PART ONE:
Hey, thanks for walking me through the audit problem! So, to break it down in the
best way i know how, we can think of each client as having a small chance being around 7-
10% chance of needing extra audit work. Statistically, I’d call this a random variable with a
binomial distribution, but you don’t need to worry about the jargon related to that. What it
says in a basic sense is a yes or a no among all 10,000 samples.
Now, with 10,000 clients and a 7.3% chance, we can expect, on average, about 730 clients
to need extra work (that’s 10,000 × 0.073). But remember, that’s just the average and we
won’t hit that exact number every time because there’s always some uncertainty and
natural variation. This is often referred to as a confidence interval. What that means is just
that out of all of our test we will be able to be “x” certain that the median falls into the
sample mean.
As for the chance of more than 1,000 clients needing extra work — that’s actually pretty
unlikely. Obviously, 1,000 is way above the expected 730, and with that many trials, the
results usually stay clustered close to the average, not way off in the tails. So, we probably
don’t need to stress about going over 1,000. Since the sample follows a normal
distribution, it would be unreasonable to expect a number at 1000.
PART TWO:
That makes a lot of sense. The reason you got an error when trying to calculate the average
is because the state of residence is a categorical variable, not a numerical one. Meaning,
states like “Texas” or “California” are names, not numbers. Different pieces of data can be
ordered into several different categories. Nominal data can be labels, categories, or zip
codes – data that can be labeled or categorized with no specific order. Ordinal data is data
that has a meaningful order, but is not numerical. This can be survey ratings or education
levels. To reiterate, these two categories are categorical levels of measurement. The next
two categories, interval and ratio, are numerical variables, meaning the data is measurable
and can be described with numbers. Interval data is data with no true zero, like how
temperature in Celsius can go negative. Ratio is similar to interval data, but it actually does
have a true zero. An example could be hours worked or revenue made. In this case, you
should be using mode as your measure of central tendency. It will tell you which state has
the highest number of clients. Instead of averaging your data, think: “What’s the most
common state for our clients to live in?” I would also recommend a bar chart as far as
visualizing your data. Each bar could represent a different state, and the height of the bar
would reflect the number of clients from that state.
PART THREE:
a)
Yeah, so a confidence interval, in this case a 95% one, just means that based on the data
they collected, they’re confident that the actual average return falls somewhere between
$11.7M and $12.3M. It is not saying there's a 95% chance of hitting those exact numbers,
it's more like, “If we repeated this process a bunch of times with different samples, 95% of
the time the true average would land in that range.” Also, I can’t really guarantee you’ll hit
$11.8M. Since the lower end of the interval is $11.7M, there’s a chance the real return
could be just under what you need. I get that you’re looking for a solid “yes” or “no,” but
statistics don’t, they just help you manage uncertainty.
b)
One thing that sticks out to me is that the team only used data from 20 investments. That’s
a pretty small sample, which can make the results less reliable. When you’re working with
a small sample size, you tend to get wider intervals and more variability. So, if they want a
clearer picture, especially when decisions like shareholder confidence are on the line, I’d
recommend gathering data from more investments if possible. Bigger samples usually
mean more accurate estimates and narrower confidence intervals.
c)
No, I wouldn’t say the results are wrong just because they used 20 samples. It’s more that
the results come with more uncertainty. A small sample can still give you useful
information, but it’s like looking at a blurry photo, you can kind of see what’s going on, but
it’s not as sharp or detailed as it could be. So, I wouldn’t throw out the results, but I’d
definitely treat them with a bit more caution and consider following up with more data if
decisions depend on it.
PART FOUR:
To check if advertising is affecting the number of job applications, we can use a correlation
analysis which is just a simple tool that shows if two variables move together kind of like a
visual separation of the two so we can see easily how they react to echaother.
A scatterplot would be a great visual for this. You’d plot ad spending on the x-axis and
number of applications on the y-axis, and then see if the dots form a pattern.
If you want to go a step further, you can calculate the correlation coefficient which is often
refer to as just “r”, which gives a number between -1 and 1 to show how strong the
relationship is. A number close to 1 or -1 means a strong relationship; a number near 0
means no relationship.