QM Flashcards
What is the QM323 Decision making framework?
define objectives/choices, analyze data/model, define risks & perform sensitivity analysis, and communicate results
What are the four decision making traps?
availability heuristic, confirmation trap, overconfidence, and overfitting
What are word clouds?
they allow you to quickly summarize data from large amounts of text. they break a piece of text into individual words & count the frequency with which they are observed
What does QM incorporate?
an understanding of risk & uncertainty
What are they key steps to creating a good work book?
plan out the structure, build the spreadsheet, review, and communicate your findings
What is an influence chart?
helps draw sketch of workbook. shows the outcome variable the model will generate and how outputs are calculated from inputs
When making Excel, should you isolate the parameters?
Yes, put parameters at the top, followed by decisions, outputs/objectives, and following calculations
What are the three ways to communicate results?
spreadsheets, visualizations, and dashboards
What are trace precedents in Excel?
They help understand the relationship between cells and which cells were used as inputs for a formula in another cell. Use formula auditing function
What are trace dependents?
they help understand which cells are affected by the value in a specific cell. They track how one cell impacts another
How to design structure of workbook?
define objective, sketch spreadsheet, organize spreadsheet into modules, start small, then isolate input parameters
What are the different variables for an influence chart?
decision variable, objective, intermediate quantities, and input parameter
What elements can be preceded by others in an influence chart?
intermediate calculations and outputs
What are major tasks in launching a product?
identifying a segment, targeting consumers, & positioning
How do you identify a segment?
You need to use cluster analysis
What is k-means clustering?
It helps assign customers to given segments using information on characteristics
Why do we use z score when doing k-means clustering?
Because the euclidean distance can be skewed cause of variables being measured on different scales
How to calculate z score?
Value - Mean/Stdev
What are the different types of analysis?
descriptive (describing observations), predictive (using a model from past data to make a prediction), and prescriptive (helps us identify the best course of action)
What are the steps for k means clustering?
Randomly put customers in clusters, find z-score, calculate the cluster centroids, and calculate distance from z score to given cluster centroid & reassign
How to find z-score with Excel?
=standardize(x,mean,stdev)
What are cluster centroids?
they are the average values of the z scores for each variable of each cluster(coordinates), need to rearrange data for this
How do you calculate distance from z-score to given cluster?
=sqrt((x1 - x2)^2 + (x3-x4)^2 + …))
How do you reassign?
Whichever cluster the observation is closer to is the one you reassign it to
How do you normalize data on XL miner?
you have to click standardization
What is inter cluster distance?
The euclidean distance between cluster centroids (how far you are from another cluster)
What is intra cluster distance?
average distance between points within the cluster. lets you see how homogenous each cluster is
What happens with clusters when you plot them on a scatter plot?
You usually only plot two variables, so it won’t show the full pictureH
How do you decide how many clusters to use?
Use trial and error to see which you make more conclusions from
How do you decide to group clusters together?
Look at the inter cluster distance & group the ones with the smallest distance
Does k-means clustering tell you which variable is most valuable?
No, you use the clusters you form to make segments which can then be used for conjoint analysis
What is conjoint analysis beneficial for?
it helps you better tailor your product to the customer
What are products and services considered?
They are a “bundle” of attributes. You need to decide which one is most valuable to you
What helps identify attributes most important to consumers?
HOQ
What are the steps in conjoint analysis?
identify a set of relevant attributes (needs to be something you can quantify), define reasonable levels for attributes (generally levels should be mutually exclusive and need to be greater than 2 per attribute), create product profile (hypothetical products/services described as bundles of attributes), obtain consumer preferences for profiles, and then analyze the data
What is an example of an attribute?
Material, foam thickness, color
What is an example of a level?
Plastic or wood for material attribute
What is an example of a product profile?
An american car with a sedan body type and gasoline engine priced at 20,000 (4 dif attributes)
What do you use to analyze conjoint analysis?
Use regression analysis, use one less variables
What is the range of an attribute?
the max partworth (intercept) - min partworth. make sure you consider baseline options
What is the formula for the importance of an attribute?
Range/(sum of range across all attributes)
What is partworth?
the regression coefficient
Do you use average when analyzing what target customers value?
No
What are the 3 methods to analyze what the target consumer values?
standard deviation, ABC analysis, and compare relationships across attributes
What does strong deviation mean?
It shows customer cares deeply about attribute because largely varied. If low all customer prefer the same thing
What is ABC analysis?
It is used only for specific bundles & gives you the market share of each product. If you are hesitant on design, use ABC to see market share (pick the one with the highest value)
What are the steps of ABC analysis?
Find value of each option, find market share, and analyze
What are association rules?
AKA market basket & affinity analysis. Asks what goes with what. Uses an if/then statement
What is an item set?
collection of items in store (basket of items)
What is an antecedent?
item corresponding to the if portion of the rule
What is a consequent?
item corresponding to the then portion of the rule
What are two approaches to targeting customers?
one to one target marketing or using predictive regression modeling (Use multiple variables to predict probability of purchase and identify best customers to target)
What is unsupervised Learning?
Can’t let the model do the work on own to discover information , can find all kinds of unknown patterns in data , can be unpredictable
What is supervised learning?
Produce an output from previous experience , Helps you to optimize performance criteria using experience
When making a dummy variable for purchase intention, what should you do?
Code definitely + probably will buy as 1 and the rest as 0
What does sum of purchdum show in pivot table?
tells you for each row, how many were purchased (ex: if male is a variable, tells you how many men purchased)
What does the grand total under sum of purchdum show in pivot table?
the total amount of people that purchased the items
What does count of purchdum show?
just tells you the count of the variable (ex: how many males in data set if that’s the variable in the row)
What does grand total in count of purchdum show?
the total number of transactions
What are pitfalls of association rules?
They are descriptive, not causal, evidence, use randomized experiments to validate potential rules & need to sift through a lot of rules before finding useful
Beware of confirmation bias
How to automatically calculate regression predictions?
When doing regression, check off residuals
What is a residual?
Actual - Predicted
Why would groups of people have the same predicted probabilities?
Because they might have the same values for the X variables
What does predicted purchase probability help you find?
Helps you find the best group to target
How do you find the cumulative purchdum when dealing with residuals?
Do actual purch dum + previous actual purchdum
How to find % of cumulative purchdum?
(cumpurchdum)/total purchasers
What is the point of doing the cumulative purchdum & %?
It helps you figure out how much a group of people will generate probability of potential sales (ex: looking at top 11 people and looking at % next to it, shows top 11 people will generate 15% probability of sales)
What is breakeven response rate for?
The % given helps you see what predicted probability of purchase you should target (ex: if rate is 10%, should target ppl with purch prob of 10% or higher)
What is support?
the amount of times you see antecedent or consequent in set
With ABC analysis, what are you doing?
You are deciding you will offer either A, B, or C (bundle of attributes) to the market
What are the downfalls of ABC?
Not establishing the best of ALL your options, just the best of the 3 options you are looking at
How do you calculate value for each