Is there a term for this?
March 5, 2019 6:24 AM   Subscribe

My google skills are not up to snuff for this. I'm looking for a term where people are asked for input (workplace, surveys, etc) but never get any results or feedback about their input, resulting in less input provided by those people. This decreased input is then interpreted as "well, there must not be a problem if we're not hearing about it."

Best example I can think of is the unemployment rate. In the news a year or two ago, there were reports of a stable or slightly decreasing unemployment rate. This was interpreted by some as unemployment improving. In actuality, people that were unemployed for long stretches of time were less likely to report being unemployed. So the problem didn't really get better, the data was just bad.

Is there a term for this process/situation?
posted by defenestrated to Human Relations (7 answers total) 7 users marked this as a favorite
I'm only familiar with the terms "bad data" and "bad strategy" from general business (about failure to keep reliable data, and failure to recognize bad data as a problem in the system).

From data analytics, you are describing one aspect of what's called "overfitting" though most data analysts/data scientists I know of would probably object that this was bending the meaning of the term since "overfitting" actually means engineering your predictive model too well to known data - so much so that its predictive value is supposedly very good, but turns out very bad when compared to test data with known outcomes that it wasn't trained with.

Perhaps you will get more (more useful) answers.

(FWIW, I googled "mistaking poor data for success" and got a lot of hits back for "bad data" which it looks like, thankfully, some business decision-makers are finally concerned about - the honeymoon period with the benefits of machine learning seems to be waning.)
posted by kalessin at 6:47 AM on March 5, 2019

I think this is an example of participation bias/non-response bias.
posted by capricorn at 7:16 AM on March 5, 2019 [5 favorites]

Seconding capricorn. You might also try “selection bias,” which I think is taking over as an umbrella term. Ngrams link
posted by eirias at 8:43 AM on March 5, 2019

Your first example of decreasing responses to workplace surveys is specifically non-response bias, which is a type of selection bias.

But I'm not sure if that completely covers your example of unemployment rates, which seems to also involve poor interpretation of data.
posted by theory at 9:01 AM on March 5, 2019

I'm not aware of a name for that whole dynamic. Sampling bias is just one aspect of the broader situation.

The specific situation you refer to is called "permanently discouraged workers" or just "discouraged workers," so perhaps by analogy we could call this the "discouragement & mis-sampling cycle."
posted by adamrice at 10:49 AM on March 5, 2019

It could also be ``cooling the mark out'', depending on how intentional you think it is.
posted by clew at 11:32 PM on March 5, 2019

In case a second example would help:

If you handle the first few sexual harassment complaints poorly, people will stop complaining (to you). And then it may look like you have no harassment issues. Unfortunately, "zero" for harassment appears to be nonzero. If you really measure zero, you're just not getting the measurement.
posted by meaty shoe puppet at 6:18 AM on March 7, 2019

« Older Importance of Physical Attraction When Looking at...   |   Yes, it's another "How Do I Television" question Newer »

You are not logged in, either login or create an account to post comments