library(data.table)
library(ggplot2)
library(ggiraph)
library(readxl)IMF Renewable Energy Statistics: World Power, Countries, Groups and Regions
Reproducible analysis with data.table, ggplot2 and ggiraph
1 Purpose and conventions
This report analyses the IMF/IRENA Renewable Energy dataset (IMF_Renewable_Energy.xlsx). It covers installed capacity, electricity generation and public investment for 247 labelled entities over 2000 to 2023. The questions are:
- How big is the world renewable fleet, and how fast is it growing?
- Which countries lead, and where does India stand against the global powers?
- How do groups (G20, G7, Advanced Economies, Emerging Market and Developing Economies) and regions compare?
- Which statistical results hold up once the data are reconciled?
Code conventions: data.table for all wrangling, ggplot2 plus ggiraph for interactive plots, no user-defined functions, no loading of any other package apart from readxl to read the workbook.
1.1 Interactive economics theme
The theme is a plain object, not a function. Every plot below adds it and is wrapped with girafe() using the shared hover and tooltip options.
theme_econ <- theme_minimal(base_size = 12) +
theme(
plot.title = element_text(face = "bold", size = 14, colour = "#1a1a2e"),
plot.subtitle = element_text(size = 10.5, colour = "#4a4a6a", margin = margin(b = 8)),
plot.caption = element_text(size = 8.5, colour = "#888888", hjust = 0),
axis.title = element_text(size = 10.5, colour = "#333355"),
axis.text = element_text(size = 9.5, colour = "#444466"),
panel.grid.major = element_line(colour = "#e8e8f0", linewidth = 0.4),
panel.grid.minor = element_blank(),
legend.position = "bottom",
legend.title = element_text(face = "bold"),
strip.text = element_text(face = "bold", colour = "#1a1a2e"),
plot.background = element_rect(fill = "#fafafe", colour = NA),
panel.background = element_rect(fill = "#fafafe", colour = NA)
)
ggi_opts <- list(
opts_hover(css = "stroke-width:2.5px;opacity:1;"),
opts_hover_inv(css = "opacity:0.35;"),
opts_tooltip(css = "background:#1a1a2e;color:#ffffff;padding:6px 9px;border-radius:5px;font-size:12px;"),
opts_toolbar(saveaspng = TRUE)
)
pal_src <- c(Solar = "#FFB300", Wind = "#42A5F5", Hydro = "#26C6DA", Bio = "#66BB6A", Fossil = "#EF5350")
src_cap <- "Source: IMF/IRENA Renewable Energy Statistics. Renewable = solar + wind + hydropower (excl. pumped storage) + bioenergy."2 Data preparation and inspection
2.1 Read and inspect the wide table
The workbook is wide: one column per year. The first header carries a byte-order mark, so it is renamed. Year headers arrive as 2000.0, 2001.0, and so on, so they are matched with a pattern instead of typed.
raw <- as.data.table(read_excel("~/imf/renewable_energy/IMF_Renewable_Energy.xlsx"))
setnames(raw, 1, "DATASET")
# Some entity names are double-encoded in the workbook (e.g. "Türkiye"). Re-read their bytes as UTF-8.
raw[, COUNTRY := fifelse(grepl("Ã", COUNTRY, fixed = TRUE),
{ y <- iconv(COUNTRY, "UTF-8", "latin1"); Encoding(y) <- "UTF-8"; y },
COUNTRY)]
ycols <- grep("^20", names(raw), value = TRUE)
raw[, .(rows = .N, entities = uniqueN(COUNTRY), sources = uniqueN(ENERGY_SOURCE), year_columns = length(ycols))] rows entities sources year_columns
<int> <int> <int> <int>
1: 3708 247 11 24
raw[, .N, by = INDICATOR] INDICATOR N
<char> <int>
1: Electricity Installed Capacity, Megawatt 1029
2: Electricity Generation, Gigawatt hours 1023
3: Public Investment in Renewable Energy as share of GDP 1656
2.2 Reshape wide to long with melt()
Why. Analysis needs YEAR as a variable. melt() turns 24 year columns into rows so that grouping, joins and plotting work on one value column.
How. id.vars keep the identifying columns, measure.vars are the year columns. Each original row becomes up to 24 rows of (entity, indicator, source, year, value).
long <- melt(raw, id.vars = c("COUNTRY", "INDICATOR", "ENERGY_SOURCE"),
measure.vars = ycols, variable.name = "YEAR", value.name = "VALUE")
long[, YEAR := as.integer(as.numeric(as.character(YEAR)))]
long[, VALUE := as.numeric(VALUE)]
long[, IND := fcase(grepl("Capacity", INDICATOR), "CAP",
grepl("Generation", INDICATOR), "GEN",
default = "INV")]
srcmap <- data.table(
ENERGY_SOURCE = c("Solar energy", "Wind energy", "Hydropower (excl. Pumped Storage)", "Bioenergy", "Fossil fuels"),
SRC = c("Solar", "Wind", "Hydro", "Bio", "Fossil"))
long[, SRC := ENERGY_SOURCE]
long[srcmap, SRC := i.SRC, on = "ENERGY_SOURCE"]
long[, .(cells = .N, missing = sum(is.na(VALUE)), missing_pct = round(mean(is.na(VALUE)) * 100, 1)), by = IND] IND cells missing missing_pct
<char> <int> <int> <num>
1: CAP 24696 3333 13.5
2: GEN 24552 4235 17.2
3: INV 39744 2043 5.1
2.3 Classify entities: country, region, sub-region, group
Why. The dataset mixes countries with aggregates (regions, sub-regions, G20, G7, income-style groups). Adding an aggregate to its own members double counts. Every world total, share or rank below depends on this classification.
How. Three name vectors list the aggregates, and fcase() labels every other entity a country. A small lookup join shortens long IMF names for plots.
regs <- c("Africa", "Americas", "Asia", "Europe", "Oceania")
subregs <- c("Central Asia", "Eastern Asia", "Eastern Europe", "Latin America and the Caribbean (LAC)",
"North America", "Northern Africa", "Northern Europe", "Oceania sub-regions",
"Southern Asia", "South-eastern Asia", "Southern Europe", "Sub-Saharan Africa (SSA)",
"Western Asia", "Western Europe")
grps <- c("G20", "G7", "Advanced Economies", "Emerging Market and Developing Economies")
long[, TYPE := fcase(COUNTRY %in% regs, "Region", COUNTRY %in% subregs, "Sub-region",
COUNTRY %in% grps, "Group", default = "Country")]
lab <- data.table(
COUNTRY = c("China, People's Republic of", "United States", "Korea, Republic of", "Russian Federation",
"Egypt, Arab Republic of", "Iran, Islamic Republic of", "Netherlands, The",
"Taiwan Province of China", "Emerging Market and Developing Economies", "Advanced Economies"),
LABEL = c("China", "USA", "South Korea", "Russia", "Egypt", "Iran", "Netherlands",
"Taiwan", "EMDE", "Advanced Econ."))
long[lab, LABEL := i.LABEL, on = "COUNTRY"]
long[is.na(LABEL), LABEL := sub(",.*$", "", COUNTRY)]
long[, .(entities = uniqueN(COUNTRY)), by = TYPE][order(-entities)] TYPE entities
<char> <int>
1: Country 224
2: Sub-region 14
3: Region 5
4: Group 4
2.4 Clean, de-duplicate and write the tidy file
n_before <- nrow(long)
long <- long[!is.na(VALUE)]
data.table(rows_before = n_before, rows_after = nrow(long), dropped_missing = n_before - nrow(long)) rows_before rows_after dropped_missing
<int> <int> <int>
1: 88992 79381 9611
# duplicate key check: any entity-indicator-source-year seen twice?
long[, .N, by = .(COUNTRY, IND, ENERGY_SOURCE, YEAR)][N > 1]Empty data.table (0 rows and 5 cols): COUNTRY,IND,ENERGY_SOURCE,YEAR,N
fwrite(long, "imf_re_long_clean.csv")
cap <- long[IND == "CAP"]
gen <- long[IND == "GEN"]
inv <- long[IND == "INV"]
re_src <- c("Solar energy", "Wind energy", "Hydropower (excl. Pumped Storage)", "Bioenergy")
long[, .(first_year = min(YEAR), last_year = max(YEAR), rows = .N), by = IND] IND first_year last_year rows
<char> <int> <int> <int>
1: CAP 2000 2023 21363
2: GEN 2000 2023 20317
3: INV 2000 2022 37701
3 Data reconciliation
Why. The dataset has no “World” row. A world total can be built from regions or from the Advanced plus Emerging groups. If both agree, and country rows also agree, the totals are trustworthy. A mismatch would mean double counting or missing entities.
How. For 2023 capacity, each source is summed three ways with data.table conditional sums inside j, and the percentage gap is computed.
rec <- cap[YEAR == 2023, .(
regions_GW = round(sum(VALUE[TYPE == "Region"]) / 1000, 1),
groups_GW = round(sum(VALUE[COUNTRY %in% c("Advanced Economies", "Emerging Market and Developing Economies")]) / 1000, 1),
countries_GW = round(sum(VALUE[TYPE == "Country"]) / 1000, 1)
), by = SRC]
rec[, `:=`(gap_regions_vs_groups_pct = round((regions_GW - groups_GW) / groups_GW * 100, 2),
gap_countries_vs_regions_pct = round((countries_GW - regions_GW) / regions_GW * 100, 2))]
rec SRC regions_GW groups_GW countries_GW gap_regions_vs_groups_pct
<char> <num> <num> <num> <num>
1: Fossil 4528.1 4528.1 4528.1 0
2: Hydro 1264.7 1264.7 1264.7 0
3: Wind 1017.4 1017.4 1017.4 0
4: Solar 1418.0 1418.0 1418.0 0
5: Bio 148.8 148.8 148.8 0
gap_countries_vs_regions_pct
<num>
1: 0
2: 0
3: 0
4: 0
5: 0
Why the naive sum is wrong. Summing every row for 2023 mixes countries, sub-regions, regions and groups. The next table measures the size of that error against the reconciled world total.
naive <- cap[YEAR == 2023 & SRC != "Fossil", .(all_rows_GW = round(sum(VALUE) / 1000, 0))]
true_world <- cap[YEAR == 2023 & TYPE == "Region" & SRC != "Fossil", .(world_GW = round(sum(VALUE) / 1000, 0))]
data.table(naive_sum_GW = naive$all_rows_GW, reconciled_world_GW = true_world$world_GW,
overstatement_x = round(naive$all_rows_GW / true_world$world_GW, 1)) naive_sum_GW reconciled_world_GW overstatement_x
<num> <num> <num>
1: 19778 3849 5.1
From here on, World = sum of the five regions, countries come only from TYPE == "Country", and aggregates are used only as labelled groups.
4 World power
4.1 World capacity by source
Why. The trajectory of the world fleet, and the renewable share of it, is the headline indicator of the energy transition.
How. Capacity is filtered to the five regions, summed by year and source, and cast wide to compute totals, shares and year-on-year growth. The compound annual growth rate is
\[\text{CAGR} = \left(\frac{V_T}{V_0}\right)^{1/T} - 1 .\]
world <- cap[TYPE == "Region", .(GW = sum(VALUE) / 1000), by = .(YEAR, SRC)]
world[, GROUP := fifelse(SRC == "Fossil", "Fossil", "Renewable")]
world_grp <- world[, .(GW = sum(GW)), by = .(YEAR, GROUP)]
world_w <- dcast(world_grp, YEAR ~ GROUP, value.var = "GW")
world_w[, `:=`(TOTAL = Renewable + Fossil, RE_SHARE = Renewable / (Renewable + Fossil) * 100)]
world_w[, `:=`(RE_YOY = (Renewable / shift(Renewable) - 1) * 100, FF_YOY = (Fossil / shift(Fossil) - 1) * 100)]
world_w[YEAR %in% c(2000, 2005, 2010, 2015, 2020, 2023), .(YEAR, Renewable = round(Renewable), Fossil = round(Fossil),
RE_SHARE = round(RE_SHARE, 1), RE_YOY = round(RE_YOY, 1), FF_YOY = round(FF_YOY, 1))]Key: <YEAR>
YEAR Renewable Fossil RE_SHARE RE_YOY FF_YOY
<int> <num> <num> <num> <num> <num>
1: 2000 744 2290 24.5 NA NA
2: 2005 892 2733 24.6 4.6 3.7
3: 2010 1215 3352 26.6 7.7 4.5
4: 2015 1840 3910 32.0 9.1 2.4
5: 2020 2805 4338 39.3 10.7 1.6
6: 2023 3849 4528 45.9 14.0 1.7
world_w[YEAR %in% c(2000, 2023), .(
RE_CAGR_pct = round(((Renewable[2] / Renewable[1])^(1 / 23) - 1) * 100, 2),
FF_CAGR_pct = round(((Fossil[2] / Fossil[1])^(1 / 23) - 1) * 100, 2),
RE_multiple = round(Renewable[2] / Renewable[1], 1))] RE_CAGR_pct FF_CAGR_pct RE_multiple
<num> <num> <num>
1: 7.41 3.01 5.2
world[YEAR == 2023, .(SRC, GW = round(GW, 1), SHARE_OF_TOTAL_pct = round(GW / sum(GW) * 100, 1))][order(-GW)] SRC GW SHARE_OF_TOTAL_pct
<char> <num> <num>
1: Fossil 4528.1 54.1
2: Solar 1418.0 16.9
3: Hydro 1264.7 15.1
4: Wind 1017.4 12.1
5: Bio 148.8 1.8
yrs <- c(2000, 2005, 2010, 2015, 2020, 2023)
w1 <- world[YEAR %in% yrs]
w1[, SRC := factor(SRC, levels = c("Fossil", "Hydro", "Bio", "Wind", "Solar"))]
w1[, tip := paste0(SRC, " ", YEAR, ": ", round(GW), " GW")]
tot1 <- w1[, .(GW = sum(GW)), by = YEAR]
p1 <- ggplot(w1, aes(x = factor(YEAR), y = GW, fill = SRC)) +
geom_col_interactive(aes(tooltip = tip, data_id = paste(SRC, YEAR)), width = 0.7) +
geom_text(data = w1[GW >= 100], aes(label = round(GW)), position = position_stack(vjust = 0.5), size = 3, colour = "#1a1a2e") +
geom_text(data = tot1, aes(x = factor(YEAR), y = GW, label = paste0(round(GW), " GW")),
inherit.aes = FALSE, vjust = -0.5, fontface = "bold", size = 3.4) +
annotate("text", x = 1.1, y = max(tot1$GW) * 0.93, hjust = 0, size = 3.4, colour = "#4a4a6a",
label = "World = sum of 5 regions (reconciled)") +
scale_fill_manual(values = pal_src, name = "Source") +
labs(title = "World installed electricity capacity by source",
subtitle = "Gigawatts; renewables grew faster than fossil capacity throughout",
x = "Year", y = "Installed capacity (GW)", caption = src_cap) +
theme_econ
girafe(ggobj = p1, width_svg = 9, height_svg = 5.5, options = ggi_opts)5 Countries and India versus the global powers
5.1 Country league table, 2023
Why. Ranking countries on renewable capacity identifies the global powers and sets the benchmark India is compared with.
How. Only TYPE == "Country" rows are used, so aggregates cannot appear. Renewable capacity is summed over the four renewable sources per country-year, then ranked with frank().
cty_re <- cap[TYPE == "Country" & SRC != "Fossil", .(GW = sum(VALUE) / 1000), by = .(LABEL, YEAR)]
cty_ff <- cap[TYPE == "Country" & SRC == "Fossil", .(FF_GW = sum(VALUE) / 1000), by = .(LABEL, YEAR)]
cty <- merge(cty_re, cty_ff, by = c("LABEL", "YEAR"), all.x = TRUE)
cty[, RE_SHARE := fifelse(is.na(FF_GW), NA_real_, GW / (GW + FF_GW) * 100)]
c23 <- cty[YEAR == 2023]
c23[, RANK := frank(-GW, ties.method = "min")]
world_re_2023 <- world_w[YEAR == 2023, Renewable]
c23[, SHARE_WORLD := GW / world_re_2023 * 100]
top15 <- c23[order(RANK)][1:15, .(RANK, LABEL, RE_GW = round(GW, 1), SHARE_OF_WORLD_pct = round(SHARE_WORLD, 1), RE_SHARE_OWN_pct = round(RE_SHARE, 1))]
top15 RANK LABEL RE_GW SHARE_OF_WORLD_pct RE_SHARE_OWN_pct
<int> <char> <num> <num> <num>
1: 1 China 1453.7 37.8 51.9
2: 2 USA 382.5 9.9 34.7
3: 3 Brazil 194.1 5.0 86.5
4: 4 India 175.9 4.6 36.1
5: 5 Germany 166.9 4.3 65.2
6: 6 Japan 128.4 3.3 41.0
7: 7 Canada 108.7 2.8 75.6
8: 8 Spain 80.1 2.1 67.8
9: 9 France 67.7 1.8 78.5
10: 10 Italy 64.4 1.7 52.9
11: 11 Türkiye 56.8 1.5 54.5
12: 12 Russia 56.6 1.5 23.4
13: 13 United Kingdom 55.7 1.4 56.6
14: 14 Australia 54.8 1.4 51.9
15: 15 Vietnam 46.0 1.2 55.6
c23[LABEL == "India", .(RANK, of = nrow(c23), RE_GW = round(GW, 1), SHARE_OF_WORLD_pct = round(SHARE_WORLD, 1))] RANK of RE_GW SHARE_OF_WORLD_pct
<int> <int> <num> <num>
1: 4 221 175.9 4.6
t3 <- c23[order(RANK)][1:15]
t3[, IS_INDIA := fifelse(LABEL == "India", "India", "Other")]
t3[, tip := paste0(LABEL, " (rank ", RANK, ")\n", round(GW, 1), " GW, ", round(SHARE_WORLD, 1), "% of world")]
p3 <- ggplot(t3, aes(x = reorder(LABEL, GW), y = GW, fill = IS_INDIA)) +
geom_col_interactive(aes(tooltip = tip, data_id = LABEL), width = 0.7, show.legend = FALSE) +
geom_text(aes(label = paste0(round(GW), " GW")), hjust = -0.15, size = 3.4, fontface = "bold") +
annotate("text", x = 3, y = max(t3$GW) * 1.05, size = 3.4, colour = "#4a4a6a", hjust = 1,
label = paste0("Top 3 = ", round(sum(t3[RANK <= 3, SHARE_WORLD])), "% of world renewable capacity")) +
scale_fill_manual(values = c(India = "#E63946", Other = "#90A4AE")) +
scale_y_continuous(expand = expansion(mult = c(0, 0.12))) +
coord_flip() +
labs(title = "Top 15 countries by renewable installed capacity, 2023",
subtitle = "Gigawatts; India highlighted in red", x = NULL, y = "Renewable capacity (GW)", caption = src_cap) +
theme_econ
girafe(ggobj = p3, width_svg = 9, height_svg = 6, options = ggi_opts)5.2 India against global powers over time
Why. A single-year ranking hides trajectory. Plotting India with the leading economies shows whether it is converging on or diverging from them.
How. Six economies are selected and each series is labelled at its 2023 end point.
peers <- c("India", "China", "USA", "Brazil", "Germany", "Japan")
d4 <- cty[LABEL %in% peers]
d4[, tip := paste0(LABEL, " ", YEAR, ": ", round(GW, 1), " GW")]
e4 <- d4[YEAR == 2023][order(GW)]
e4[, YL := GW]
for (i in 2:nrow(e4)) set(e4, i = i, j = "YL", value = max(e4$GW[i], e4$YL[i - 1] + 45))
pal_peer <- c(India = "#E63946", China = "#2196F3", USA = "#4CAF50", Brazil = "#9C27B0", Germany = "#FF9800", Japan = "#00BCD4")
p4 <- ggplot(d4, aes(x = YEAR, y = GW, colour = LABEL, group = LABEL)) +
geom_line_interactive(aes(data_id = LABEL), linewidth = 1.1) +
geom_point_interactive(aes(tooltip = tip, data_id = LABEL), size = 1.6) +
geom_text(data = e4, aes(y = YL, label = paste0(LABEL, " ", round(GW), " GW")), hjust = -0.08, size = 3.2, fontface = "bold", show.legend = FALSE) +
scale_colour_manual(values = pal_peer, name = "Economy") +
scale_x_continuous(breaks = seq(2000, 2023, 4), limits = c(2000, 2028)) +
labs(title = "Renewable installed capacity: India versus global powers",
subtitle = "Gigawatts, 2000 to 2023", x = "Year", y = "Renewable capacity (GW)", caption = src_cap) +
theme_econ
girafe(ggobj = p4, width_svg = 9, height_svg = 5.5, options = ggi_opts)5.3 Growth: CAGR 2010 to 2023 among the largest countries
Why. CAGR removes the effect of starting size so fast and slow builders are comparable.
How. The 20 largest renewable countries in 2023 are cast to a 2010 versus 2023 table, and the formula above is applied with T = 13.
top20 <- c23[order(RANK)][1:20, LABEL]
cg <- dcast(cty[LABEL %in% top20 & YEAR %in% c(2010, 2023)], LABEL ~ YEAR, value.var = "GW")
setnames(cg, c("2010", "2023"), c("GW_2010", "GW_2023"))
cg[, CAGR_pct := ((GW_2023 / GW_2010)^(1 / 13) - 1) * 100]
cg[, CAGR_RANK := frank(-CAGR_pct, ties.method = "min")]
cg[order(CAGR_RANK), .(CAGR_RANK, LABEL, GW_2010 = round(GW_2010, 1), GW_2023 = round(GW_2023, 1), CAGR_pct = round(CAGR_pct, 1))] CAGR_RANK LABEL GW_2010 GW_2023 CAGR_pct
<int> <char> <num> <num> <num>
1: 1 Netherlands 3.0 35.6 20.9
2: 2 South Korea 2.8 30.2 20.1
3: 3 China 233.2 1453.7 15.1
4: 4 United Kingdom 9.6 55.7 14.5
5: 5 Vietnam 8.8 46.0 13.6
6: 6 Australia 11.2 54.8 13.0
7: 7 Japan 35.5 128.4 10.4
8: 8 India 52.3 175.9 9.8
9: 9 Türkiye 17.3 56.8 9.6
10: 10 Germany 56.5 166.9 8.7
11: 11 USA 135.6 382.5 8.3
12: 12 Mexico 12.6 32.0 7.5
13: 13 Italy 28.8 64.4 6.4
14: 14 Brazil 89.6 194.1 6.1
15: 15 France 31.5 67.7 6.1
16: 16 Spain 42.2 80.1 5.1
17: 17 Sweden 22.7 40.6 4.6
18: 18 Canada 81.2 108.7 2.3
19: 19 Norway 30.2 40.2 2.2
20: 20 Russia 48.0 56.6 1.3
CAGR_RANK LABEL GW_2010 GW_2023 CAGR_pct
<int> <char> <num> <num> <num>
cg[, IS_INDIA := fifelse(LABEL == "India", "India", "Other")]
cg[, tip := paste0(LABEL, "\nCAGR ", round(CAGR_pct, 1), "%\n", round(GW_2010, 1), " GW to ", round(GW_2023, 1), " GW")]
p6 <- ggplot(cg, aes(x = reorder(LABEL, CAGR_pct), y = CAGR_pct, fill = IS_INDIA)) +
geom_col_interactive(aes(tooltip = tip, data_id = LABEL), width = 0.7, show.legend = FALSE) +
geom_text(aes(label = paste0(round(CAGR_pct, 1), "%")), hjust = -0.15, size = 3.3, fontface = "bold") +
geom_hline(yintercept = ((world_w[YEAR == 2023, Renewable] / world_w[YEAR == 2010, Renewable])^(1 / 13) - 1) * 100,
linetype = "dashed", colour = "#1a1a2e") +
annotate("text", x = 2, y = ((world_w[YEAR == 2023, Renewable] / world_w[YEAR == 2010, Renewable])^(1 / 13) - 1) * 100 + 0.3,
label = "World CAGR", hjust = 0, size = 3.2, colour = "#1a1a2e") +
scale_fill_manual(values = c(India = "#E63946", Other = "#90A4AE")) +
scale_y_continuous(expand = expansion(mult = c(0, 0.12))) +
coord_flip() +
labs(title = "Renewable capacity CAGR, 2010 to 2023: 20 largest countries",
subtitle = "Dashed line = world CAGR", x = NULL, y = "CAGR (%)", caption = src_cap) +
theme_econ
girafe(ggobj = p6, width_svg = 9, height_svg = 6, options = ggi_opts)5.4 India’s energy mix and rank by source
Why. Totals hide composition. India’s mix shows which technology drove the build-out, and its rank per source shows where it is a global power.
How. Country rows are ranked within each source with frank() by group, and India’s rows are extracted.
rk <- cap[TYPE == "Country" & YEAR == 2023 & VALUE > 0]
rk[, `:=`(RANK = frank(-VALUE, ties.method = "min"), OF = .N), by = SRC]
rk[LABEL == "India", .(SRC, GW = round(VALUE / 1000, 1), RANK, OF)][order(RANK)] SRC GW RANK OF
<char> <num> <int> <int>
1: Fossil 311.9 3 224
2: Wind 44.7 4 156
3: Bio 10.8 4 144
4: Solar 73.1 5 220
5: Hydro 47.3 6 163
rk[RANK <= 3 & SRC != "Fossil", .(SRC, RANK, LABEL, GW = round(VALUE / 1000, 1))][order(SRC, RANK)] SRC RANK LABEL GW
<char> <int> <char> <num>
1: Bio 1 China 31.3
2: Bio 2 Brazil 17.6
3: Bio 3 USA 11.0
4: Hydro 1 China 370.6
5: Hydro 2 Brazil 109.9
6: Hydro 3 USA 84.3
7: Solar 1 China 609.9
8: Solar 2 USA 139.2
9: Solar 3 Japan 89.1
10: Wind 1 China 441.9
11: Wind 2 USA 148.0
12: Wind 3 Germany 69.5
im <- cap[LABEL == "India" & SRC != "Fossil" & YEAR %in% seq(2000, 2023, 3), .(YEAR, SRC, GW = VALUE / 1000)]
im <- rbind(im, cap[LABEL == "India" & SRC != "Fossil" & YEAR == 2023, .(YEAR, SRC, GW = VALUE / 1000)])
im <- unique(im)
im[, SRC := factor(SRC, levels = c("Bio", "Hydro", "Wind", "Solar"))]
im[, tip := paste0(SRC, " ", YEAR, ": ", round(GW, 1), " GW")]
imt <- im[, .(GW = sum(GW)), by = YEAR]
p5 <- ggplot(im, aes(x = factor(YEAR), y = GW, fill = SRC)) +
geom_col_interactive(aes(tooltip = tip, data_id = paste(SRC, YEAR)), width = 0.7) +
geom_text(data = im[GW >= 8], aes(label = round(GW)), position = position_stack(vjust = 0.5), size = 3) +
geom_text(data = imt, aes(x = factor(YEAR), y = GW, label = paste0(round(GW), " GW")), inherit.aes = FALSE, vjust = -0.5, fontface = "bold", size = 3.3) +
scale_fill_manual(values = pal_src, name = "Source") +
labs(title = "India's renewable capacity mix", subtitle = "Gigawatts; solar overtakes wind and hydro in the latest years",
x = "Year", y = "Installed capacity (GW)", caption = src_cap) +
theme_econ
girafe(ggobj = p5, width_svg = 9, height_svg = 5.5, options = ggi_opts)5.5 India: solar and wind year-on-year growth
Why. Year-on-year growth shows acceleration and slowdown phases that totals smooth over.
How. After sorting by year within each source, growth is (value / shift(value) - 1) * 100. Early years start from a near-zero base, so the plot window is capped and labels appear only for visible bars.
iy <- cap[LABEL == "India" & SRC %in% c("Solar", "Wind")][order(SRC, YEAR)]
iy[, YOY := (VALUE / shift(VALUE) - 1) * 100, by = SRC]
iy <- iy[YEAR >= 2010 & !is.na(YOY)]
iy[, tip := paste0(SRC, " ", YEAR, ": ", round(YOY, 1), "% (", round(VALUE / 1000, 2), " GW)")]
p10 <- ggplot(iy, aes(x = YEAR, y = YOY, fill = SRC)) +
geom_col_interactive(aes(tooltip = tip, data_id = paste(SRC, YEAR)), position = "dodge", width = 0.75) +
geom_text(data = iy[YOY <= 150], aes(label = round(YOY)), position = position_dodge(width = 0.75), vjust = -0.3, size = 2.7) +
geom_hline(yintercept = 0, colour = "#444466") +
annotate("text", x = 2010.3, y = 140, hjust = 0, size = 3.1, colour = "#4a4a6a",
label = "Bars above 150% are cut off (low base); see tooltips") +
coord_cartesian(ylim = c(-5, 150)) +
scale_fill_manual(values = pal_src, name = "Source") +
scale_x_continuous(breaks = 2010:2023) +
labs(title = "India: year-on-year capacity growth, solar versus wind", subtitle = "Percent change on previous year",
x = "Year", y = "YoY growth (%)", caption = src_cap) +
theme_econ
girafe(ggobj = p10, width_svg = 9, height_svg = 5, options = ggi_opts)6 Groups and regions
6.1 Group and region comparison, 2023
Why. G20 and G7 describe the economic powers, and Advanced versus Emerging economies describes the development divide. Comparing each group’s size and renewable share of its own mix separates a big fleet from a clean fleet.
How. For each aggregate entity, renewable and fossil capacity are summed with conditional sums in j. India and China are added as country reference rows.
ents <- c(grps, regs, "India", "China, People's Republic of")
ag <- cap[YEAR == 2023 & COUNTRY %in% ents, .(
TYPE = TYPE[1],
RE_GW = sum(VALUE[SRC != "Fossil"]) / 1000,
FF_GW = sum(VALUE[SRC == "Fossil"]) / 1000,
SOLAR_GW = sum(VALUE[SRC == "Solar"]) / 1000,
WIND_GW = sum(VALUE[SRC == "Wind"]) / 1000), by = LABEL]
ag[, `:=`(RE_SHARE_OWN = RE_GW / (RE_GW + FF_GW) * 100, SHARE_WORLD_RE = RE_GW / world_re_2023 * 100)]
ag[order(-RE_GW), .(TYPE, LABEL, RE_GW = round(RE_GW), FF_GW = round(FF_GW), SOLAR_GW = round(SOLAR_GW), WIND_GW = round(WIND_GW),
RE_SHARE_OWN_pct = round(RE_SHARE_OWN, 1), SHARE_WORLD_RE_pct = round(SHARE_WORLD_RE, 1))] TYPE LABEL RE_GW FF_GW SOLAR_GW WIND_GW RE_SHARE_OWN_pct
<char> <char> <num> <num> <num> <num> <num>
1: Group G20 3408 3692 1319 978 48.0
2: Group EMDE 2509 3064 891 626 45.0
3: Region Asia 2054 2701 871 522 43.2
4: Country China 1454 1346 610 442 51.9
5: Group Advanced Econ. 1340 1464 527 391 47.8
6: Group G7 974 1147 382 303 45.9
7: Region Europe 840 622 290 258 57.4
8: Region Americas 831 960 210 214 46.4
9: Country India 176 312 73 45 36.1
10: Region Oceania 63 56 33 14 53.2
11: Region Africa 61 189 13 9 24.4
SHARE_WORLD_RE_pct
<num>
1: 88.5
2: 65.2
3: 53.4
4: 37.8
5: 34.8
6: 25.3
7: 21.8
8: 21.6
9: 4.6
10: 1.6
11: 1.6
g7 <- melt(ag[, .(LABEL, TYPE, `Share of world renewable capacity (%)` = SHARE_WORLD_RE, `Renewable share of own capacity (%)` = RE_SHARE_OWN)],
id.vars = c("LABEL", "TYPE"), variable.name = "METRIC", value.name = "PCT")
g7[, tip := paste0(LABEL, " (", TYPE, ")\n", METRIC, ": ", round(PCT, 1))]
p7 <- ggplot(g7, aes(x = reorder(LABEL, PCT), y = PCT, fill = TYPE)) +
geom_col_interactive(aes(tooltip = tip, data_id = LABEL), width = 0.7) +
geom_text(aes(label = round(PCT, 1)), hjust = -0.15, size = 3) +
scale_fill_manual(values = c(Group = "#5C6BC0", Region = "#26A69A", Country = "#E63946"), name = NULL) +
scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
coord_flip() +
facet_wrap(~METRIC, scales = "free") +
labs(title = "Groups, regions and reference countries, 2023",
subtitle = "Left: who owns the world fleet. Right: how clean each fleet is (groups overlap, so shares do not add to 100)",
x = NULL, y = NULL, caption = src_cap) +
theme_econ
girafe(ggobj = p7, width_svg = 10, height_svg = 6, options = ggi_opts)6.2 Renewable versus fossil capacity by region
Why. The regional balance of clean versus fossil capacity shows where the transition is furthest along.
How. The five regions are summed per type, and the renewable share of own mix is labelled above each pair.
r8 <- cap[YEAR == 2023 & TYPE == "Region", .(GW = sum(VALUE) / 1000), by = .(LABEL, GROUP = fifelse(SRC == "Fossil", "Fossil", "Renewable"))]
r8s <- dcast(r8, LABEL ~ GROUP, value.var = "GW")
r8s[, SHARE := Renewable / (Renewable + Fossil) * 100]
r8[r8s, SHARE := i.SHARE, on = "LABEL"]
r8[, AXIS := paste0(LABEL, "\nRE share ", round(SHARE), "%")]
r8[, tip := paste0(LABEL, " ", GROUP, ": ", round(GW), " GW")]
p8 <- ggplot(r8, aes(x = reorder(AXIS, -GW), y = GW, fill = GROUP)) +
geom_col_interactive(aes(tooltip = tip, data_id = paste(LABEL, GROUP)), position = "dodge", width = 0.7) +
geom_text(aes(label = round(GW)), position = position_dodge(width = 0.7), vjust = -0.4, size = 3.2) +
scale_fill_manual(values = c(Fossil = "#EF5350", Renewable = "#42A5F5"), name = NULL) +
scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
labs(title = "Renewable versus fossil installed capacity by region, 2023", subtitle = "Gigawatts",
x = NULL, y = "Installed capacity (GW)", caption = src_cap) +
theme_econ
girafe(ggobj = p8, width_svg = 9, height_svg = 5.5, options = ggi_opts)7 Statistical analysis
7.1 Descriptive statistics of the country distribution
Why. Before any test, the shape of the distribution matters. Renewable capacity across countries is very uneven, so the mean, median and spread show how misleading an “average country” would be. The skewness coefficient and top-5 share quantify concentration.
How. Country-level renewable capacity for 2023 is summarised in one j expression: mean, median, standard deviation, IQR, 90th percentile, skewness \[g_1=\frac{\tfrac1n\sum (x_i-\bar x)^3}{s^3},\] and the share of the five largest countries. India’s percentile is the share of countries at or below India’s value.
d23 <- c23[GW > 0]
d23[, .(countries = .N,
mean_GW = round(mean(GW), 2), median_GW = round(median(GW), 3), sd_GW = round(sd(GW), 2),
IQR_GW = round(IQR(GW), 3), p90_GW = round(quantile(GW, 0.9), 2), max_GW = round(max(GW), 1),
skewness = round(mean((GW - mean(GW))^3) / sd(GW)^3, 2),
top5_share_pct = round(sum(tail(sort(GW), 5)) / sum(GW) * 100, 1),
india_percentile = round(mean(GW <= GW[LABEL == "India"]) * 100, 1))] countries mean_GW median_GW sd_GW IQR_GW p90_GW max_GW skewness
<int> <num> <num> <num> <num> <num> <num> <num>
1: 221 17.42 0.824 103.58 4.961 21.44 1453.7 12.32
top5_share_pct india_percentile
<num> <num>
1: 61.7 98.6
b11 <- cap[TYPE == "Country" & YEAR == 2023 & VALUE > 0 & SRC != "Fossil"]
b11[, GW := VALUE / 1000]
i11 <- b11[LABEL == "India"]
i11[, tip := paste0("India ", SRC, ": ", round(GW, 1), " GW")]
p11 <- ggplot(b11, aes(x = SRC, y = GW, fill = SRC)) +
geom_boxplot(width = 0.55, alpha = 0.7, outlier.shape = 21, outlier.colour = "grey50") +
geom_point_interactive(data = i11, aes(tooltip = tip, data_id = SRC), colour = "#E63946", size = 4.5, shape = 18) +
geom_text(data = i11, aes(label = paste0("India ", round(GW, 1))), colour = "#E63946", hjust = -0.2, size = 3.1, fontface = "bold") +
scale_y_log10() +
scale_fill_manual(values = pal_src, guide = "none") +
labs(title = "Distribution of country capacity by source, 2023", subtitle = "Log scale; box = IQR; red diamond = India",
x = NULL, y = "Capacity (GW, log scale)", caption = src_cap) +
theme_econ
girafe(ggobj = p11, width_svg = 9, height_svg = 5.5, options = ggi_opts)7.3 Log-linear trend regression
Why. A regression on log capacity gives the average annual growth rate with a measure of fit. Levels grow exponentially, so a straight-line fit on raw capacity would misstate growth.
How. For each series, lm(log(GW) ~ YEAR) is fitted inside a data.table grouped call. The slope \(b\) is converted to a growth rate \(g=(e^{b}-1)\times100\), with \(R^2\) and the slope’s p-value.
ts_dt <- rbind(
world_w[, .(SERIES = "World renewable", YEAR, GW = Renewable)],
world_w[, .(SERIES = "World fossil", YEAR, GW = Fossil)],
cty[LABEL == "India", .(SERIES = "India renewable", YEAR, GW)],
cty[LABEL == "China", .(SERIES = "China renewable", YEAR, GW)],
cty[LABEL == "USA", .(SERIES = "USA renewable", YEAR, GW)])
ts_dt <- ts_dt[GW > 0]
trend <- ts_dt[, {
m <- lm(log(GW) ~ YEAR)
s <- summary(m)$coefficients
.(growth_pct_per_year = round((exp(s["YEAR", 1]) - 1) * 100, 2), r_squared = round(summary(m)$r.squared, 3),
p_value = signif(s["YEAR", 4], 3), n_years = .N)
}, by = SERIES]
trend SERIES growth_pct_per_year r_squared p_value n_years
<char> <num> <num> <num> <int>
1: World renewable 7.54 0.982 1.00e-20 24
2: World fossil 3.18 0.977 1.80e-19 24
3: India renewable 9.07 0.992 1.07e-24 24
4: China renewable 14.17 0.996 5.18e-28 24
5: USA renewable 6.94 0.972 1.45e-18 24
t14 <- ts_dt[SERIES %in% c("World renewable", "World fossil", "India renewable")]
t14[, tip := paste0(SERIES, " ", YEAR, ": ", round(GW, 1), " GW")]
tl <- trend[SERIES %in% c("World renewable", "World fossil", "India renewable")]
tl[, YPOS := fcase(SERIES == "World fossil", 12000, SERIES == "World renewable", 7500, default = 4700)]
p14 <- ggplot(t14, aes(x = YEAR, y = GW, colour = SERIES)) +
geom_smooth(method = "lm", formula = y ~ x, se = TRUE, linewidth = 0.7, alpha = 0.15) +
geom_point_interactive(aes(tooltip = tip, data_id = paste(SERIES, YEAR)), size = 2) +
geom_text(data = tl, aes(x = 2000, y = YPOS, label = paste0(SERIES, ": ", growth_pct_per_year, "%/yr, R2 ", r_squared)),
inherit.aes = FALSE, hjust = 0, size = 3.2, fontface = "bold") +
scale_y_log10() +
scale_colour_manual(values = c("World renewable" = "#1565C0", "World fossil" = "#EF5350", "India renewable" = "#E63946"), name = NULL) +
labs(title = "Log-linear trends in installed capacity", subtitle = "Straight lines on a log axis = constant percentage growth",
x = "Year", y = "Capacity (GW, log scale)", caption = src_cap) +
theme_econ
girafe(ggobj = p14, width_svg = 9, height_svg = 5.5, options = ggi_opts)7.4 Panel regression with country fixed effects
Why. Country-by-country slopes can be noisy. A panel with country fixed effects estimates the common growth rate across the largest 30 renewable countries while removing permanent size differences between countries.
How. lm(log(GW) ~ YEAR + factor(country)) is fitted on the 30 largest countries from 2010 onward. The YEAR coefficient is the within-country growth rate. A second table ranks individual country growth rates so India’s position is visible.
top30 <- c23[order(RANK)][1:30, LABEL]
pn <- cty[LABEL %in% top30 & YEAR >= 2010 & GW > 0]
m_fe <- lm(log(GW) ~ YEAR + factor(LABEL), data = pn)
sf <- summary(m_fe)$coefficients
data.table(common_growth_pct = round((exp(sf["YEAR", 1]) - 1) * 100, 2), std_error = round(sf["YEAR", 2], 4),
t_value = round(sf["YEAR", 3], 1), p_value = signif(sf["YEAR", 4], 3),
r_squared = round(summary(m_fe)$r.squared, 3), observations = nobs(m_fe)) common_growth_pct std_error t_value p_value r_squared observations
<num> <num> <num> <num> <num> <int>
1: 8.68 0.0028 29.8 2.55e-102 0.964 420
gr <- pn[, {
m <- lm(log(GW) ~ YEAR)
.(growth_pct = round((exp(coef(m)[2]) - 1) * 100, 1), r2 = round(summary(m)$r.squared, 3))
}, by = LABEL][order(-growth_pct)]
gr[, RANK := .I]
gr[RANK <= 8 | LABEL == "India"] LABEL growth_pct r2 RANK
<char> <num> <num> <int>
1: South Korea 22.4 0.990 1
2: Netherlands 22.1 0.975 2
3: Poland 18.4 0.960 3
4: Taiwan 14.8 0.938 4
5: China 14.5 0.997 5
6: United Kingdom 13.9 0.904 6
7: Australia 13.7 0.978 7
8: Vietnam 13.4 0.939 8
9: India 10.3 0.990 11
7.5 Correlation: India solar versus wind
Why. Do solar and wind move together because of shared policy, or are they substitutes? Both series trend upward, so a correlation in levels is inflated by the shared trend. Correlating annual additions removes the trend.
How. Capacity is cast wide by year. cor.test() is run on levels and on first differences (annual additions in MW) so the two answers can be compared.
isw <- dcast(cap[LABEL == "India" & SRC %in% c("Solar", "Wind")], YEAR ~ SRC, value.var = "VALUE")
isw[, `:=`(D_SOLAR = Solar - shift(Solar), D_WIND = Wind - shift(Wind))]
ct_lvl <- cor.test(isw$Solar, isw$Wind)
ct_dif <- cor.test(isw$D_SOLAR, isw$D_WIND)
data.table(basis = c("Levels (MW)", "Annual additions (MW)"),
pearson_r = round(c(ct_lvl$estimate, ct_dif$estimate), 3),
p_value = signif(c(ct_lvl$p.value, ct_dif$p.value), 3),
n = c(ct_lvl$parameter + 2, ct_dif$parameter + 2)) basis pearson_r p_value n
<char> <num> <num> <num>
1: Levels (MW) 0.864 5.35e-08 24
2: Annual additions (MW) 0.370 8.24e-02 23
7.6 Normality of the country distribution
Why. Many tests (t-test, ANOVA) assume roughly normal data. If the raw distribution is strongly skewed, a log transform or a rank-based test is needed.
How. The Shapiro-Wilk statistic \(W\) is computed on raw and on \(\log_{10}\) country renewable capacity. \(W\) near 1 means close to normal, and a small p-value rejects normality.
sh_raw <- shapiro.test(d23$GW)
sh_log <- shapiro.test(log10(d23$GW))
data.table(scale = c("Raw GW", "log10 GW"), W = round(c(sh_raw$statistic, sh_log$statistic), 4),
p_value = signif(c(sh_raw$p.value, sh_log$p.value), 3), n = nrow(d23)) scale W p_value n
<char> <num> <num> <int>
1: Raw GW 0.1375 1.42e-30 221
2: log10 GW 0.9819 6.16e-03 221
7.7 Hypothesis test: do emerging economies grow renewables faster than advanced ones?
Why. The development divide is central to energy policy. A paired test asks whether, year by year, the emerging group’s capacity growth differs from the advanced group’s.
How. Annual percentage growth of renewable capacity is computed per group, cast wide by year, and compared with a paired t-test and the rank-based Wilcoxon signed-rank test. Pairing by year removes global shocks common to both groups. Annual growth rates are serially correlated, so p-values are indicative rather than exact.
gg <- cap[COUNTRY %in% c("Advanced Economies", "Emerging Market and Developing Economies") & SRC != "Fossil",
.(GW = sum(VALUE) / 1000), by = .(LABEL, YEAR)][order(LABEL, YEAR)]
gg[, YOY := (GW / shift(GW) - 1) * 100, by = LABEL]
gw <- dcast(gg[!is.na(YOY)], YEAR ~ LABEL, value.var = "YOY")
setnames(gw, c("Advanced Econ.", "EMDE"), c("AE", "EMDE"))
tt <- t.test(gw$EMDE, gw$AE, paired = TRUE)
wt <- wilcox.test(gw$EMDE, gw$AE, paired = TRUE, exact = FALSE)
gw[, .(mean_yoy_AE_pct = round(mean(AE), 2), mean_yoy_EMDE_pct = round(mean(EMDE), 2), years = .N)] mean_yoy_AE_pct mean_yoy_EMDE_pct years
<num> <num> <int>
1: 6.13 8.34 23
data.table(test = c("Paired t-test", "Wilcoxon signed-rank"),
statistic = round(c(tt$statistic, wt$statistic), 2), p_value = signif(c(tt$p.value, wt$p.value), 3)) test statistic p_value
<char> <num> <num>
1: Paired t-test 4.91 6.59e-05
2: Wilcoxon signed-rank 259.00 2.47e-04
7.8 ANOVA: do world growth rates differ by technology?
Why. If solar, wind, hydro and bioenergy grew at the same pace, technology would not matter for planning. ANOVA tests whether mean annual growth differs by source.
How. World annual growth per source is computed with shift(), then aov(YOY ~ source) gives the F test, and the rank-based Kruskal-Wallis test confirms it without normality assumptions.
wy <- world[SRC != "Fossil"][order(SRC, YEAR)]
wy[, YOY := (GW / shift(GW) - 1) * 100, by = SRC]
wy <- wy[!is.na(YOY)]
wy[, .(mean_yoy_pct = round(mean(YOY), 1), sd_pct = round(sd(YOY), 1), years = .N), by = SRC][order(-mean_yoy_pct)] SRC mean_yoy_pct sd_pct years
<char> <num> <num> <int>
1: Solar 36.9 17.2 23
2: Wind 19.7 8.0 23
3: Bio 7.4 2.6 23
4: Hydro 2.6 0.9 23
av <- summary(aov(YOY ~ SRC, data = wy))[[1]]
kw <- kruskal.test(YOY ~ factor(SRC), data = wy)
data.table(test = c("One-way ANOVA", "Kruskal-Wallis"),
statistic = round(c(av[1, "F value"], kw$statistic), 2),
p_value = signif(c(av[1, "Pr(>F)"], kw$p.value), 3)) test statistic p_value
<char> <num> <num>
1: One-way ANOVA 58.35 7.35e-21
2: Kruskal-Wallis 78.71 5.81e-17
8 Key findings
- Reconciled world. Regions and the Advanced plus Emerging groups agree exactly. World renewable capacity in 2023 is about 3849 GW against 4528 GW of fossil capacity. Summing every row without removing aggregates overstates the world renewable fleet about 5.1 times.
- Growth. World renewable capacity grew about 5.2 times from 2000 to 2023, a log-linear rate near 7.54% a year, against 3.18% for fossil. The renewable share of capacity moved from 24.5% to 45.9%.
- Global powers. The top three countries (China, USA, Brazil) hold 53% of world renewable capacity.
- India. India ranks 4 of 221 countries with 175.9 GW, 4.6% of the world fleet. Its 2010 to 2023 CAGR is 9.8% (rank 8 among the 20 largest), and its log-linear trend is 9.07% a year.
- Investment. India’s public renewable investment in the latest year is 0.00023% of GDP against a median of 0.0011% among the 136 countries that report a positive value.
- Groups and regions. Emerging and Developing Economies own 65% of world renewable capacity and Advanced Economies the rest. The G20 holds 89% and the G7 25%. The renewable share of each group’s own capacity is similar (47.8% advanced, 45% emerging), but emerging economies grow renewables faster: 8.3% a year against 6.1% (paired t-test p = 6.6e-05). Europe has the highest renewable share among regions and Africa the lowest.
- Statistics. Country capacity is extremely skewed (skewness above 12) and non-normal, so log transforms and rank-based tests are used. India’s solar and wind capacity correlate at r = 0.86 in levels, but only r = 0.37 (p = 0.082) in annual additions, so most of the apparent co-movement is the shared upward trend. Technologies differ significantly in growth: solar averaged the fastest annual growth, hydro the slowest.
9 Limitations
- Capacity covers four renewable sources plus fossil fuels. Geothermal, marine and pumped storage are not in the capacity indicator, so renewable capacity is slightly understated.
- Hong Kong, Taiwan and small territories are listed as separate economies, as in the source file.
- Annual growth series are serially correlated, so p-values from tests on yearly rates are indicative.
- Interactive ggiraph output needs HTML. PDF or Word export keeps tables and code but not the hover behaviour.