Bounce and momentum

Lifted from Robert Waldmann’s Stochastic thoughts:

Brendan Nyhan autopsies the mittmentum narrative here.

I comment.

Thank you for this very good post.  I had been puzzled by the “bad narrative about momentum” narrative on progressive blogs (largely because I interact with the rest of humanity largely by reading progressive blogs).  You demonstrate that there was indeed a mittmentum narrative.

I’d like to add some thoughts.  First, aside from the general tendency to seek facts which confirm the story we want to tell, people have huge amounts of trouble with time series which have no momentum.  This has been demonstrated by psychologists.

The experiment is to show people a random walk — a time series in which the changes are uncorrelated so the best forecast of where it will end up is always exactly the current level.  Subjects just can’t resist perceiving mean reversion or  momentum. These are opposite.  Mean reversion is called “a bounce” in the campaign literature.  The idea is that a recent change shall partly fade away and things will go back to the way they were.    So immediately after the first debate, there was much discussion of whether Romney’s gains were a bounce — destined to vanish.  Momentum is the opposite, people see a trend and extrapolate it so they think a shift will not just last but grow stronger over time.  Basically people always see one or the other when they are shown data with neither.

These opposite errors are linked.  The idea that a random series of changes shouldn’t move in the same direction for a while (so changes are perceived as bounce) makes people think something funny is going on when they happen to be in the same direction for a while (we see a new trend).   Notably, since polls contain sampling error which is independent across polls, polling averages contain independent changes and give the sort of series which we just can’t mentally accept.

Tracking polls make things trickier still.  Because they are moving averages, the will have trends even if overall public opinion doesn’t.  Say the Gallup poll will tend to move in the same direction for a week, because it is a weekly moving average.  It is psychologically hard to see something trend for a week and not extrapolate the trend.  This is true even if one knows that the data are a weekly moving average.
I don’t know votamatic (I check Jackman, Talking Points Memo, fivethirtyeight ,and Real Clear Politics uh often).  Jackman and the TPM group add to the problem.  Their smoothed average is LOESS which means they calculate the value for t as follows: estimate a time trend with data with weights which decline for polls further from t then report the fitted value for time t.  this means that they extrapolate trends.  The approach assumes that there is momentum.  Then people extrapolate the trend in the extrapolated trend.    I  have checked and confirmed that dropping a medium old poll with huge Romney – Obama can cause the current estimate of Romney minus Obama to go *up*.  I try and try to find such cases and have found one or two IIRC think the McLaughlin for Allen Virginia poll was one (the web user clicking buttons experiment depends on what other data are used so can’t trivially be reproduced using tools at the sites).

There is also just a lag in reporting.  Total webaholic political junkies consider early October ancient history.  The ink, paper and TV reliant not so much.  This post stresses how a 10 day old pattern was noted 3 days after it was confidently asserted that there was no such pattern.  7 days  has not always a huge amount of time.

Finally there is spin frankly reported as spin.  A whole lot of the momentum stories quoted (often anonymous) Romney campaign staff claiming they had momentum.  It is just a fact that they made those claims.  Importantly the Obama campaign didn’t push back (Ezra Klein claims this and he would know).  Reporters just don’t report that while he said and she said the same thing, the data show something else.   I think many consider reporting the mutually denied facts to be unethical — they feel they can’t contest Romney campaign claims more than the Obama campaign does.  Now I have no idea why anyone granted anonymity to flacks pushing the campaign’s narrative and I don’t like the he said she said approach.  But it is not the same as perceiving momentum and reporting that there is momentum.   They said Romney flacks said Romney had momentum, because this is just what happened.  We have a problem, but it is not in our common psychology but in the manipulability (sp) of reporting based on   simple rules.
Anyway sorry for the overlong comment on your excellent post.