Data Analysis
16.621
March 13, 2003
Data Analysis Process
Experiment
(Raw Data)
Data Reduction
(Reduced Data)
Assess
Hypothesis
Report
Findings
Data Interpretation
Graphical
Analysis
Compare
with theory
or other data
Incorporate
knowledge
Extract
findings
Conceptually
separate, but can be
iterations/feedback
between them
Keep the end goal in mind!
Experiment Step
? Experiment produces “raw data”
– In measured units (volts, seconds, lbs, ….)
– Organization determined by recording method
(notebook, tabular form, database,….)
? Take data early to check-out experiment
? Do some “quick data reduction/analysis” during
experiment to see if results look ok.
? Goal: Assure data set is complete and all
information is collected while experiment is still
set up.
Largely a recording process.
Data Reduction Step
? Convert and normalize or nondimensionalize
raw data to “meaningful variables”
(temperature, time of flight, force
coefficients,...)
? Statistical analysis as appropriate (next class)
? Error analysis as appropriate
? Goal: assure data is “valid” before
interpretation
Largely a deductive process.
Data Interpretation Step
? Interaction/Iteration of four sub elements
– Graphical analysis: helps visualize data to see trends, patterns,
relationships to theory,…
– Compare with theory or other data to determine agreement or
differences (both important)
– Incorporate knowledge learned in classes or from experience
– Extract findings which represent the knowledge generated from the
experiment
? Implementation tailored to each project.
? This step is where “value” is added by researcher
A highly inductive process!
Error Analysis - A General Approach
z During the design of the experiment
1 Identify all possible sources of error:
– Experiment set up: facility effects, environmental effects, human
subjects, …..
– Measurement system: velocity, temperature,...
2 Estimate possible severity of each source
– Discuss with advisor.
3 For those that are considered “important”, identify mitigation
strategies.
– Experimental design and/or test protocols (e.g. repeat tests)
4 Plan for quantitative analysis of reduced data
– Quantitative analysis relies on math model of the system
– Not possible for all situations: human factors tests, s/w studies
– Often good for measurement systems: pitot probe, strain gauge,...
? Sometimes quoted by manufacturer or supplier
Keep the end goal in mind!
Error Analysis - A General Approach II
? During the experiment
– Execute experiment according to protocols
– Record notes in lab notebook
– Check for mistakes
? During data reduction
– Calculate error bars for measurements
– Check for outlier points
? During data interpretation/reporting
– Consider errors when interpreting data
? Assure findings are beyond uncertainty of experiment
– Display error bars in way that aids in understanding findings
Goal: To “qualify” the accuracy of your data to support findings.
Exercise
? Make a sketch of a figure that you might put
in your final report which displays your
experimental data in a way that can be used
to assess your hypothesis.
? What questions do you have about data
reduction? For example
– Statistical analysis
– Graphical analysis
– Error analysis
? Turn in your sketch and questions.