Target Scale Degree and Accuracy in Cued Two-Note Sight-Singing: Data and Code

Description

Data and reproducible code for “Target Scale Degree and Accuracy in Cued Two-Note Sight-Singing” (RECIEM). The main dataset contains 6,656 observed binary scores from 104 music students completing 64 staff-notated two-note items in C major (C4–C5), without notated rhythm or a time limit. Each item followed a C4–E4–G4 tonic chord and a supplied starting pitch; item order was fixed. Correct = 1 only when both starting and target pitches were correct on their final renditions; spontaneous corrections were allowed. Scoring was auditory. A separate preliminary sample (n = 57) is not included. Forty randomly selected main-study recordings were independently rescored: 99.29% agreement over 2,240 paired decisions (Cohen’s κ = .986, 95% participant-bootstrap CI [.977, .993]). Original main-study scores were retained. Primary logistic mixed-effects models use 56 non-unison items (5,824 responses) to assess categorical target degree, interval category, and melodic direction. A post hoc 49-item comparison (5,096 responses) examines tonic-chord targets versus other diatonic targets. In this single-key, one-octave design, degree also identifies exact pitch/register. Sensitivity checks cover C4 exclusion, item order, interval quality, expanded random effects, and starting degree. A further post hoc comparison uses 11 cells containing both directions (22 items; 2,288 responses); starting pitches are not matched. Analyses are exploratory where indicated and were not preregistered. The ZIP contains 17 files: three unchanged observed-data inputs, twelve R/Python scripts, a README, and checksums. Six analysis workflows, agreement statistics, and figures can be regenerated; outputs are produced under rerun/. No responses were imputed or altered. Findings are task-specific associations, not evidence of a causal direction effect or an isolated tonal-hierarchy mechanism.

Steps to reproduce

1. Extract Target_Scale_Degree_N104_Data_and_Code_RECIEM.zip and open a terminal in the MP104 folder. Install R packages lme4, DHARMa, numDeriv and Python packages numpy, pandas, matplotlib, scikit-learn (Python 3.9+). See README.md for tested R/package versions and full specifications. 2. Audit inputs and create analysis subsets (no observed scores are changed): `python code/prepare_data.py` 3. Run the primary model stages in order: `Rscript code/run_models.R primary audit .` `Rscript code/run_models.R primary fit .` `Rscript code/run_models.R primary profile .` `Rscript code/run_models.R primary diagnostics .` Repeat these stages using order, random, quality, or focused49 instead of primary. For quality, an additional report stage is available. 4. Run the post hoc starting-degree and 22-item overlapping-cell analyses: `Rscript code/run_models.R posthoc56 audit .` `Rscript code/run_models.R posthoc56 fit .` `Rscript code/run_models.R posthoc56 diagnostics .` 5. Reproduce scoring agreement and figures: `python code/run_agreement.py` `python code/build_figures.py` Run primary and order profile stages before building Figure 3. Outputs appear in rerun/. Compare 56- and 49-item models descriptively, not by cross-sample LRT or AIC. Keep singularity warnings; consult README.md for limitations. The new posthoc56 script was integrated but not rerun during ZIP preparation because R was unavailable in that environment.

Institutions

• Atatürk University

  Erzurum, Erzurum

Categories

Funders

• Atatürk University BAP

Licence