Introduction to Statistical Analysis Yale Braunstein School of Information Management & Systems
Additional items related to MGO Mission statements in the corporate world A comparison of Coke & Pepsi Refining objectives Can ALWAYS be measured Specify date by which it will be accomplished Therefore a problem can arise when deadlines depend on each other. Example: Release Version 2.0 after completing beta testing AND complete beta tests prior to release of Version 2.0
Introduction The general idea is to summarize and analyze data so that it is useful for decision-making We do this by calculating “measures of central tendency” and by looking for relationships (We will NOT cover formal tests of hypotheses) Primary vs. secondary data sources Data on uses (system) vs. data on users (people)
Data Data may be continuous or discrete Just looking at the data often does not enable one to ascertain what is actually happening Solution: Use appropriate descriptive statistics to summarize and present results Another Data
Analysis--Introduction The BIG Question: What are you trying to discover or show? From survey to report Flow of information Sample survey of California ISPs Brief comparison of Excel & SPSS
Data Collection Instruments Questionnaires & surveys Transactions logs Bills & invoices Census forms & reports Pre-packaged data sets
Issues in Research Design Case study vs. statistical sample What is the universe ? (uses, users, etc.) Current political debate over “average tax cut” vs. “tax cut for the average family” Is the sample representative ? Volumes vs. titles in the library Does correlation imply causality? Do we need to identify the pathogen? Controlling for outside factors
Sample Size How large a sample is needed? The larger the sample the more accurate the results (unless the response rate becomes very low) The larger the sample the more the cost/effort Sample size does NOT depend on the size of the population Rules of thumb 100 for 95% confidence, 5% tolerance, 90-10 expected split 400 for 95% confidence, 5% tolerance, 50-50 expected split 30 – 50 in each cell on n x m discrete classes Exact formula (use with care): Size = 0.25 * (certainty factor/acceptable error)^2 Where the certainty factor = 1.96 for 95%; 2.576 for 99%
Sources of Error The respondent The investigator Sampling error Change in the system itself Coding & analysis Other