Issues in data analysis

Issues in data analysis

1. trust

   • trust the data itself
     • origins,documentation, security, curation, quality of maintenece
   • trust in the processing of the data
     • has code been verified?
     • have appropriate, proven methods been applied?
     • How transparently have the results been reported?

2. data quality and fitness for purpose

   • Accuracy
     • Is the data captured, cleaned and stored in such a way that it still represents the original ‘thing’ it was intended to represent?
   • Validity
     • How well do stored data values agree with an external repository of ‘correct’ or reference data. (For example, do a person’s stored National Insurance number,
   • Provenance
     • Are the sources of the original data, and the individuals and tools responsible for updating and entering new data, known?

3. bias

   • Human bias
     • in human decision making favouring individuals or outcomes
   • Bias in data capture
     • ie when a dataset captured for one purpose is used for another
   • Bias in data cleaning
     • removing ‘incorrect’ data elements during data cleaning and preparation
   • Bias in data handling
     • Software for processing data must be fit for purpose and the analyst must be familiar with it

4. completeness and correctness

   • ie how is a null value handled?

5. reproducibility and provenance.

   • scientific protocol demands that analysts can track and record transformations to datasets
   • Benefit of using IPython
   • Provence shows how data is tracked