Python Documentation

All operations were performed in a Google Colab environment

Import Libraries

An up-to-date version of Plotly is required. The following libraries are used for data processing and visualization:

!pip install --upgrade plotly
#--upgrade is required for specific hover menu functions

import plotly.express as px
import plotly.graph_objects as go

import pandas as pd
import numpy as np

Reference Lists and Dictionaries

List of all election years, used for columns, y-axis labels and sorting:

#used for columns, y-axis labels and sorting
electionYears = [1949, 1953, 1957, 1961, 1965, 1969, 1972,	1976,	1980,	1983,	1987,	1990,	1994,	1998,	2002,	2005,	2009,	2013,	2017,	2021,	2025]

Color codes for graphing, keyed by party name:

partyColor = { "CDU/CSU" : "#2d3c4b", "SPD" : "#E3000F", "FDP" : "#ffed00", "GRÜNE" : "#46962b",
              "AfD" : "#0099ff", "DIE LINKE" : "#FF0000", "SSW" : "#003c8f",
              "NPD" : "#8b4726", "PIRATEN" : "#FF8800", "BSW" : "#792351", "REP" : "#0075be",
              "KPD/DKP":"#FF355E", "DRP" : "#8b4726", "GB/BHE" : "#A04B90", "DP" : "#EFBF04",
               "BP" : "#3158b0", "Zentrum" : "#3389CC", "Sonstige" : "#878787"
              }

Dictionaries mapping chancellors and coalition parties to election years:

#name of chancellor
chancellorNames = {"Adenauer" : [1949, 1953, 1957, 1961],"Erhard" : [1965], "Brandt" : [1969, 1972],
                   "Schmidt" : [1976, 1980], "Kohl" : [1983, 1987, 1990, 1994], "Schröder" : [1998, 2002],
                   "Merkel" : [2005, 2009, 2013, 2017], "Scholz" : [2021], "Merz" : [2025]}

coalitionYears = {1949 : ["CDU/CSU", "FDP","DP"], 1953 : ["CDU/CSU", "FDP","GB/BHE","DP"],
                  1957 : ["CDU/CSU", "DP"], 1961 : ["CDU/CSU", "FDP"], 1965 : ["CDU/CSU", "FDP"],
                  1969 : ["SPD", "FDP"], 1972 : ["SPD", "FDP"], 1976 : ["SPD", "FDP"], 1980 : ["SPD", "FDP"],
                  1983 : ["CDU/CSU", "FDP"], 1987 : ["CDU/CSU", "FDP"], 1990 : ["CDU/CSU", "FDP"],
                  1994 : ["CDU/CSU", "FDP"], 1998 : ["SPD", "GRÜNE"], 2002 : ["SPD", "GRÜNE"],
                  2005 : ["CDU/CSU", "SPD"], 2009 : ["CDU/CSU", "FDP"],   2013 : ["CDU/CSU", "SPD"],
                  2017 : ["CDU/CSU", "SPD"], 2021 : ["SPD", "GRÜNE", "FDP"], 2025 : ["CDU/CSU", "SPD"]}

#parties relevant for discrepancy chart
discParties = ["CDU/CSU", "SPD", "FDP", "GRÜNE", "AfD"]

Import Data

Data is scraped directly from the wahlrecht.de results table:

url = 'https://www.wahlrecht.de/ergebnisse/bundestag.htm'

tables = pd.read_html(url)
print (f'Total tables: {len(tables)}')

Verify import and transfer to DataFrame:

tables[1].head(3)

electionResults = tables[1]
electionResults.head(3)

Clean Up Dataframe

The raw HTML table contains annotated text values and ambiguous separators that need to be resolved before analysis:

#replace annotated text values
electionResults.replace('5,1¹',51,inplace=True, regex=True)
electionResults.replace('DIE LINKE²','DIE LINKE',inplace=True, regex=True)
electionResults.replace('KPD/DKP³','KPD/DKP',inplace=True, regex=True)

#set first col as index
electionResults = electionResults.set_index(('Unnamed: 0_level_0', 'Unnamed: 0_level_1')).copy()
electionResults.index.name = 'Partei'

#drop irrelevant row
electionResults = electionResults.drop('Wahlbeteiligung')
#drop last, redundant, column
electionResults = electionResults.drop(electionResults.columns[-1], axis=1)
#proper NaNs
electionResults = electionResults.replace(['–', '-'], np.nan)

electionResults.head(3)

Create Dataframes for Analysis

Separate DataFrames are created for seat counts and vote percentages, then normalized and transposed for plotting:

#select seat columns
seats = electionResults.loc[:, (slice(None), 'Sitze')]
#drop multi-index level
seats.columns = seats.columns.droplevel(1)

#assign years as column names
seats.columns = electionYears

seats.head(3) #verify dataframe creation

#select percentage columns
percentage = electionResults.loc[:, (slice(None), '%')]
percentage.columns = percentage.columns.droplevel(1)

percentage.columns = electionYears

#decimal ',' was not recognized during import. replace values
percentage = percentage.astype(float)
percentage = percentage / 10
percentage.loc['BSW',2025] = percentage.loc['BSW',2025]/100 #BSW value was imported incorrectly
percentage.head(3)

percentagePlot = percentage.transpose()

#convert all values in the DataFrame to numeric, coercing errors to NaN
seatShare = seats.apply(pd.to_numeric, errors='coerce')

#calculate the sum of seats for each year, then calculate percentage of total
totalSeats = seatShare.sum(axis=0)
seatShare = seatShare.div(totalSeats, axis=1) * 100

#transpose, set index and sort by year
seatShare = seatShare.transpose()
seatShare.index = seatShare.index.astype(int)
seatShare = seatShare.sort_index(ascending=True)
#drop columns containing only NaN values to limit legend entries
seatShare = seatShare.dropna(axis=1, how='all')

Vote- and Seat Percentage Comparison

Combined and derived DataFrames are prepared for use across all four charts:

#combine percentage and seat dataframes for comparison
percSeatComp = pd.concat([percentagePlot, seatShare])
percSeatComp = percSeatComp.sort_index(level=0) #sort years in index

#create df to analyze discrepancy between votes and seats
discrepancy =  percentagePlot - seatShare
discrepancyRate = (seatShare / percentagePlot) *100

#needed for boxplot
seatConv = discrepancyRate.dropna(axis=1, how='all')

Chart 1: General Overview — Plot and Menu Creation

Each election year contributes three rows to the plot (a spacer, a votes row, and a seats row) with interactive dropdown menus for filtering by chancellor and coalition:

rows = []
meta = []  #keep metadata for y-axis labels and hover menus
#loop through every election year, adding from vote and seat dfs
for y in electionYears:
    #1st spacer row
    spacer = pd.Series(np.nan, index=percentagePlot.columns)
    rows.append(spacer)
    meta.append({"year": y, "row_type": "spacer"})
    #2nd votes row
    votePerc = percentagePlot.loc[y].copy()
    rows.append(votePerc)
    meta.append({"year": y, "row_type": "Stimmanteil"})
    #3rd seats row
    seatPerc = seatShare.loc[y].copy()
    rows.append(seatPerc)
    meta.append({"year": y, "row_type": "Sitzanteil"})

compPlot = pd.DataFrame(rows).reset_index(drop=True)
meta = pd.DataFrame(meta)

#build y categories
y = np.arange(len(compPlot))

#create tick labels only on every 3rd row (your "seats" rows)
#2 instead of 0 or 1 so year label is on top of row entry
tick_vals = np.arange(2, len(compPlot), 3)
tick_text = [str(y) for y in electionYears]

fig = go.Figure() #create plotly figure element

Create primary filtering arrays for votes/seats rows:

#meta["rowType"] should be one of: "spacer", "votes", "seats"
row_type = meta["row_type"].to_numpy()
votesMask = (row_type == "Stimmanteil")
seatsMask = (row_type == "Sitzanteil")
bothMask  = votesMask | seatsMask  # spacer rows remain hidden anyway

#precompute masked x arrays per party
xAll   = [compPlot[c].where(bothMask, None).to_numpy()  for c in compPlot.columns]
xVotes = [compPlot[c].where(votesMask, None).to_numpy() for c in compPlot.columns]
xSeats = [compPlot[c].where(seatsMask, None).to_numpy() for c in compPlot.columns]

Create coalition-based filtering system:

xCoalition = [] #create dropdown menu options for coalitions

for party in compPlot.columns:
    partyCoalitionMask = np.zeros(len(compPlot), dtype=bool)
    #for each row, check if the party was in coalition for that year
    for idx, row_meta in meta.iterrows():
        year = row_meta["year"]
        if year in coalitionYears and party in coalitionYears[year]:
            partyCoalitionMask[idx] = bothMask[idx]

    xCoalition.append(compPlot[party].where(partyCoalitionMask, None).to_numpy())

Refine chancellor dropdown — this fixes an issue arising from selecting "coalition only" after having selected a chancellor:

n_traces = len(compPlot.columns)

#helper: indices (rows) belonging to certain election years
def RowsForYears(years):
    idxs = []
    for i, yr in enumerate(electionYears):
        if yr in years:
            base = i * 3
            idxs.extend([base, base+1, base+2])  # spacer, votes, seats
    return idxs

#precompute chancellor-specific arrays for all menus
chancellorArrays = {}

for chancellor, years in chancellorNames.items():
    yearMask = meta["year"].isin(years).to_numpy()

    #create masks for proper button functions
    bothChancMask  = bothMask  & yearMask
    votesChancMask = votesMask & yearMask
    seatsChancMask = seatsMask & yearMask

    #coalition mask (full length)
    coalition_ch = np.zeros(len(compPlot), dtype=bool)
    for idx, row_meta in meta.iterrows():
        yr = row_meta["year"]
        if yr in coalitionYears:
            #only mark rows that belong to this chancellor and are votes/seats
            if (yr in years) and (row_meta["row_type"] in ("Stimmanteil", "Sitzanteil")):
                coalition_ch[idx] = True

    #build x arrays per trace
    xAll_ch   = [compPlot[c].where(bothChancMask,  None).to_numpy() for c in compPlot.columns]
    xvotesChancMask = [compPlot[c].where(votesChancMask, None).to_numpy() for c in compPlot.columns]
    xseatsChancMask = [compPlot[c].where(seatsChancMask, None).to_numpy() for c in compPlot.columns]

    #coalition per trace must be party-aware per year
    xCoalition_ch = []
    for party in compPlot.columns:
        partyMask = np.zeros(len(compPlot), dtype=bool)
        for idx, row_meta in meta.iterrows():
            yr = row_meta["year"]
            if yr in years and yr in coalitionYears and party in coalitionYears[yr]:
                partyMask[idx] = bothMask[idx]  # votes/seats rows only
        xCoalition_ch.append(compPlot[party].where(partyMask, None).to_numpy())

    #y-axis zoom + ticks for those years (use the seats row of each year: base+2)
    idxs = RowsForYears(years)
    lo, hi = min(idxs), max(idxs)

    tick_vals_ch = [i*3 + 2 for i, yr in enumerate(electionYears) if yr in years]
    tick_text_ch = [str(yr) for yr in electionYears if yr in years]

    chancellorArrays[chancellor] = dict(
        xAll=xAll_ch, xVotes=xvotesChancMask, xSeats=xseatsChancMask, xCoalition=xCoalition_ch,
        tickvals=tick_vals_ch, ticktext=tick_text_ch,
        yrange=[lo - 0.5, hi + 0.5],
        height= 1000
    )

Add dropdown for each chancellor:

chancellorDropdown = []

#default menu including all chancellors
chancellorDropdown.append(
    dict(
        label="Alle Kanzler",
        method="update",
        args=[
            {"x": xAll, "y": [y] * n_traces},
            {
                "yaxis.tickvals": tick_vals,
                "yaxis.ticktext": tick_text,
                "yaxis.range": [-0.5, len(compPlot) - 0.5],
                "title.text": "<b>Bundestagswahlen: Sitzanteil vs. Stimmanteil - Alle Jahre</b>",
                "height": 1000,
                #set menu button args back to global arrays
                "updatemenus[0].buttons[0].args": [{"x": xAll}],
                "updatemenus[0].buttons[1].args": [{"x": xVotes}],
                "updatemenus[0].buttons[2].args": [{"x": xSeats}],
                "updatemenus[2].buttons[0].args": [{"x": xAll}],
                "updatemenus[2].buttons[1].args": [{"x": xCoalition}],
            }
        ]
    )
)

#add one option per chancellor
for chancellor in chancellorNames.keys():
    A = chancellorArrays[chancellor]
    chancellorDropdown.append(
        dict(
            label=chancellor,
            method="update",
            args=[
                {"x": A["xAll"], "y": [y] * n_traces},
                {
                    "yaxis.tickvals": A["tickvals"],
                    "yaxis.ticktext": A["ticktext"],
                    "yaxis.range": A["yrange"],
                    "title.text": f"<b>Bundestagswahlen: Sitzanteil vs. Stimmanteil - {chancellor}</b>",
                    "height": 1000,
                    #rewrite other menus so subsequent clicks use the same chancellor subset
                    "updatemenus[0].buttons[0].args": [{"x": A["xAll"]}],
                    "updatemenus[0].buttons[1].args": [{"x": A["xVotes"]}],
                    "updatemenus[0].buttons[2].args": [{"x": A["xSeats"]}],
                    "updatemenus[2].buttons[0].args": [{"x": A["xAll"]}],
                    "updatemenus[2].buttons[1].args": [{"x": A["xCoalition"]}],
                }
            ]
        )
    )

Create all dropdown and button menus:

fig.update_layout(
  updatemenus=[
        #data type selector (Alle/Stimmanteile/Sitzanteile)
        dict(
            type="buttons",
            direction="right",
            x=0.0, xanchor="left",
            y=1.07, yanchor="bottom",
            pad = {"t":10,"b":0},
            buttons=[
                dict(
                    label="Alle",
                    method="restyle",
                    args=[{"x": xAll}],
                ),
                dict(
                    label="Nur Stimmanteile",
                    method="restyle",
                    args=[{"x": xVotes}],
                ),
                dict(
                    label="Nur Sitzanteile",
                    method="restyle",
                    args=[{"x": xSeats}],
                ),
            ],
        ),
        #chancellor selector dropdown
        dict(
            type="dropdown",
            direction="right",
            x=0, xanchor="left",
            y=1, yanchor="bottom",
            pad = {"t":0,"b":10},
            buttons=chancellorDropdown,
            active=0,
        ),
        #coalition filter button
        dict(
            type="buttons",
            direction="right",
            x=0.41, xanchor="left",
            y=1.07, yanchor="bottom",
            pad = {"t":10,"b":0},
            active=0,
            buttons=[
                dict(
                    label="Alle Parteien",
                    method="restyle",
                    args=[{"x": xAll}],
                ),
                dict(
                    label="Nur Koalitionsparteien",
                    method="restyle",
                    args=[{"x": xCoalition}],
                ),
            ],
        )
    ],
)

Add one stacked horizontal bar trace per party:

#add one stacked trace per party (column)
for i, party in enumerate(compPlot.columns):
    fig.add_trace(
        go.Bar(
            y=y,
            x=xAll[i],
            orientation="h",
            name=party,
            visible=True,
            marker=dict(color=partyColor.get(party, "#CCCCCC")),
            #richer hover: year + whether it's votes or seats
            customdata=np.stack([meta["year"], meta["row_type"]], axis=-1),
            hovertemplate=(
                "%{customdata[1]}<br>"
                f"{party}<br>"
                "Anteil: %{x:.2f}%<extra></extra>"
            ),
        )
    )

Add final styling elements, axis labels and display the chart:

#absolute majority line at 50%
fig.add_vline(x=50, line_width=1, line_color="black")

fig.update_layout(
    barmode="stack",
    title="<b>Bundestagswahlen: Sitzanteil vs. Stimmanteil</b>",
    xaxis_title="<b>Anteil (%)</b>",
    yaxis_title="",
    xaxis=dict(range=[0, 100]),
    yaxis=dict(
        tickmode="array",
        tickvals=tick_vals,
        ticktext=tick_text,
        range=[-0.5, len(compPlot) - 0.5],
    ),
    legend_title="Partei",
    bargap=0,
    plot_bgcolor="#FfFffF",
    hovermode="closest",
    height=1000,
    margin=dict(t=200),
)

fig.update_layout(legend_traceorder="normal")
fig.show()

Export chart:

# fig.write_html("bundestagswahlen_chart.html")

Chart 2: Vote and Seat Share Discrepancy

The discrepancy DataFrame is plotted as a line chart for the major parties. The values are inverted to show the seats gained relative to vote share:

discrepancy.sample(3)

discFig = px.line(-discrepancy[discParties]) #inverse to show seats gained
#discParties is limited to (former) governing or high share parties

for trace in discFig.data: #loop through plot columns
    if trace.name in partyColor:
        #match column/party name and get color coding from dict
        trace.line.color = partyColor[trace.name]

discFig.update_traces(line=dict(width=3),
                       hovertemplate =
                      "<b>%{y:.2f}%</b>")

discFig.update_yaxes(
    zeroline=True,
    zerolinewidth=2,
    zerolinecolor='rgba(0, 0, 0, 0.5)')

discFig.update_layout(
    hovermode="x unified",
    title="<b>Diskrepanz zwischen Stimm- und Sitzanteil</b>",
    yaxis_title="<b>Diskrepanz (%)</b>",
    xaxis_title="<b>Wahljahr</b>",
    xaxis=dict(range=[1948, 2026], unifiedhovertitle=dict(text="<b>Diskrepanz im Wahljahr %{x}</b>")),
    yaxis=dict(range=[-4, 8]),
    legend_title="<b>Partei</b>",
    plot_bgcolor="#ffffff",
    height=700,
    margin=dict(t=100),
)

discFig.show()

Export chart:

discFig.write_html("diskrepanzen_chart.html")

Chart 3: Gallagher Index

The Gallagher Index is computed from the squared discrepancy values and plotted with color-coded background bands indicating proportionality levels:

discrepancyGallagher = discrepancy**2
discrepancyGallagher['Gallagher Score'] = np.sqrt((discrepancyGallagher.sum(axis=1))*.5)
discrepancyGallagher.head(3)

gallFig = px.line(discrepancyGallagher["Gallagher Score"])

#prevent stacking of hrect shapes from rerunning graph
gallFig.layout.shapes = ()

gallFig.update_traces(line=dict(width=3),
              hovertemplate = '<b>Wahljahr %{x} | %{y:.2f}%</b><extra></extra>',
              line_color = 'white')

gallFig.update_yaxes(tickmode = 'linear',dtick = 2)

#hex values taken from Wikipedia shading for Gallgher Index
gallFig.add_hrect(y0=0, y1=1, fillcolor="#0c3091",
                  opacity=1, layer="below", line_width=0)
gallFig.add_hrect(y0=1, y1=3, fillcolor="#2f5cd5",
                  opacity=1, layer="below", line_width=0)
gallFig.add_hrect(y0=3, y1=5, fillcolor="#6bd2df",
                  opacity=1, layer="below", line_width=0)
gallFig.add_hrect(y0=5, y1=7, fillcolor="#c3eded",
                  opacity=1, layer="below", line_width=0)

gallFig.update_layout(
    hovermode="closest",
    title="<b>Gallagher Index nach Wahljahr</b>",
    yaxis_title="<b>Index (%)</b>",
    xaxis_title="<b>Wahljahr</b>",
    xaxis=dict(range=[1948, 2026], showgrid=False),
    yaxis=dict(range=[0, 7]),
    height=600,
    showlegend=False,
)

gallFig.show()

Export chart:

discFig.write_html("gallgher_chart.html")

Chart 4: Seat Conversion Rate

The seat conversion rate (seat share / vote share × 100) is displayed as a boxplot per party, showing the distribution across all elections in which a party was represented:

seatConv.sample(3)

orderedCols = [col for col in partyColor.keys() if col in seatConv.columns and col not in ["Sonstige"]]
seatConv = seatConv[orderedCols]

boxFig = go.Figure()

for col in orderedCols:
    color = partyColor[col]
    boxFig.add_trace(go.Box(
        y=seatConv[col].dropna(),
        name=col,
        marker_color=color,
        line_color=color,
        boxpoints="all",
        pointpos=0,
        text=seatConv[col].dropna().index.astype(str),
        hovertemplate='<b>%{text} | %{y:.2f}</b><extra></extra>'
    ))

boxFig.update_traces(
    yhoverformat=".2f"
)

boxFig.update_layout(
    hovermode="closest",
    title="<b>Sitzkonvertierung von Stimmanteil zu Sitzanteil</b>",
    yaxis_title="<b>Konvertierungsrate (%)</b>",
    xaxis_title="<b>Partei</b>",
    yaxis=dict(range=[20, 140]),
    plot_bgcolor="#ffffff",
    height=600,
    showlegend=False,
)

boxFig.show()

Export chart:

boxFig.write_html("boxplot_chart.html")