Scripts for processing and analyzing the Swiss Household Energy Demand Survey (SHEDS) data in R and Python. SHEDS data is stored in SPSS format (.sav) which includes value labels and variable descriptions. Descriptions and more information can be found at SHEDS - Sweet Cross.

Figure 1. SHEDS variable counts over time

Note: This Figure shows the number of variables (columns) available in SHEDS datasets in each category. These do not correspond to numbers of questions because each item of a question is coded in its own column.

The project provides example scripts demonstrating how to work with SHEDS data in R and Python. It also includes a CSV file listing all question identifiers across survey years, indicating when each question was used, to improve transparency and facilitate longitudinal analysis.

sheds_data_scripts/
├── sheds_questions_up2025.csv # Table of question identifiers over the years
├── README.md
├── .gitignore
└── src/
    ├── python/
    │   ├── utils.py # read_clean_sheds, get_data_summary, save_plot, conditional_ffill, build_car_history, check_finished, analyze_ev_ownership_data
    │   ├── sheds_explore.ipynb # Explore dataset and metadata
    |   ├── car_owners_paper.ipynb # Explore car owners
    │   ├── longitudinal_exploration.ipynb # EV ownership
    │   ├── longitudinal_history_variables.py # Build longitudinal history files (see below)
    │   └── read_sav_example.ipynb # Simple example of how to read and use metadata
    └── R/
        ├── utils.R # read_clean_sheds, get_data_summary, save_plot, build_car_history_stata, analyze_ev_ownership_data, theme_publication, validate_car_history, filter_by_included_vars, generate_latex_row
        ├── sheds_explore.Rmd # Explore dataset and metadata
        ├── outlier_detection.Rmd # Outlier detection
        ├── outliers_functions.R
        ├── car_owners_paper.R # Explore car owners
        ├── wemf_regions_distribution_map.Rmd # Create region distribution map
        └── longitudinal_exploration.rmd # EV ownership

longitudinal_history_variables.py builds two panel datasets (one row per respondent × wave) saved as .csv and .pkl:

• sheds_accom_history — accommodation type (accom1: owner / tenant / cooperative) across all waves, plus additional housing variables. Includes a filled column accom1_filled to handle survey skip logic:
  • In 2018 and 2019, returning respondents who did not move were not asked accom1 again (coded -2 DNA). Their previous answer is carried forward into accom1_filled.
  • In 2020, a small number of respondents have a missing accom1; these are also filled from the previous wave.
  • If a respondent reported a move (accom_change == 1), no fill is applied — the new value must be explicit.
• sheds_identifier_history — selected target variables (e.g. accom3, accom5, heat5a1_2, accom9a1_2, renov2_*) across all waves in wide format. accom9a1_2 and renov2_* are structural equivalents asked in different waves; for any given row one set is populated and the other is NaN.

Car history is built via build_car_history in utils.py (Python) and equivalently in utils.R / longitudinal_exploration.rmd (R):

• Returning car owners are only asked for their fuel type (mob3_3) if they changed their car (mob3_change == 1). Otherwise the previous answer is carried forward into mob3_3_filled.

pip install pandas numpy pyreadstat matplotlib seaborn

install.packages(c("haven", "tidyverse", "zoo", "scales", "yaml", "skimr", "corrplot", "ggthemes", "kableExtra", "extrafont", "patchwork", "forcats"))

Saving plots as PDF/EPS requires Cairo. Install the system library for your OS first, then install.packages("Cairo") in R:

• macOS: install XQuartz, then brew install cairo
• Linux (Debian/Ubuntu): sudo apt install libcairo2-dev libxt-dev
• Windows: Cairo is bundled with R — no extra steps needed.

Before running any scripts, create config.yaml and set the paths for your machine. Example:

paths:
  data_dir: "/path/to/your/sheds_data/"   # folder containing the .sav files
  plots_dir: "plots"                       # output folder for saved figures
  questions_csv: "sheds_questions_up2025.csv"
  geodata_bezirke: ""                      # path to BFS Bezirke shapefile

sheds_files:
  "2016": "SHEDS2016.sav"
  "2017": "SHEDS2017.sav"
  "2018": "SHEDS2018.sav"
  "2019": "SHEDS2019.sav"
  "2020": "SHEDS2020.sav"
  "2021": "SHEDS2021.sav"
  "2023": "SHEDS2023.sav"
  "2025": "SHEDS2025.sav"

• data_dir — absolute path to the directory where your SHEDS .sav files are stored. All scripts read data from here.
• sheds_files — maps each wave year to its filename inside data_dir. Remove entries for waves you do not have.
• plots_dir — relative or absolute path where figures are saved by save_plot(). Created automatically if it does not exist.
• geodata_bezirke — only needed for the region distribution map. Leave empty to auto-download.

Note: config.yaml contains your local paths and is listed in .gitignore — do not commit it.

For the reading the file library pyreadstat has to be first installed. Then the the file can be loaded with metadata:

import pyreadstat

# Load with metadata
df, meta = pyreadstat.read_sav("/path/to/SHEDS2025.sav", encoding="UTF-8")

# Get variable label (question text)
meta.column_names_to_labels['accom11_1']
# -> "How satisfied are you with your current heating system?"

# Get value labels (response options)
meta.variable_value_labels['accom11_1']
# -> {1: 'Very dissatisfied', 2: 'Dissatisfied', ..., 5: 'Very satisfied'}

# Apply labels to create readable values
df['accom11_1_label'] = df['accom11_1'].map(meta.variable_value_labels['accom11_1'])

library(haven)

sheds <- read_sav("/path/to/SHEDS2025.sav")

# Get variable label
attr(sheds$accom11_1, "label")

# Get value labels
attr(sheds$accom11_1, "labels")

# Apply labels
library(dplyr)
sheds %>%
  mutate(accom11_1_label = as_factor(accom11_1))

If you use this repository, please cite the following paper:

@TechReport{repec:irn:wpaper:26-05,
    type={IRENE Working Papers},
    institution={IRENE Institute of Economic Research},
    author={Amy Liffey and Rene Schumann and Sylvain Weber and Mehdi Farsi},
    title={The Swiss Household Energy Demand Survey: Panel updates and evidence after eight waves},
    year={2026},
    month={Mar},
    number={26-05},
    abstract={This paper serves as a comprehensive reference for users of the Swiss Household Energy Demand Survey (SHEDS). SHEDS was conducted annually from 2016 to 2021, followed by two waves in 2023 and 2025; two further waves are scheduled in 2027 and 2029. So far, it provides a panel dataset of eight waves that span a 9-year period. We present recent updates to the panel, including data processing and data quality guidelines, and provide an overview of its evolving structure. In addition, we provide two applications that illustrate the research possibilities enabled by SHEDS. We use the first application to provide a description of how the data can be used to inform an agent-based model. The second application is a descriptive study of the evolution of car ownership patterns showcasing the potential of the survey's longitudinal dimension.},
    keywords={},
    doi={None},
    url={https://ideas.repec.org/p/irn/wpaper/26-05.html},
}