Data Analysis
Download as PPT, PDF31 likes3,744 views
The document outlines the practices and requirements of data analysis, emphasizing the need for familiarity with analysis techniques, tools, and computing resources based on data scale and complexity. It discusses various types of data analysis, including exploratory, confirmatory, and qualitative analyses, along with the design, preparation, and revision of analyses. Additionally, it highlights the significance of metadata and proper documentation in the data analysis process and the importance of effective reporting of results.
Data Analysis
- 1. Data Analysis Andrea Wiggins IST 400/600 April 14, 2008
- 2. Data Analysis Data are collected, created, and kept for the purpose of analysis Without analysis, it’s just a bunch of bits Data managers need familiarity with analysis practices http://flickr.com/photos/techne/100055322/
- 3. Overview Types of data analysis Requirements for analysis Basic steps in data analysis Types of tools Scientific analysis workflows Types of analysis output
- 4. Requirements for Analysis Data Analysis design Analysis tools Computing resources to run the analysis Human expertise http://www.flickr.com/photos/anikarenina/369089979/
- 5. Computational Resources The computing resources required for analysis will depend on scale and complexity Scale refers to the the data Complexity refers to the analysis
- 6. Scale of Analysis Physical locations of the data and the analysis machines Communication networks Volume of data, number of data streams Time to complete reflects size of data and complexity of analysis Hours or days may be required
- 7. Complexity Both analytic and computational complexity are relevant Some operations are “cheap” and others are “expensive” Number of calculations required - every function is made of other functions Execution in serial versus parallel processing: how many tasks at once?
- 8. Serial & Parallel Processes Serial Parallel
- 9. Small Scale Computing Regular microcomputers like your laptop Ordinary consumer PCs are able to do some significant computational work http://www.flickr.com/photos/cayusa/431036565/
- 10. Moderate Scale Computing Relatively small, locally-managed clusters Google’s smallest cluster: 13 servers Reservoir Simulation Joint Industry Project’s cluster -> http://www.cpge.utexas.edu/rsjip/
- 11. Macro Scale Computing High performance and grid computing NYSGrid TeraGrid NEESGrid Etc. http://ajlopez.wordpress.com/2007/12/03/grid-computing-programming/
- 12. Types of Data Analysis Two primary types for quantitative data EDA: exploratory data analysis CDA: confirmatory data analysis Third type for non-numerical data QDA: qualitative data analysis Photographs, words, observations Traditionally found in social sciences
- 13. Confirmatory Data Analysis Uses statistical tests to confirm or falsify hypotheses You know what you’re looking for Analysis is usually carefully planned in advance http://flickr.com/photos/activitystory/105110622/
- 14. Exploratory Data Analysis Methods used for data mining Nontrivial knowledge discovery from data Looking at data to form hypotheses for CDA testing (on a different data set) Don’t always know what you’re looking for, analysis evolves over time Caution: sometimes you find what you’re looking for, even if it isn’t there!
- 15. Qualitative Data Analysis Most common in social sciences, where data sets are usually smaller Uses a variety of methods to analyze non-numerical data Many qualitative analysis methods are difficult or impossible to automate http://flickr.com/photos/valix/939388335/
- 16. Context of Analysis Scientific inquiry Business intelligence Monitoring Carefully planned regular reporting As-needed ad hoc analysis http://flickr.com/photos/makou0629/1145908929/
- 17. Data Analysis: Design Examine (some of) the raw data Especially important with data meshing, when multiple different data sources are used together Design the analysis & instantiate it Test existing hypotheses Explore data to form hypotheses Use databases & analysis tools
- 18. Data Analysis: Prepare Data Clean the data - preprocessing Remove “noise” Sample the data Select the portion to use for analysis Validate the analysis Use a subsample to check your analysis Do the results make sense? Can you check intermediate values?
- 19. Data Analysis: Revise & Run Revise the analysis & test again Also known as debugging Good idea to compare manually and automatically computed results when possible to verify that everything works Repeat as needed Run the full analysis when ready
- 20. Data Analysis: Save Output Save/export the analysis results, artifacts, and appropriate metadata Data selection criteria, sample, analysis design version When analysis was run, by whom System details Time to run, exceptions Other relevant details dictated by your context of inquiry
- 21. Data Analysis: Use Write up the results Often requires returning to the raw data, analyzed data, and other information about the analysis Questions always arise… Something looks out of place, doesn’t make sense, can’t possibly be true Double-check everything: results, analysis records, analysis metadata
- 22. Very Important Details Data formats Format/s of raw data in source/s Format/s required for analysis Format/s of outputs: image, csv, statistics, descriptive text, etc. Data manipulation Moving from source to analysis to usable results, without losing/abusing anything
- 23. Data Analysis Tools Vary by preferences, skills, demands of the data Custom solutions Collections of small modular scripts Customized vendor software installations Stats packages SPSS, SAS, R eScience tools http://flickr.com/photos/alistairmcmillan/2102898220/
- 24. Open Science Movement Not just open data, also open analysis: Open Notebook Science Closed Open Traditional Lab Notebook (unpublished) Traditional Journal Article Open Access Journal Article Open Notebook Science (full transparency) From Jean-Claude Bradley in Nature Proceedings : doi:10.1038/npre.2007.39.1 : Posted 11 Jun 2007
- 25. Analysis Workflows Scientists “need access to tools and services that help ensure that metadata are automatically captured or created in real-time” - Cyberinfrastructure Vision for 21st Century Discovery Taverna Workbench demo video Example of a scientific workflow analysis tool, used for genetics - and social science! http://floss.syr.edu/Presentations/TavernaDemoRedux.m4v
- 26. Analysis Outputs Most analysis starts as numbers and ends up as words Scholarly articles White papers Technical reports Visualizations More on Wednesday http://flickr.com/photos/thiru/278930492/
- 27. Dashboard Reports At-a-glance reports for regular, ongoing monitoring Uses many visualizations Usually intended for managers & executives http://flickr.com/photos/jauladeardilla/345883088/
- 28. Concluding Thoughts Understanding how data is used will help you manage it better Planning ahead makes data analysis go more smoothly Data analysis almost never goes perfectly Analysis is the fun part of research, when discoveries are made
- 29. Questions for Discussion What can data managers contribute to data analysis? What are some of the factors that are relevant to designing data analysis? How is metadata relevant to designing data analysis? How is metadata relevant to reporting data analysis results?