Search
Browse
Create
Log in
Sign up
Log in
Sign up
Upgrade to remove ads
Only $2.99/month
EXAM 2 Analytics
STUDY
Flashcards
Learn
Write
Spell
Test
PLAY
Match
Gravity
Terms in this set (144)
correlation
a statistical measure that indicates the extent to which two factors vary together and thus how well one factor can be predicted from the other. Correlations can be positive or negative.
positive correlation
indicates the extent to which those variables increase or decrease in parallel-one goes up the other goes up-rains out, more umbrellas
negative correlation
indicates the extent to which on variable increases as the other decreases
1.0 correlation
perfect correlation
means that there is perfect positive correlation
-1.0 correlation
as one moves up, the other moves exactly in the opposite direction
-means that there is perfect negative correlation
0 value correlation
value indicates there is no (zero) relationship between the variables at all, they have nothing to do with each other
CAUSATION
BEWARE: CORRELATION DOES NOT NECESSARILY MEAN
: in a commonly studied medical case it showed that women taking hormone replacement therapy (HRT) had a lower-than-average incidence of coronary heart disease
This led doctors to propose that HRT was protective against CHD
But randomized trials showed that HRT caused a small but statistically significant risk of CHD
Re-analysis revealed that women who were taking HRT were members of a higher socio-economic group that ate healthier and exercised more regularly
Correlation is not a causation ex
Correlation is not a causation ex
definitely true; where there is causation, there is likely correlation (reversed)
So yes - correlation can definitely indicate causation but only after it is systematically explored using large enough data sets and research
So why use correlation in marketing analytics?
It is always correct to say "Correlation does not imply causation" in that it is not always certain
The idea that correlation and causation are connected is
The faster windmills are observed to rotate, the more wind there is. Therefore wind is caused by windmills rotating and windmills are machines to produce wind.
NOT TRUE, you have to understand the data to be logical
Higher temperatures produce higher umbrella sales during the summer. Therefore, warm weather causes consumers to purchase umbrellas.
NOT TRUE! Its the rain!
Is it "Children that watch a lot of TV are the most violent, therefore TV makes children more violent" OR "Violent children like watching more TV than less violent ones."
In the middle ages people believed that lice was beneficial because there was rarely any lice on sick people. But, in fact, lice are sensitive to temperature and avoid humans with fever
More examples:
A bald (or previous balding) state leader of Russia has succeeded a non-bald (hairy) one, and vice versa, for nearly 200 years
There is a high correlation between the number of films Nicolas Cage does in a year and the number of drownings by falling into a pool
The result of the last home game by the Washington NFL team prior to the presidential election predicted the outcome of every presidential election from 1936 to 2000. Despite the fact that football games have nothing to do with the outcome of an election.
Coincidental Correlation
causation
So what, in Marketing Analytics, do we do? Is correlation not reliable?Correlation IS real regardless of .
some causation that we can determine and measure...
Does our display ad result in higher sales conversions?
Are ads that appeal to middle-aged men giving us a higher click through rate?
Do competitive actions impact our unit sales?
Can we research and analyze buyer behavior?
But for Marketing analytical purposes we want to determine if the correlation reveals
...
Calculating Correlation
collect the data you want to examine for correlation. Obviously in this case it would be the daily temperature (let's refer to as data set "x") and the unit sales of gloves (let's refer to as data set "y").
Problem: Is there correlation between the daily temperature and the sale of gloves?
First -
calculate a coefficient of correlation and see the results
second
-After a sufficient amount of data is collected then
x y chart
temperature (x) and unit sales (y)
Then you present data:
take temp (x) MINUS the average temperature
then take units MINUS average unit sales
The calculation: step 1
a x b
then you take
square a
square b
after you do axb,
add up ALL of the a^2 and b^2 and axb values
lastly
the totals of (axb) then divide by the square root of all (the totals of the a^2 x b^2)
The calculation step 2
...
SEE THE SLIDES FOR MORE INFO
scatter plot
You can also use a ______ to visualize the data
correlation formula
Correlation: a statistical measure that indicates the extent to which two or more variables fluctuate together
Results
-A "0" value indicates there is zero correlation
-a "+1" indicates there is perfect positive correlation
-a "-1" indicates there is perfect negative correlation
Correlation is NOT Causation!
-It is not always certain to be causation
-There are many good examples of coincidental causation
-But causation should definitely have high correlation between variables
-As marketers we must strive to determine causation to better effect important activities such as buyer behavior, ad placement, targeting, etc.
Calculating and charting correlation will reveal the relationship between the variables
Summary
...
CHP 7 REGRESSION
regression
a form of predictive modeling (want to see how things will act in the future-investigates the relationship btwn a dependent variable and an independent variabled
-ex: does weather affect sales?
forecasting, time series, and determining causal effect relationship between variables
Regression is used for
independent variables using different scales
multi-regression analysis allows us to review the effect of several
"linear regression" calculation using marketing data
There are a number of different regression calculations. We will be using a
calculated marketing sales
The best way to calculate a regression is to start with
you take x (IV) variable squared and then take x*y
you do this for all of the observation data
then you total all of the y, x, x^2, and xy data
Calculating the regression:step 1
SEE REGRESSION LINE SLOPE AND FOR B INTERCEPT
Step 2: calculating the line slope
you use your new line that you just calculated
step 3: now we plot the results
...
we now have a formula that can help us forecast by plotting a line around these data points
we can predict new values
based on what we know with the regression forecasting
y intercept
the y-coordinate of a point where a graph crosses the y-axis
y = -4.959183673 (35) + 468.45
^plug in 35 and calculate new number
If the equation for the line slope is y = -4.959183673 x + 468.45, then what would be the forecasted unit sales if the temperature was 35 degrees?
...
this helps us forecast our sales based on what is happening with the independent variable
-for example, the temperature fluctuations determines how many glove sales we will have
P-values
work to tell us which relationships in our model are statistically significant and the nature of those relationships
-It describes the mathematical relationship between the independent variable(s) and the dependent variable.
p-value
The probability level which forms basis for deciding if results are statistically significant (not due to chance).
good estimates
statistical output
After running your regression model check your plots first to be sure that you have _______. After that you need to interpret the ______.
inferential statistics.
Regression analysis is a form of ______
the larger population.
The p-values help us determine whether the relationships that you observe in your sample also exist in
null hypotheses
The p-value for each independent variable test the
null hypothesis
that the variable has no correlation with the dependent variable.
no association between the changes in the independent variable and the shifts in the dependent variable.
In other words, there is insufficient evidence to conclude that there is effect at the population level.
If there is no correlation, there is
reject the null hypothesis for the entire population.
Your data favors the hypotheses that there IS a non-zero correlation - or there is "significance" - there is a significant correlation between the two
If the p-value for an independent variable is less than your significance level your sample data provides enough evidence to
...
if there is insufficient evidence, i get a bad p value
if there is sufficient evidence, i get a good p value
is insufficient evidence in your sample to conclude that a non-zero correlation exists.
On the other hand, a p-value that is greater than the significance level indicates that there
significance level
A critical probability associated with a statistical hypothesis test that indicates how likely an inference supporting a difference between an observed value and some statistical expectation is true. The acceptable level of Type I error.
multiple regression
multiple independent variables, trying to determine what affects the y value the most
...
In this example, the P-value results indicate there IS significance in one variable and there is a non-zero correlation value. The significant variable in this example is Temp (x1) with a P-Value of 0.00207215
The regression line (using the Coefficients) is y = "sales" = 396.395 + (-4.8087603 Temp) + (-.372685 Wind) + (20.6814615 * Price)
the coefficients show us the line
...
So what does that all mean with your work? The output below shows that the Temp predictor variable (independent) is statistically significant because its p-value equal 0.00207215.
the usual significance level of 0.05.
On the other hand, Wind and Price are not statistically significant because their p-values (0.84114198 and .41407215) are greater than
...
It is standard practice to use the coefficient p-values to decide whether to include variables in the final model. For the results above you would consider removing Wind and Price.
precision.
Keeping variables that are not statistically significant can reduce the model's
...
you can have more than one significant variables but you throw out the insignificant variables
Regression: a form of predictive modeling that investigates the relationship between a dependent variable (y) and an independent variable(s) (x)
-Used for forecasting, time series and causal effect relationship between variables
-Multi-regression analysis allows us to review the effect of several independent variables at once
-There are many different regression calculations
- we have reviewed Pearson's
-The trend line can be used for -simple forecasting
Summary
Results
A simple linear regression calculation (Pearson) will allow us to see a trend line with a slope that shows the relationship between our dependent variable (y) and independent variable(s) (x)
That slope formula is of the form y = mx + b
Slope (m) formula:(NΣxy − Σx Σy) N(Σx2) − (Σx)2
b intercept formula: Σy − m(Σx)N
Summary
...
CHP 8 RESOURCE ALLOCATION
Resource Allocation
ties together the various marketing analytics techniques to a firm's strategic decisions
resource allocation
Assigning available resources, or factors of production, to specific uses chosen among many possible and competing alternatives. It involves answering "What to produce" and "How to produce".
- Advertising in so many formats, Promotions, Merchandising, etc.
Marketing managers are often faced with the decision of the level of investment in different marketing activities
Resource allocation
is the endgame of analytics for any company. Using marketing analytics properly any organization should be able to determine the optimal level of spending it should make on each of its marketing channels to maximize success
dividing up our dollars to best maximize success
critical contributor to its customer relationship efforts and to its bottom line
We want to build a data-driven organization where marketing analytics is a
1. Determine the Objective Function - what are we trying to accomplish
2. Connect Marketing Inputs to the Objective
3. Estimate Relationships Identified in Step 2
4. Identify the Optimal Value of Marketing Inputs
Resource Allocation Process 4 Steps
...
Step 1 - Objective Function
Step 1 - Objective Function
What is the metric the company wants to set as its goal for optimization?
Conversion rates to sales?Incremental margins / profits?Customer Lifetime Value (CLV)?Near-term sales lift?New buyers?Market share?Retention rates?Future growth potential?
What is the metric the company wants to set as its goal for optimization?
fails to meet the organization's objective!
Without determining your objective first, you are at a high risk of developing a resource allocation model that
...
Step 2 - Connect Marketing Inputs to the Organization Objectives
Step 2 - Connect Marketing Inputs to the Organization Objectives
Intuition and experience are of huge importance in this step
It allows the marketer to correctly decompose a metric
connecting inputs to objectives
the attributes of the business that contribute to those profits?
What are the relationships between the various components within the business?
It allows the marketer to correctly decompose a metric
For example, if a company is examining gross profits (aka the objective), what are
Sales
for example, is a function of price, advertising, sales force, trade promotions, etc.how much we sell a year is dependent on our price, on our advertising, etc
sales DV
price, advertising, sales force, trade promotions, etc are IV
...
Because gross profits minus marketing yields net profits, manipulating marketing channels can improve sales. But marketing is also a cost center as well...
accounting identities (ID) vs. empirical (EM)
Once you connect the Marketing Inputs / Objectives you need to identify those relationships that are
Accounting relationships (ID)
are easily calculated and known.they are numbers on the financials
unit cost
unit margin
marketing $
for example somethings are easy from an accounting standpoint to measure...it is an accounting identity
empirical require a lot more work because i know there is a relationship between unit price, advertising, sales force, and trade promotion but i dont know which one of these effects unit sales the most...we dont know how much they effective, it is an empirical quantity.
If we add the cost up of those units, we do know the identity relationship (it is just the sum of how much those items cost)
RELATIONSHIPS between variables are what is difficult to quantify
Empirical, however, requires much more work. Why?
added up and equated to sales
The empirical relationships (Unit Price, Advertising, Sales Force, Trade Promotions) cannot be
marketing inputs into sales (for example, between price and sales, or advertising and sales, etc.)
Marketing managers must analyze historical data to develop a function that transforms the
It will require a "best guess" or prediction.
The results are not going to be certain between the empirical relationships and the objective.
are certain.
ID relationships, however,
multiple regression
we need to know what effects what in our model so we understand how to allocate our resources in the best way
we can do this with
Unit Margin = Unit Price - Unit Cost. So it has an ID relationship to Unit Margin as a quantifiable accounting identity (ID)
Price is a known quantity. If my price is $7.98 there is nothing empirical about that
when it comes to calculating "Unit Margin" - it is quantifiable.
not certain! We know it has some relationship / affect on sales but it's not exactly quantifiable. So we indicate that in our model with an EM relationship.
But price also has some affect on sales. If I price my product too high it will definitely influence sales (decrease) and the converse is true - if I price my product too low it will likely influence sales (increase). But how much price influences sales is
If I spend $1,000 on advertising, $40,000 on sales force and $15,000 on trade promotion my marketing spend will be, in fact, exactly $56,000. Hence an ID relationship.
The marketing spend on advertising, sales force and trade promotion are known quantities and have an ID relationship with our Marketing $.
That is not quantifiable yet. We know it has some affect but it's empirical. By empirical we mean this - we KNOW that marketing spend has an affect on unit sales but we cannot say exactly how much (EM). More importantly which one affects sales the most? Which one has no affect on sales? Again - think about your multiple regression homework.
On the other hand, how would those dollars affect "Unit Sales"?
decision items for marketers
Square boxes are ______ - what should we price at? How much should I spend on trade promotion? How many salespeople should I hire?
quantifiable results from my actions
The ovals are metrics - ________. I know, for fact, what my sales were this past month. I don't know for sure which of my decisions (pricing, advertising, trade spend, sales force, etc.) affected sales the most.
objective
The ______ here is "Net Profit" hence the rounded square box.
...
Step 3 - Estimate Relationships Identified (EM) in Step 2
econometric (regression) model
A common method for Estimating Relationships Identified (EM) in Step 2 is to build an
model to visually see which inputs have an affect on an output.
Which marketing inputs of interest (price? advertising? sales calls?) should be considered as having an affect on the dependent variable? That is the whole point of the previous slides and discussion. You have to first create a
predict the precise shape of an objective function
Once a regression model is obtained, the marketing manager can
the independent variables (price, advertising, sales calls) and the dependent variable (market share, profits, unit sales, customer lifetime value - CLV, etc.)
This mathematical model will describe the relationship between
...
Step 4 - Identify the Optimal Value of Marketing Inputs
optimal value and mix of marketing inputs (decision items) to maximize the objective function (net profit)
This gives a detailed picture of what the organizations precise marketing spend should be on each channel it uses to market its product
With a good econometric model based on regression, we can reverse the process and identify the
Resource allocation ties together the various marketing analytic techniques to an organization's strategic decisions (think pricing, trade promotions, sales staff, advertising)
The resource allocation process has 4 steps
Determine the objective function (Gross profit? Unit sales?)
Connect Marketing Inputs to the Objective
-ID relationships?
-EM relationships?
Estimate relationships identified in step 2
Using statistical analysis to build econometric models with statistical calculations like regression
Identify the Optimal value of Marketing inputs
Regression Summary
Marketing managers must understand their marketing efforts as precisely as possible to determine how much to spend on each marketing channel
-If paid search advertising is the most effective way of getting a firm's message in front of the right customer, why would we spend more on print advertising?
-If sales calls are profitable only up to a point, you must know at which point the calls start costing you money instead of making it
The only way to measure the effects of marketing efforts on profitability is through the relationships revealed through marketing analytics and resource allocation modeling
By using statistical analysis marketing managers can then optimize spending on each channel
Regression Summary
...
CHP 10 Product Analytics
Trend and forecast
using various techniques including moving averages, centered moving averages (monthly sales amounts over many years and removing the "noise" from the data)
multiple independent variables
Establish the significance of ______ (promotions, price, discounts, billboards) and their relationship with a dependent variable (sales) and the importance of experience and judgement to determine those quantifiable (ID) and qualitative (EM) effects of our marketing mix
significance of our marketing mix (promos, discounts, price, promotions, etc.) on sales
As marketers these techniques are extremely valuable for forecasting and making decisions on the
Product analytics
can help a manager with their product management decisions
conjoint analysis
Can be connected to resource allocation by demonstrating that stronger brands and better-quality products can charge a higher price without a discernable decrease in demand
conjoint analysis
Identification of product features that appeal to consumers to help a product manager determine the right balance between features and price
conjoint analysis
It seeks to determine how consumers value the different attributes that make up a product and the trade-offs they are willing to make among the different attributes or features that compose the product
Features and attributes
could also be described as "options" that come with a particular product.-adjectives that describe a noun
-the dimensions of a door
very tangible attributes that can be easily described or quantified (color for example)
Conjoint analysis is best suited for products that have
product analytics
There are a number of different _____ that can be used to look at these features (methods: k-means clustering, conjoint analysis, etc.)
...
It's popularity has grown tremendously over the last few years as easy to use software has allowed its widespread implementation
Examples:
Predicting the market share of a proposed new product given the current offerings of competitors
Predicting the impact of a new competitive product on the market share of any given product in that marketplace
Determining consumer's willingness to pay for a proposed new product
Quantifying the trade-offs customers or potential customers are willing to make among the various attributes or features that are under consideration in the new product design (what are consumers willing to pay for hard cover vs soft cover)
examples of use of conjoint analysis
features considered jointly
Literally, conjoint analysis means an analysis of
the features and observing consumers ratings for that product or it's choices
Although it is difficult for consumers to state directly how much each feature of a product is worth to them (hard cover vs. paperback, higher price vs. lower price, used vs. new, short delivery vs. long delivery, etc.) you can infer the value by experimenting with
Think about it though - you do conjoint analysis all the time but you can't always put numbers on your thinking
with conjoint analysis, i can see how much consumers value each of the different options and come up with quantifiable answers
1. Research Product
2. Design Experiment, select attributes and values
3. survey consumers
4. analyze results
Steps in the conjoint analysis
research product
what do you know about the company and product? What are the specific attributes?
the attribute is the "length" of a boat
the value is the actual numerical number of the 50 ft
attribute vs value
...
2. experimental design
attributes and the values of the attributes that will be tested
A conjoint analysis begins with an experimental design that includes all
it doesnt matter, you can have many values, the dont have to be numerical
They can have values (see above) or just be simply a "yes" or "no" - doesn't matter. And they can have many values - maybe "Price" has values of $23,000, $25,000, $27,000...
survey consumers on their preferences of these attributes.
The next step would be after we design this experiment, we then need to
The Utility value
- or part worth - corresponds to an average consumer preference for the level of any given attribute (basically which choices did they prefer over others). Within any attribute the utilities are scaled in a way that they add up to zero.For Horsepower, -2.24 + 1.06 + 1.08 = 0.
negative utility; it means that this level is on average LESS preferred than a level with an estimated utility that is positive (ex: higher prices have negative utility values)
Also, a negative number does NOT mean a ____ utility
trade-off analysis.
Obviously you can see that people prefer a lower priced vehicle of $23,000 (Utility of 2.10) over a higher priced vehicle of $25,000 (Utility of 1.15). Are you surprised? These values become extremely useful, however, when you are doing a
A Toyota with 280 HP, leather interior, no sunroof and a price of $23,000 has a utility of 0.75 + 1.18 + 1.60 -0.68 + 2.10 = 4.95If the same options were applied to a Volkswagen the overall utility would drop to 0.65 + 1.18 + 1.60 - 0.68 + 2.10 = 4.85The drop is directly related to the difference between the utility for the Toyota brand and the Volkswagen brand of .10
For example...
you want to make sure the benefit of providing a sunroof exceeds the cost of adding it on
Moreover, we can use these valuable utility values and interpolation to determine the value of a sun roof - OR what is the consumer willing to pay for a sun roof? This would greatly help a product manager when trying to determining product offerings / options and pricing!
...
Our prior example was a Toyota with 280HP with leather, no sunroof and a price of $23,000. Let's see what an average buyer says they would pay for a sunroof.
4.95 + 1.36 = 6.31
1) If you add a sunroof to the car it would change it's total utility from 4.95 (no sunroof) to 6.31 (with a sunroof) by adding 1.36 (difference between no sunroof of -.68 and having a sunroof of +.68)
3) 2.10 (current price) - 1.36 (additional cost of sun roof) = .74 -> note that .74 is between $25,000 and $27,000
im going to try to figure out how much they would be willing to pay for a car with .74 utility value, since we dont have that value, we have to use interpolation to find the value
2) The Toyota price was $23,000 with a utility value of 2.10. If you add a sun roof then the price utility would fall from 2.10 to .74. Why???
SEE MATH ON SLIDE 13 FOR EXPLANATION
4) The rest is interpolation. The result? Consumers say that they would be willing to pay $2,302.58 for a sunroof.
...
In your research you determine thatPrice, Condition, Cover, Delivery date are very important to consumers who buy books.Then you determine the different values for each attribute to use in a survey and we will keep it simple. For example
:-Price - $90 or $110
-Condition - New / Used
-Cover - Hard / Soft
Condition: New (+1) Used (-1)
Cover: Hard (+1) Soft (-1)
Price: $95 (+1) $110 (-1)
Delivery: 1 day (+1) 5 day (-1)
remember these attributes HAVE to add to zero
When you determine the different values for the model you conduct a survey of consumers and their assessment of each attribute and it's importance to them.
In our example, the survey would assess which of the attributes values they prefer over the other value and assign a numerical value to each:
Condition: New Used
Cover: Hard Soft
Price: $95 $110
Delivery: 1 day 5 day
After getting the attributes and values, create a combination of all product attributes possible and create a matrix:
ranking column
After creating your model, survey it and collect the results...the results are in the
Then, using our -1 and +1 ratings within each attribute we build a matrix of results
calculate the results
ranking
Each attribute value is multiplied times it's _____ to produce another matrix with weighted values.
"part worth" utility value for each
Sum the calculations, then divide by 8 (number of options) to get the
the most important to consumers in our survey.
Now that we have the part worth values of -.25 (cover), -2.0 (condition), -1.0 (delivery) and -.25 (price) we can chart them to see which attribute is
tells us that based on our results we see that condition is the most important attribute for customers, price and cover type are the lowest important attributes
What does this tell us as a product manager at ABC books?
Product analytics can help a manager with their product management decisions
There are a number of marketing analytic techniques to do this including
-K-Means clustering and Conjoint Analysis are just two
Conjoint analysis allows marketers / product managers to
-Demonstrate that stronger brands and better-quality products can charge a higher price without a discernable decrease in demand
-Identify product features that appeal to consumers to help a product manager determine the right balance between features and price
-Determine how consumers value the different attributes that make up a product and the trade-offs they are willing to make among the different attributes or features that compose the product
There is a myriad of possible applications with conjoint analysis!
Product analytics Summary
THIS SET IS OFTEN IN FOLDERS WITH...
CHP 11: Digital Analytics
84 terms
CHP 13: Website Analytics
88 terms
Big Data 101
76 terms
CHP 1: MARKETING ANALYTICS
67 terms
OTHER SETS BY THIS CREATOR
EXAM 3 Study Guide
232 terms
EXAM 2 REVIEW
121 terms
CHP 13: PR and sponsorships
42 terms
CHP 14: Display Ads
49 terms
OTHER QUIZLET SETS
IFR Cross Country Flying/ Airway Altitudes, MOCA,…
16 terms
Hamlet Act 1 Scene 1 and 2
15 terms
द्वितीयः सर्गः ।।२।। Sentence
57 terms
veterinary laws and ethics
33 terms