WPC 300 – Mid Term Exam Study Guide
Discuss what makes analytics science
oDefining characteristics of science: questioning, exploring, observing, comparing, measuring,
experimenting, discovering
oFacts (evidence): update beliefs with new evidence, need it to accept/reject a hypothesis,
supports hypothesis’/theories
oFalsification: science is a testable and unfalsified hypothesis, ideas can only be tested this
way, irrefutable theories not scientific (evidence can be used to rule out ideas, not support
them) accept hypothesis being true or reject it being false
oParsimony: simple explanations over complex ones
oWhat Makes Analytics Science?: you are using experimentation to test a hypothesis
Different analytical modeling techniques, be able to understand what is the best analytical
technique for different business scenarios.
oDescriptive: Allows one to summarize data, grab essential features of data (mean, mode,
median, max, min, standard deviation) what do I have, look into the past
oPredictive: What is likely to happen by how much and what are the expected costs, look into
the future, used for liner regression and data mining and simulation. What do we expect to
happen
oExplanatory: This causes that, used for trying to understand a casual relationship (t tests,
regression analysis, correlation) ex. Does smoking cause cancer. What is going on
oPrescriptive: recommended actions and timing, possible automation of actions, used in liner
and integer optimization
Be able to design an experiment given a scenario
oGolden standard: double blind randomized experiment, when experimenter knows which
group is the control
oWhat does randomization mean? The best defense against bias, ppl given a fair and random
chance at selection. (An equal opportunity to be placed in every group)
oControl group: Group that doesn’t receive the experimental treatment
oTreatment group: Group that receives treatment
oOutcome measures: comparison of your data and measuring the success, measuring the thing
you are trying to make a difference to, the dependent variable
oA/B testing: Method of comparing two website/apps against each other to determine which
one performs better. Two variants of a page are shown to users at random and statistical
analysis determines which one preforms better for the goal
Given a scenario, discuss to experiment or to observe.
oExplain limitations of observations: Confounding/Spurious Correlation (there is a third
unknown factor that is contributing to the correlation)
-EX: Breastfeeding linked to higher IQ of infants, but could be due to moms who
breastfeed are better educated. Tomatoes source of salmonella, but could be due to what
tomatoes are eaten with which are the source of salmonella
oExplain strength of observations: You can examine LT effects, experiments sometimes are not
ethical or possible, can do casual estimations, promise of big data, don’t intervene just watch
oExplain limitations of experiments: Sometimes not ethical or possible
oExplain strength of experiments: Easy to establish casual inference, golden standard, analysis
simple and straight forward, design it and intervene, less chance for bias/confounding
Understand sampling from population
oWhy sampling: Feasibility. Testing a population is expensive and a lot of effort
oSampling methods: random, convenience (easy but doesn’t represent whole population),
stratified
oSources of sampling biases: When random sampling is violated
-Convenience sampling (sampling based off convenience, not randomization)
-Self-selection bias (online polls) depends on who visits site and sees it and who actually
want to respond
-Social desirability, non-responsive
oBe able to discuss potential sampling biases given a scenario:
-Self-selection bias: In online polls, polls are only done by visitors to the site; Non-response
(another type of self-selection bias) is the fact that some people are less likely to respond to a
survey example weed smokers more likely to offer their opinion on weed because they are
more interested in the study
Understand group comparisons
oUnderstand the concept of statistical significance: A difference not caused by chance, there is
a specific cause!!! Results are real, important to analyze the data!!
oT Value: used to determine whether the means of two samples are statistically different from
each other. When you want to compare the means of two samples
-If 2 samples are taken from the same population then they should have similar means
-If 2 means are statistically different then they are likely from two different populations
oBe able to compute t values: take mean from each sample, find the difference, divide by their
shared standard error
Then compare to the distribution and come up with a p value
oP Value: Probability you can accept or reject the null hypothesis, if you reject it doesn’t mean
you can automatically accept the alternative
Cluster analysis
oWhat is cluster analysis, what is it for?: Grouping data so that elements in a group will be
similar to one another and different from elements in other groups, Used for descriptive
reasons
oHow does k-means algorithm work: A way of partitional clustering
-split them into k number of groups, select initial centroids, assigning points to clusters
based on distance, recalculate centroids and reassign points, keep doing until you settle
on final set of clusters
oUnderstand Euclidean distance: the straight-line distance between two points (metric used for
clustering)
oCompute Euclidean distance: difference between a general point and the average, square
other two sides to get distance
oCompute SSE, explain cohesion and separation:
-cohesion in a cluster everything should be as close as possible to the centroid
-separation in a cluster everything should be as far as possible from the centroid
Powered by TCPDF (www.tcpdf.org)
