Sep 6 2011

Really Simple Statistics: What is Ordinal Data? #MRX

Welcome to Really Simple Statistics (RSS). There are lots of places online where you can ponder over the minute details of complicated equations but very few places that make statistics understandable to everyone. I won’t explain exceptions to the rule or special cases here. Let’s just get comfortable with the fundamentals.

Today we tackle another kind of number. Unlike nominal numbers, ordinal numbers have real meaning behind them. The name itself hints at the meaning. Ordinal numbers portray ordered numbers.

But, the only thing we know about the numbers is that there is an order to them. For example, there are more cookies in the first picture than there are in the second. But, we can’t see the whole picture, so we don’t know how many more cookies are in the first picture. We could assign a a 2 to the first picture and a 1 to the second picture, but we wouldn’t be able to say that there are twice as many cookies in the first picture. Just that there are more. Here are some examples of ordinal data.

A big handful of rice vs a small handful of rice. Why: We don’t know how much rice is in each hand but we can see there is more in one than the other.
Someone who is a bit shy vs someone who is really shy. Why: We don’t how much more shy the really shy person is, but we know they are more shy.
Questions on surveys where the answers look like: Strongly agree, somewhat agree, somewhat disagree, strongly disagree. Why: We don’t know how much more “strongly” is compared to “somewhat” but we do know it’s more.
This is more than that. This is lighter than that. This is heavier than that. This is taller than that. This is bluer than that. This is tastier than that. This feels more rough than that. This smells worse than that. This is longer than that. This is earlier than that. This is faster than that.

The key indicators are these:

Something is more or less than the other thing
We don’t know how much more or less it is

It’s just that simple!

May 16 2011

Actions based on personal opinions not data?!

Image via Wikipedia

The scientist inside each of us demands factual, solid, indisputable proof. Facts are golden truths that tell us which brands people buy, how often they buy them, and where they buy them. Facts come from scientific research that abides by established validated methods.

But facts are boring. We often already know the facts but we do our research to confirm them. We already know that girls like Hello Kitty and boys like Transformers but the research must be done. So when crazy people come along and make decisions without regard for research findings or without doing any research at all, we are appalled.

In fact, I don’t believe that decisions without research are possible. Even companies that “do no research” (I’m talking to you Steve Jobs) are constantly doing research. They’re just using a different methodology.

Their datasets are decades of experience with real consumers, thousands of sales figures tied to market events, intuition applied to raw data, opinions based on boring datatables. All of these things are simply alternate forms of insight tools analyzed and interpreted by insight generators, people. They just happen to be insight tools that survey researchers dream about but rarely have access to.

So don’t be fooled. The next time someone says they didn’t do any research before coming to their conclusions, think about what datasets they did have the privilege of using.

Surprise, surprise! A non-rep sample is as good as a ‘rep’ sample (lovestats.wordpress.com)
Mobile Research Conference 2011 – Take Aways (natashaallden.wordpress.com)
Statistics Speak Louder Than The Estimations! (seodoz.com)

Nov 2 2010

#MRA_FOC #MRX Effective Data Visualization by Naomi Robbins, Part #2

Image via Wikipedia

nbrgraphs

After a really neat bag lunch, it was time to jump back into part 2 of Naomi’s data visualization session. So many fun tidbits this afternoon and here are just a few.

Banking to 45 degrees is something to consider – Choose a scale so that the angles within a chart are around 45 degrees. Neato!
You can create charts without a zero but you need to have a good reason, e.g., if the chart with zero fails to show a change in slope. But your audience needs to have an understanding of how to read a chart without zero. BUT, bar graphs ALWAYS need zero because they are a judge of length which always starts at zero.
People are confused by double Y axes, particularly if it’s the same variable on both sides. A double axis is where the right side of your chart uses one scale, e.g., from 0 to 10, and the left side of your chart uses another, e.g., from 0 to 20.
Naomi has called a textbook publisher to complain that its charts were rife with chart lies. Let’s see if the next edition takes heed.
Scary – people using textbooks with bad and misleading charts are usually designed for people who have no exposure to charts before. Publishers want the charts to be fun and engaging. Sad, sad, sad.
Cindy Brewer has a great website on choosing colours for maps and it works great for charts as well. http://www.colorbrewer.org
Sequential hues are good for sequential data (e.g., time series) – light blue, medium blue, dark blue. Diverging hues are good for categorical data – red, green, blue
8% of male population and 0.5% of female population has trouble with colour so reconsider the use of red and green in charts even if you use red to represent positive/go and negative/stop. Check your individual charts at www.vischck.com. VERY cool.
Blue and orange are better colours for charts because there are fewer problems for people reading them.
Graphs are for the forest, tables are for the trees. Tables are not better than graphs and graphs are not better than tables. Use the tool for the purpose.
In 2010, no one is impressed that you can change a font (that’s just stupid) or make a 3d chart in excel.

If you haven’t been to a charting class before, you really should try Naomi’s course. You will learn far more tidbits than you can possibly remember but you will definitely learn something that you can implement immediately. I’ve taken the Tufte course and even I learned some new things. Yay charts!

Naomi@nbr-graphs.com, http://www.nbr-graphs.com

This slideshow requires JavaScript.

Any errors in representation are my own.

____________________________________________________________________
Conversition Strategies Social Media Research: By researchers, For researchers

Nov 2 2010

#MRA_FOC #MRX Effective Data Visualization by Naomi Robbins, Part #1

nbrgraphs

I was caught between a pillow and a soft place this morning with a choice between a session on social media and the other a session on charting. But, as a fan of Edward Tufte, a legendary charting specialist, I couldn’t resist attending Naomi’s data visualization session.

She began the session by testing the room lighting to see if the colours on her presentation would show correctly on the screen, something I can appreciate having given presentations myself where a variety of colours ended up looking the same. It is something everyone should do particularly if you are presenting charts. If your labels, gridlines, or distinguishing chart features don’t show up, you might as well not do the presentation at all.

Here are just a few of my favorite points:

The best chart is the one where the information is detected most quickly
If perceiving the information is not important, then a pie chart is fine, e.g., when the chart is used as decoration
The way you read a chart depends on which software you use and labeling the data points does not make a bad chart ok. See the chart below to see if you can determine what the data points are. Does the line match up with the front of the bar, the back of the bar, or neither!
Graphs are to show relationships and trends, not exact numbers. If you need exact numbers, then use a table. Hence, bar charts do not need numbers.
All bar graphs should start at zero because bars reflect length which has a zero.
Alphabetical order is rarely the best way to order data.
There is no substitute for colour.
People know what number comes between 88 and 90 so you don’t need to label every point.
When we use error bars, we often use 68%. But 68% makes sense in a table for self-calculation. Doesn’t 95% make more sense in a chart?
Museums want to show data honestly and accurately. Corporations….. have other ideas.

Naomi presents in a style reflective of a professional statistics geek with tons of charts and examples and I got quite a kick of the morning session. She showed us a lot of tricks that I like to play on my colleagues such as having them guess chart values on really bad charts. She showed us a number of charts that I have never seen before and am now anxious to try. She showed many examples of bad charts turned good with just a couple minutes of work. She provided a set of notes that is probably the best set I have EVER come across. She suggested that though Edward Tufte is a charting genius, he is not the only expert in charting and she introduced us to William Cleveland, one of her favourite experts.

This slideshow highlights just a few of the huge range of charts that Naomi highlighted. You really need her commentary to see just how funny some of the charts are but I’m sure you’ll enjoy them anyways.

This slideshow requires JavaScript.

____________________________________________________________________
Conversition Strategies Social Media Research: By researchers, For researchers

Edward Tufte: Sparklines: theory and practice (edwardtufte.com)
Using Modern Data Visualization Methods in Business Reports (businessmanagement.suite101.com)
50 Free Web Apps to Make Beautiful Graphs (bestconstructionmanagementdegree.com)

Aug 18 2010

Laugh at yourself and then cry at our flailing industry

Well, once you manage to catch your breath after laughing solid for 4 minutes, let’s really think about all the people involved in this little prank.

1: Interviewer: First of all, this interviewer deserves a raise, a bonus, and a promotion for going through this interview without laughing, getting upset, or antagonizing the survey responder. I’m sure he deals with this sort of thing, whether real or fake, all day long every day. And yet, the utmost professionalism on his part. Kudos for a great job.

2: Responder: How did our industry get to such a state where surveys are written so poorly that people leave a tape recorder at their telephone waiting for researchers to call in order to make fun of them? This is nothing for us to be proud of.

3: Data Analyst: How exactly is the data analyst going to handle data which is clearly horrible quality? Will the analyst think of checking for outliers in each question? Will the analyst review the entire set of responses to recognize that it is an across the board outlier and probably a troublemaker? Will these responses lead to completely invalid analysis and conclusions?

4: Survey Author: Of course, we understand the need to use standardized questions in surveys. But, no matter how convinced you are, the world does not consist of people who know how surveys work. There are absolutely people out there who need to be taken through a survey with far more care than what we

permit when writing surveys. Telephone surveys need to be written so that interviewers can speak naturally and help those people who actually need some help. That’s where good data comes from. I’m really curious if the survey author left a place for the interviewer to indicate that this instance was possibly an outlier.

So, enjoy. But the next time you write a survey, keep this in mind. Are you antagonizing yet another survey responder or are you responsible for creating a more positive market research experience?