I am enjoying the new BBC series "
How Do You Solve a Problem Like Maria?" and have been using it so often as a catch phrase in my work conversations that I couldn't resist sharing my experience at the
F2009 conference earlier this month in the same context! If you haven't caught the show - where Andrew Lloyd Webber is searching for his next Maria for The Sound of Music - you are missing a treat.
AND if you missed the latest gathering for forecasting experts at
F2009 earlier this month, I wanted to share a few tasty experiences. I'll keep you posted on the F2010 schedule so you can plan ahead. (Or so that you can forecast your 2010 conference attendance, pun totally intended.)
One of the interesting F2009 sessions featured
Susie Fortier from Statistics Canada.
Statistics Canada publishes all official statistics like labor, wages, retail wholesale, Gross Domestic Product, etc. for Canada.
Statistics Canada has several underlying objectives that have to be factored into the reporting of these important statistics - which include exposing the underlying math to the public so it is
exactly reproducible, presenting several estimates or alternatives and stating all findings as conditional.
There are certain situations where the technique or philosophy you adopt to solve the problem could impact those objectives to provide the public with accurate, reproducible data. Statistics Canada is not formally in the business of forecasting but they have adopted an approach to addressing these situations for a consistent methodology to non-overlapping time series data.
Sometimes the need to link a time series is driven by artificial breaks such as a survey redesign, survey category reclassifications, etc and the linking is performed to join the data. Steps must be performed in these situations to test the significance of the break . There are three approaches you can use according to Fortier to test significance:
1) End user view ( a.k.a. gut feel )
2) Sampling error
3) Statistical significance of the shift
Fortier discussed all three approaches. In detail she discussed how to approach the problem when the break is caused by survey redesign – one of those artificial break scenarios. Fortier overviewed the approach of wedging which can also be used for operational reasons – for example if the data must not change prior to a certain date.
Her example of the Canadian Travel Tourism survey was one most marketers have experienced. For example, when a customer satisfaction survey or market data survey is fielded over a period of time and the survey questions, categories are tweaked. In this particular case, Fortier was dealing with changes to the tourism survey that included changes to mode of travel, expenditure classifications, etc. This artificial break due to classification changes and survey design had to be considered.
Fortier walked through the example using the data wedging approach to show a gradual level shift from one survey time series data to the next and I found it very interesting from my past-life as a field marketing person. These issues and challenges often came up with comparing “adjusted” surveys over time.
That said, being relatively new to the forecasting arena and focusing more on analytics, time series data and approaches has been of interest. I’ve been paying attention to my colleague, Mike Gilliland’s, new blog
The Business Forecasting Deal. Check it out! I’ll post more about my time at F2009 tomorrow!
As Business Intelligence Product Marketing Manager for SAS, I’m responsible for the Global message direction, strategy and marketing of SAS business intelligence. More on me. About the pot pie, I'm a southern girl, and that's comfort food. But food for thought -- I can explain.