Knowing your data …

Lots of good things come from New Zealand … Bungy jumping, Easiyo® make-your-own yoghurt, the All Blacks®, Bega® and Mainland® cheese, good Sav-blanc wines, Crowded House, Russell Crowe (well, maybe not the last one).

But what also comes from New Zealand, and has done so for a very long time, are creative and competitive ideas pursued by skilled and knowledgeable market researchers.

The latest I have come across is a rather cool online tool developed by Irene Rix and her business partner Josh Bondy, at their aptly named ‘CodeKiwi’ business, albeit based in Melbourne.

Some of you will know that I can go on a bit about the quality of data files that I am given to analyse.  Indeed, some of the leading interviewing packages still don’t seem to have ‘got’ the idea that their great online interview scripting capability can still result in data files that are painful to analyse. 

In most cases, asking your data provider to give you output in SPSS .sav format will minimise my pain (and that of any analyst).  But not always.  In any case, a lot of times I still receive data files in .csv or .xlsx format, and that can raise a whole lot of issues that normally just don’t occur with SPSS data, e.g. (to name just a few):

• No labelling information for code frames or questions
• Answers to multi-response questions recorded in just one field, with responses separated by commas
• ‘Pick any’ responses recorded as 1, 2, 3, etc instead of 0/1.

I could go on.  

In most cases, I therefore prefer to refer people to the excellent guide to SPSS data file preparation, available on the Survey Analysis website http://surveyanalysis.org/wiki/SPSS_Data_File_Specifications.

The CodeKiwi team’s approach, in contrast, stems from their recognition that in the wider data and analytics field, the .csv (and .txt) data file format is pretty much pervasive.  Their online tool thus allows anyone (not just market researchers and analysts) to upload these types of data files, and have  comprehensive and automatic checks and reports run on the data structure, format, missing value patterns, distributional forms and characteristics, values requiring recoding, string lengths and a host of other information.

In addition, a unique ‘Data Health Index’ is provided, that is based on the pattern of missing values, mix of variable types, and their skew, variance, and concentration of values.

In their own words:

“Knowing your data before you get started can save hours of pain.  Learn what's lurking in that shiny new datafile, and save your time for the clever stuff...”

The “Know thy Data” initiative is a first-rate effort to come to grips with a key issue that is faced by all data analysts, whether in market research or in other fields of data endeavour.

Why not have a look for yourself?  https://knowthydata.io/  

********************************************

KKO is not OK !

I've been using a new Nokia 930 Windows 8.1 phone for a month or so now.  No problems with the phone, it does all that I need it to do, albeit the Windows store is a bit light-on for useful apps.  For example, there are about 10 apps on my old Samsung S2 Android phone that I still use regularly (banking, etrade, Fire Ready from the CFA) that simply don't exist for Windows.

But that's not a real problem, nor was it unexpected when I signed up for the Nokia.

What was unexpected is the totally unforgiveable and outrageous scam that Telstra actively facilitates from a outfit of debateable business morals called KKO Store.

Specifically, you only have to accidentally click on an ad or banner within one of your downloaded (totally legal, and endorsed by Microsoft) Windows apps, to then receive a text like the one below:

Now, normally you'd just ignore this type of spamming and delete it.  BUT ... if you do that, what then happens is that Telstra kindly adds $9.99 to your phone bill every week in perpetuity, because what you should have done is reply to the above text with the word 'STOP'.

And when you contact Telstra to complain, their response is 'sorry, nothing to do with us, we're just passing on the costs, you'll have to take it up with KKO' and 'your phone seller should have warned you about Premium SMS' (they didn't ... and it was a Telstra Shop!).

As I said, outrageous.  The direct  parallel is if I were to contact you, and tell you that I am going to take $10 out of your bank account every week (because your bank has said they are happy to comply and cooperate with me), unless you specifically tell me not to.

Where is the ACCC on this?  Where is the TIO?  Nowhere to be seen, so far as I can tell.

There have been countless people caught by this scam, from relatively knowledgeable business people (like myself) through to little old grannies (like my mum).

Telstra, you've gone down in my estimation .... on the basis of this experience, I'd give you a 4 or lower on the NPS scale.  

A really very useful discussion ...

For those of you of quantitative bent, and even those of you not, this discussion on what you can and can't or maybe shouldn't attempt with choice modelling is really rather good http://www.sawtoothsoftware.com/download/Turbo_Choice_2014.pdf
 
[Sawtooth are probably the largest purveyors of choice and conjoint software globally.  They are incredibly generous with what they share with the broader research community, regardless of whether you are a user of their software (I'm not) or not.]

[Oh oh ... looks like the above link is not working any more - contact me if you would like a copy of the paper/pdf.]

Conclusive proof that social media data predict sales ?

I don't think so.

From a recent email by one of the keynote speakers at the AMSRS conference for 2014 ...

"Important news to share today! There is conclusive proof that social media data predict sales from a landmark study made public.

It proved that the quantity of social media conversations about a brand has a statistically significant relationship to changes in its sales. 

.....

This industry learning comes on top of an academic paper by Prof. Wendy Moe at the University of Maryland that showed a correlation of .8 between social media listening data and brand equity metrics derived from survey questions." 

Here are four data sets (Source: www.r-project.org) that each have an identical correlation of 0.8 ... does that look like 'proof' of anything to you?

A more appropriate use of NPS ...

http://mosttrustedadvisers.com/

Valid use of NPS? You be the judge.

I recently had cause to phone the Fairfax help line, to try and work out why I couldn't log in to my online account on 'The Age' website.

The advisor I spoke to was quite helpful, so no complaints there.

But I then received the following from the Fairfax organisation ...

Answering this question poses a number of problems:

• Do I assume that my friends and colleagues have asked me for a recommendation concerning The Age, or am I just volunteering an opinion out of the blue?
• What's the context, i.e. is it for business news coverage, sport, political commentary, etc.?  Which one (or ones)?
• Is the recommendation sought for the online edition of The Age, or is it for the print version?

NPS can be a helpful metric, but the underpinning question needs to be framed carefully, and take full account of the context in which it is asked.

A better (but not perfect) question in this case might have been something along the lines of:

"Based on your (recent) service experience, if they asked you for your opinion would you be more or less likely to recommend the online edition of The Age newspaper to friends or colleagues, or does it make no difference to what you would say to them?"

Spurious correlation ...

Go here ... http://www.tylervigen.com/

[Thank you, John Gandar, for the link.]