PDA

View Full Version : Statistical Analysis



Keith Starosta
06-18-2004, 9:46 AM
OK....bear with me here, folks...my brain hurts already with this one. :o

I have been tasked by my boss to research some statistical analysis techniques, and how we could possibly apply them to solve some customer issues on my current project. Here is an example of what I am trying to understand:

During election time, the media comes on the tube and states that, based upon a sampling of voters, Candidate X is projected to win the election.

There is some statistical science behind that type of projection, and I need to try to understand what that is. I'm researching the hell outta this right now, but if any of you guys have any insight, I'd greatly appreciate it! Thanks!

Keith

Kevin Arceneaux
06-18-2004, 10:14 AM
Wow, something on this forum I can be semi-usefull.

First a couple of things.

1. What you want and what political polls are is two completely different critters. You have a population set, your cutomers, that is fixed. A political poll (PP) is a population that is set by those taking the poll. The PP population can be easily skewed to get the results the people paying for the poll want. Yours is not. It is feedback from people that have purchased your product. An example of a skewed poll, that has been in the news lately is the LA Times presidential poll. A look at those that were polled show a definate bias for Kerry. (NOTE: THIS IS AN EXAMPLE ONLY< IT IS NOT MEANT TO OPEN A POLITCAL DISCUSSION. THE REPUBLICANS DO THE SAME THING)
2. Statistics lie. Again this will not pertain to you as you are using a fixed population. But in other polls, it can be set to throwout certain responses.

What you will need to do is set up a database of the feedback from your customers. Depending on the size of your cutomers that have provided feedback, you may not see any trends at first unless their is a glaring problem that is common in many of the responses. What you will do is start a database with each response have a column. Each customer that has that specific response, gets a check mark in that column. Responses that get a lot of checkmarks are protential areas of improvement or something that needs to be fixed now. Of course there may be responses that need not be included in the database. I wish the machine was purple, green, or etc, generally can be thrown out unless color is important.

You can also use a poll of your customers to see what they would like. Give them a set of responses and ask them to pick from important to least important and use that to set up a weighted poll. This will give you a lot of information on what they want. Again you set up a database and you will see what your cutomers want.

I hope I was able to give you an basic idea of things. Maybe their is a super math wiz who really enjoy's statistics. Personally I hated those classes.

Chris Padilla
06-18-2004, 1:19 PM
Mmmm, math...my cup of tea! Mean, media, mode, standard deviation...drool, drool.... :D

Joseph N. Myers
06-18-2004, 1:49 PM
Keith,

Kevin gave a good description of the differences between political polls and customer service (trends, problems, trends, etc). Assuming the customer service part, Kevin's recommendations are a good way to go.

If it is more complicated, I would guess that you are getting into an area called regression analyst where the object is to be able to determine what are the main items and in what order, influences the customer to buy the product(s) (or drop it or upgrade it, etc).

If I were you, I'd give the local college math department a call and talk to the department head. There always some grad students hanging around that would love to get into some real world problems and might get some college credits for doing it. And the price should be minimal.

If I can be of additional help, shot me a PM and we'll go from there.

Regards, Joe

Steven Wilson
06-18-2004, 2:34 PM
Mmmm, math...my cup of tea! Mean, media, mode, standard deviation...drool, drool.... :D

Ok Chris, what is the non-parametric statistical test that is also known as the "Vodka" test.

Chris Padilla
06-18-2004, 2:37 PM
You keep drinking vodka until your numbers look right? :D

Martin Shupe
06-18-2004, 5:06 PM
Keith,

Go to the library and get an introduction to statistics textbook.

Read the part about sample size, and how it relates to prediction accuracy. Basically, the accuracy you need of your results determines your sample size. If I remember correctly, the term is 95% "confidence interval", which is accepted by the scientific community.

I am not sure if you need to get into multiple linear regression and analysis of variance of not. But if you do, you will want a computer program to do the math.

In my day, we used SPSSX, which stood for Statistical Package for the Social Sciences. You enter the data, it spits out the regression equation, and ANOVA charts. Then you have to interpret what it all means.

It has been many years since I did this, but with a computer program, it is not that hard.

Steven Wilson
06-18-2004, 5:17 PM
You keep drinking vodka until your numbers look right? :D
No! It's the Kolmogorov-Smirnov test (Smirnov ;) ;) ). The Kolmogorov-Smirnov (KS-test) tries to determine if two datasets differ significantly. The KS-test has the advantage of making no assumption about the distribution of data. (Technically speaking it is non-parametric and distribution free.)

Chris Padilla
06-18-2004, 5:34 PM
Hmmm, I like my answer better! :p

Dennis Peacock
06-19-2004, 12:09 AM
You keep drinking vodka until your numbers look right? :D

Chris,

Yea...I failed my last vodka test.....had to be carried away and put to bed! :eek: :)

Ian Barley
06-19-2004, 5:20 PM
I lived for a long time with customer service data collection and analysis. One very simple and often neglected thing to bear in mind is the design of questions.

Yes/Nos are easy.

Choices (Red/Blue/Green0 are easy.

If you are trying to get relative scores always make sure that your scale is sufficiently fine (but that it is not 1-10) and that it is even in number - that way respondents cannot pick the middle number. And always make sure that you phrase your questions so that higher scores represent better experiences. I once had to rationalise some data from somebody elses questionnaire and he had set it so that 1 was a good score on some questions and a bad score on others. Made presentation of results extremely difficult.

Aaron Koehl
06-19-2004, 8:28 PM
OK....bear with me here, folks...my brain hurts already with this one. :o

I have been tasked by my boss to research some statistical analysis techniques, and how we could possibly apply them to solve some customer issues on my current project. Here is an example of what I am trying to understand:

During election time, the media comes on the tube and states that, based upon a sampling of voters, Candidate X is projected to win the election.

There is some statistical science behind that type of projection, and I need to try to understand what that is. I'm researching the hell outta this right now, but if any of you guys have any insight, I'd greatly appreciate it! Thanks!

Keith
Keith,

If you're interested in projecting (forecasting), it's regression you're after. Without going into the gory details, I'll start by saying you'll have to collect some data (a sample), but certain rules govern that data collection in order for your results to be significant (meaningful). Based on data you collect, you can then build a mathematical model which will be able to provide insight into the problem at hand.

Asking about statistical analysis is very much like posting in the GWW forum:
"I need to create something out of wood." For a major project, it might be worth hiring a consultant.

That said, is there any more information you could possibly provide? Roughly speaking, what is it your're trying to forecast, what kind of data do you have access to, and what is your timeframe?

This ought to be fun. :)

Martin Shupe
06-20-2004, 6:31 PM
I was at Walmart yesterday and happened across a "Cliff's Notes" guide to statistics. It was about $10.

If you want a simple explanation of statistics, it did a fairly good job of keeping it simple, but still explaining what the terms mean.

You might want to check it out and see if it would be useful to you.