data/03_data

Data for the HMC (Human Machine Communication) project

Variables

An overview of all variables on item level can be found in the item reference and in the EXCEL codebook. These files show which variables have been collected in each wave.

Folder and file organisation

Folders

• 01_raw_data contains the downloaded files from Qualtrics
• 02_anonymized_data contains the anonymized data files (otherwise this can still be considered raw data)
• 03_cleaned contains data files with harmonized data names; additionally some incorrect variable names were fixed and double entries from subjects who did a wave two or more times were removed; see cleaning.R and below for more details
• 04_preprocessed_data contains the preprocessed data files with scales created from the items; see database_api_reference.md for a description of scales.

Files

• HMC_codebook.xlsx contains all variable names for all waves, with the original descriptions as presented in Qualtrics and the original variable names as well as the harmonized variable names
• HMC_variables.xlsx contains an overview of the variables, their origin, who wanted them in the data, etc. This file is for internal use and is not commited with the public version

Data collection and data files

The data collection was done in Qualtrics. The following projects are on https://kmrc.qualtrics.com/:

• AI_Trends_Wave1
• AI_Trends_Wave2
• AI_Trends_Wave3
• AI_Trends_Wave4
• AI_Trends_Wave4_sample2
• AI_Trends_Wave5
• AI_Trends_Wave5_sample2
• AI_Trends_Wave6
• AI_Trends_Wave6_sample2

Sample

Sample 2 data files

After wave 3, we re-invited wave-1 participants for waves 4–6 to increase statistical power for questions that did not require participation in all six waves. This departed from our original plan to invite only participants from the immediately preceding wave because ongoing monitoring showed that many non-users remained non-users and that relatively few participants perceived AI as a social actor. To capture more contemporary usage and obtain sufficient variation for research questions filtering for individuals that perceived AI as a social actor, we broadened recruitment in wave 4 to all wave-1 participants.

Download settings in Qualtrics

The data were downloaded from Qualtrics as CSV files with the following settings.

Overall

• Download all fields
• Export values

CSV

• Recode seen but unanswered questions as -99
• Recode seen but unanswered multi-value fields as 0
• Split multi-value fields into columns

Data anonymization

After download from Qualtrics, files were put in the respective folders for each wave in 03_data/01_raw_data/wave*. The script 03_data/01_raw_data/anonymization.R mostly removes the PROLIFIC_IDs from the data and adds an anonymized ID subj_id with entries subj0001 - sub1009 to all data sets.

Irrelevant columns - mostly automatically created by Qualtrics - are also removed. See anonymization.R for details.

The anonymized data files are saved to 03_data/02_anonymized_data/ as CSV files with file names HMC_<wave>_anonymized.csv.

Data cleaning

After data anonymization, some more rudimentary data cleaning was done with the script 03_data/02_anonymized_data/cleaning.R. Especially, the original variable names in Qualtrics were harmonized so they all follow the same structure.

The cleaned data files are saved to 03_data/03_cleaned_data/ as CSV files with file names HMC_<wave>_cleaned.csv.

The following section gives an overview of the problems in the data, that needed some cleaning.

Problems

with variable names

• For the variables looking at what tasks subjects would delegate to AI, there were some inconsistencies in the naming. This was only in the variable naming, the items were presented correctly to the subjects. The folloing variables were renamed:

  • delg_tsk_typs_4 --> delg_tsk_typs_3
  • delg_tsk_typs_5 --> delg_tsk_typs_4
  • delg_tsk_typs_6 --> delg_tsk_typs_5
  • delg_tsk_typs_7 --> delg_tsk_typs_6
  • delg_tsk_typs_8 --> delg_tsk_typs_7
  • delg_tsk_typs_8 was deleted

• The labels of the intention variables were swapped by accident and this was corrected:

  • int_use_bhvr_fav = int_use_bhvr_noUser and vice versa

with subjects

• Two entries in wave 1: subj0762
• Three entries in wave 3: subj1009
• We kept the first entry for each subject
• subj1009 has been removed from the dataset since it only appeared in wave 3 and it is unclear how this happened; only subjects who participated in wave 1 have been invited to participate in further waves

Data preprocessing

The final data preprocessing creates scales from the collected items. It was done in Python, and the code for the preprocessing can be found in a separate code repository: https://gitea.iwm-tuebingen.de/HMC/preprocessing. The files with the final variables for each scale are then saved in the folder 03_data/04_preprocessed_data. Three versions are provided: csv, and excel versions per wave as well as an overall sqlite database containing all waves in one file. database_api_reference.md contains the documentation of the database.