Using Machine Learning to Optimize the Quality of Survey Data: Protocol for a Use Case in India
Bibliographic record
Abstract
BACKGROUND: Data quality is vital for ensuring the accuracy, reliability, and validity of survey findings. Strategies for ensuring survey data quality have traditionally used quality assurance procedures. Data analytics is an increasingly vital part of survey quality assurance, particularly in light of the increasing use of tablets and other electronic tools, which enable rapid, if not real-time, data access. Routine data analytics are most often concerned with outlier analyses that monitor a series of data quality indicators, including response rates, missing data, and reliability of coefficients for test-retest interviews. Machine learning is emerging as a possible tool for enhancing real-time data monitoring by identifying trends in the data collection, which could compromise quality. OBJECTIVE: This study aimed to describe methods for the quality assessment of a household survey using both traditional methods as well as machine learning analytics. METHODS: In the Kilkari impact evaluation's end-line survey amongst postpartum women (n=5095) in Madhya Pradesh, India, we plan to use both traditional and machine learning-based quality assurance procedures to improve the quality of survey data captured on maternal and child health knowledge, care-seeking, and practices. The quality assurance strategy aims to identify biases and other impediments to data quality and includes seven main components: (1) tool development, (2) enumerator recruitment and training, (3) field coordination, (4) field monitoring, (5) data analytics, (6) feedback loops for decision making, and (7) outcomes assessment. Analyses will include basic descriptive and outlier analyses using machine learning algorithms, which will involve creating features from time-stamps, "don't know" rates, and skip rates. We will also obtain labeled data from self-filled surveys, and build models using k-folds cross-validation on a training data set using both supervised and unsupervised learning algorithms. Based on these models, results will be fed back to the field through various feedback loops. RESULTS: Data collection began in late October 2019 and will span through March 2020. We expect to submit quality assurance results by August 2020. CONCLUSIONS: Machine learning is underutilized as a tool to improve survey data quality in low resource settings. Study findings are anticipated to improve the overall quality of Kilkari survey data and, in turn, enhance the robustness of the impact evaluation. More broadly, the proposed quality assurance approach has implications for data capture applications used for special surveys as well as in the routine collection of health information by health workers. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/17619.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.072 | 0.082 |
| Meta-epidemiology (narrow) | 0.001 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.003 | 0.003 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.044 | 0.008 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".