Professional Documents
Culture Documents
(a)
How many optimal segments emerge? What are their relative sizes?
3 optimal segments. Segment 1 sample has 14 people, segment 2 sample has 18, segment 3 sample has 8.
(b) Profile, interpret and label these segments. Rank order them in terms of market attractiveness.
Based on the sample data, if we assume that market size is directly proportional to market attractiveness,
then rank order is as follows:
Segment 2
Segment 1
Segment 3
Now, read in and run the discriminant analysis portion.
(c) Paste and interpret the confusion matrix that obtains. Which segment is most and least accurately
predicted?
Now, read in the holdout dataset Officestar Classification.csv and run it.
(d) Interpret the relative segment sizes that emerge from the predicted segment memberships.
Compare them with the test sample distribution of segment memberships.
Segment 1 has 122 people, segment 2 has 46, segment 3 has 132. The test sample was not the best
predictor of the segment size. Test sample, for instance, predicted that Segment 3 was the smallest, but in
fact it was the largest.
Market Attractiveness was recalculated by multiplying market size with Average Income of the segment.
Assuming people will spend more if they earn more, the rank order is as follows:
Segment 3
Segment 1
Segment 2
This is exactly the opposite of what we calculated with the test sample. Thus we would target Segment 3.
(e) In your chosen segment, identify and name the top 2-3 respondents who you will pursue and target
based on the targeting exercise.
I would choose customer 62, 107 and 185 to pursue since they are the closest to the mean within this
particular segment.
Mean for product returns is greater than price discounts and the std dev is roughly the same. hence,
people value product returns more than returns.
(ii) Do Co2016ers care relatively more about free shipping or about variety
available on the site?
Most of the CO2016 batch values variety more than free shipping. However, there is a fair bit of
variability in the std. dev. of the two.
Now close the App window. Open the R code ('R code hypoth testing
HW.txt' file) I have sent for this HW. Copy-paste each line into the R
console, one by one and press enter. The code solves the Qs. I'd like you to
try executing the code and answer the Qs accordingly. Feel free to take
help from peers or contact me, the AAs or Aashish in case of any issues.
Or to use JMP etc.
## Q1. Do hi spenders rely more on COD than prepayments? ##
hispenders = true = COD > Initial hypothesis > high spenders pay by COD.
But p= 0.01078<0.05
Hence initial hypothesis is false. HiSpenders prefer prepayment.
## Q2. Do hi.spenders also spend more time online? ###
hispenders = true = Hightime > Initial hypothesis > high spenders spend more time online.
p-value = 0.6923
Hence initial hypothesis is true. But correlation is weak.
## Q3. Do males spend more time online on average? (as a t-test of differences) ##
Welch Two Sample t-test
data: webtime[xm] and webtime[xf]
t = 0.78362, df = 122.44, p-value = 0.2174
alternative hypothesis: true difference in means is greater than 0
95 percent confidence interval:
-0.1650032
Inf
sample estimates:
mean of x mean of y
2.444853 2.296875
The p-value is significant enough for the null hypothesis to hold. That is, males spend more time
online on average.
## Q4. Is gender systematically associated with time spent online? (as chisq test of Assoc) ###
From the chi-squared test, p-value is 0.5969. Hence, we do not reject the null hypothesis that gender
is systematically associated with time spent online.
## Q5. Do iOS users spend more time (money) online than Android ones? ##
From the p-value obtained from the chi-squared test, it is evident that it is significant enough to not
reject the null hypothesis which is that iOS users do spend more time online than Android ones.