Showing posts with label bivariate. Show all posts
Showing posts with label bivariate. Show all posts

Tuesday, February 5, 2013

Ice Cream causes Swimming Pool Deaths!

I see this proverbial "Ice cream causes swimming pool drownings" statement made in the world of economics and politics all the time.

It's so prevalent that there's a Wikipedia article on spurious relationships.
[Ice cream] sales are highest when the rate of drownings in city swimming pools is highest.
You can look at the data over and you'll see that this phenomenon happens like clockwork:
  • Low ice cream sales... fewer swimming pool deaths.
  • High ice cream sales... many swimming pool deaths.
So there's a correlation, right? Yes.

With that correlation, some go farther to allege that ice cream causes drownings or that drownings causes ice cream sales. (Ahem, no.)

To claim that ice cream sales is an indicator of drownings or vice versa also misses the point because ice cream sales and swimming pool deaths are both results of an underlying factor; a heat wave.

Unfortunately, this statement of two symptoms indicating one another is seen all the time in the world of cell culture analysis:
  • Final ammonium (NH4+) is an indicator of culture performance
    - or -
  • Final lactate (Lac) is an indicator of product titer

credit: The Usual Suspects MGM

Seriously, who here doesn't already know that cell growth impacts culture performance?  Or that cell metabolism impacts culture performance?

Yet we are still publishing papers on how final lactate is an indicator of product titer and concluding that cell metabolism impacts culture performance.

Final ammonium or final lactate are symptoms of cell culture metabolic conditions that produce higher titers.

Unless you can:
  • Change media components
  • Change a parameter setpoint (pH, temp, dO2)
  • Change the timing of culture operations (temp shift, pH shift, timing of feeds...)
Essentially recommend specific changes the Production group can execute to improve culture conditions and you've simply uncovered a spurious relationship; there remains no action you can take to improve culture performance.

This is why it is best to start your multivariate analysis by picking actionable parameters to ensure that you have true factors.

When you pick actionable parameters to model as factors in your multivariate analysis, you have a shot at gaining control of an out-of-control campaign and meeting your Adherence-to-Plan, as Rob Johnson did.

If you're happy pontificating from ivory towers, keep making true-but-useless statements on how every time Y1 happens that Y2 also happens.