiring temporary fixes? The children in these communities can't wait for an answer.