The Philosophy of Data

advertisement
The Philosophy of Data
By DAVID BROOKS
Published: February 4, 2013
If you asked me to describe the rising philosophy of the day, I’d say it is data-ism. We now have the
ability to gather huge amounts of data. This ability seems to carry with it certain cultural
assumptions — that everything that can be measured should be measured; that data is a
transparent and reliable lens that allows us to filter out emotionalism and ideology; that data will
help us do remarkable things — like foretell the future.
Over the next year, I’m hoping to get a better grip on some of the questions raised by the data
revolution: In what situations should we rely on intuitive pattern recognition and in which
situations should we ignore intuition and follow the data? What kinds of events are predictable
using statistical analysis and what sorts of events are not?
I confess I enter this in a skeptical frame of mind, believing that we tend to get carried away in our
desire to reduce everything to the quantifiable. But at the outset let me celebrate two things data
does really well.
First, it’s really good at exposing when our intuitive view of reality is wrong. For example, every
person who plays basketball and nearly every person who watches it believes that players go
through hot streaks, when they are in the groove, and cold streaks, when they are just not feeling it.
But Thomas Gilovich, Amos Tversky and Robert Vallone found that a player who has made six
consecutive foul shots has the same chance of making his seventh as if he had missed the previous
six foul shots.
When a player has hit six shots in a row, we imagine that he has tapped into some elevated
performance groove. In fact, it’s just random statistical noise, like having a coin flip come up tails
repeatedly. Each individual shot’s success rate will still devolve back to the player’s career shooting
percentage.
Similarly, nearly every person who runs for political office has an intuitive sense that they can
powerfully influence their odds of winning the election if they can just raise and spend more money.
But this, too, is largely wrong.
The data show that in state and national elections that are well-financed, television ad buys barely
matter. After the 2004 election, political scientists tried to measure the effectiveness of campaign
commercials. They found that if one candidate ran 1,000 more commercials than his opponent in a
county — a huge disproportion — that translated into a paltry 0.19 percent advantage in the vote.
After the 2006 election, Sean Trende constructed a graph comparing the incumbent campaign
spending advantages with their eventual margins of victory. There was barely any relationship
between more spending and a bigger victory.
In May and June of 2012, the Obama campaign unleashed a giant ad barrage against Mitt Romney,
but as political scientist John Sides wrote in The Times’s FiveThirtyEight blog recently, the ads had
no lasting effect.
Likewise, many teachers have an intuitive sense that different students have different learning
styles: some are verbal and some are visual; some are linear, some are holistic. Teachers imagine
they will improve outcomes if they tailor their presentations to each student. But there’s no
evidence to support this either.
Second, data can illuminate patterns of behavior we haven’t yet noticed. For example, I’ve always
assumed that people who frequently use words like “I,” “me,” and “mine” are probably more
egotistical than people who don’t.
But as James Pennebaker of the University of Texas notes in his book, “The Secret Life of Pronouns,”
when people are feeling confident, they are focused on the task at hand, not on themselves. High
status, confident people use fewer “I” words, not more.
Pennebaker analyzed the Nixon tapes. Nixon used few “I” words early in his presidency, but used
many more after the Watergate scandal ravaged his self-confidence. Rudy Giuliani used few “I”
words through his mayoralty, but used many more later, during the two weeks when his cancer was
diagnosed and his marriage dissolved. Barack Obama, a self-confident person, uses fewer “I” words
than any other modern president.
Our brains often don’t notice subtle verbal patterns, but Pennebaker’s computers can. Younger
writers use more downbeat and past-tense words than older writers who use more positive and
future-tense words.
Liars use more upbeat words like “pal” and “friend” but fewer excluding words like “but,” “except”
and “without.” (When you are telling a false story, it’s hard to include the things you did not see or
think about.)
We think of John Lennon as the most intellectual of the Beatles, but, in fact, Paul McCartney’s lyrics
had more flexible and diverse structures and George Harrison’s were more cognitively complex.
In sum, the data revolution is giving us wonderful ways to understand the present and the past. Will
it transform our ability to predict and make decisions about the future? We’ll see.
Paul from Lower Manhattan
Downtown New York, NY
NYT Pick
I've spent decades helping government organizations use data to improve operations,
productivity, occasionally policy, and in recent years strategy. What I always encourage is using
data as part of a feedback system to improve future decisions. When data are only used up front
to determine a policy or strategy, that's insufficient. Continual data collection and analysis are
essential to determine if intended results are being achieved or if changes are needed to improve
results.
Corporations are already using "big data" to increase sales, and more powerful ways to analyze
more massive amounts of data may indeed help us get better public policies in the future,
though watch out for abuse and cherry picking data to fit pre-determined ideological policies.
But big data, even if not abused, will provide a relatively small advantage in making better policy
choices up front. Much greater benefits will come from having more data of adequate quality
flowing through well-designed feedback loops on a more timely basis so the users can keep
improving strategies, policies, and operations for maximum public benefit.
ProfWombat
Andover MA
NYT Pick
1. The most interesting recent example of use of data and its implications, far more than foul
shooting streaks, is Nate Silver's use of Bayesian statistical models to predict election outcomes
successfully, and the right's vehement rejection of his work and shock at its accuracy after the
fact, largely out of ideology and innumeracy. An odd omission, that, from a column about data,
its applications and limitations.
2. Examples abound of quantifying the unquantifiable. The US News college rankings come to
mind: a complex institution can't be ranked as greater than, less than or equal to another.
Single measures such as IQ and Spearman's G can't quantitate human intelligence, either.
3. Any given population gives rise to statistics. Any given individual within that population, at
any given time, is doubly an anecdote.
4. Perhaps the largest body of data with the most significant policy implications arise from
economics. Many economic models continue to exist despite a paucity of data for them. Again,
an odd omission.
Daniel12
Wash. D.C.
NYT Pick
Much of this is nonsense. Basketball players--indeed people in many fields--have hot and cold
streaks, inspired and bad days; bizarre interpretation to say a person's good days are just
random, accidental, like a coin just coming up heads repeatedly at a particular time, and that we
should look at a person's "career average" rather than judge by good days; any fool knows that in
the arts and sciences we do not go by any career average but the high points, the Mona Lisa
moments, the Wright brothers inventing the airplane etc. Plenty of artists do only a couple
significant things--we judge by that and not by any "career average". I suppose a newspaper
columnist though would prefer career average, the consistent turn out of average pieces, and to
call all exceptional accomplishment "just random noise", "pure luck", "not the person's usual
way of being", "a coin just coming up heads", "could have been me"...
As for students having
different learning styles, obviously some people are more verbal or spatial than average,
common sense demonstrates that fact; what we can argue about is "linear" and "holistic". Of
course we have little evidence for the latter two words being types of learning style because what
do these words signify in the first place? Are they as obvious as being verbal or spatial? Not at
all.
The problem in general is correct interpretation of data. We need more people with
brains. Data is increasing, but so often I see obvious errors in interpretation of data.
Martin
New York
NYT Pick
The idea that a conclusion is "objective" or reliable just because it appeals to measurable data is
totally bogus. Our capacity for gathering data has expanded so exponentially that it's nearly
meaningless--in the sense that I can find the data I want by crafting my question or search in a
way that suits my ends. The fact that political battles are carried out in the realm of media &
publicity intensify this tendency. That's why we see so many studies or investigations that ask
not what conclusion is consistent with the data, but what data will serve a pre-determined
conclusion.
And then when you inquire into anything that involves human behavior, you
encounter the fact that humans change their behavior in response to being studied. Life is not
just about measuring ourselves, but about deciding what we want to be.
Kurt
NY
NYT Pick
One of the most valuable lessons I learned in grad school was the imprecision of data and its
misuses. Mr Brooks says we are living in an age being driven by data, but it is probably just as
true that we are being driven by its misuse, in some ways just as contra-factual as any
superstition-ridden medieval society.
At the public level, we see summaries of academic
studies with stark results, frequently clearly pointing to specific courses of action or analysis. But
how much of that data providing such specificity rests on unproven assumptions or vague
wording of which we know nothing? Were those assumptions or weightings changed
infinitesimally, the results of the study could change dramatically, yet of that we know nothing,
only being spoon fed the preferred policy prescription, one that could be strongly inferred as
being the desired result all along.
Often data is so specific it lends a feeling of false control
and precision, when neither really exists, which is yet another problem.
Which points out
the need for transparency of process and peer review. And that process works well, so long as
everyone has not already descended into group think. But confirmation bias is a well noted
phenomenon on both sides in politics. Is it really wise to believe such does not exist in other
fields?
And how much of what we are told and believe is just as much a product of
preexisting dispositions and prejudices and how much is really susceptible to argument and
evidence?
ralph varhaug
houston
NYT Pick
Claude Shannon, the inventor of Information Theory made a distinction between data and
information: data was a fact, information was the basis of a decision. Data is not information
until we put it in context. A red light is a fact, a red light over a street intersection is the basis for
a decision and thus, information. We seem to believe that all data are equally valuable but data
without context is useless.
Phyllis Kritek
Half Moon Bay, CA
NYT Pick
Well done, Mr. Brooks. To add to the conversation I offer this quote from Einstein: ""Not
everything that counts can be counted, and not everything that can be counted counts."
By measuring things, we draw our attention to them, and hence not to other things. We may also
signal, inadvertently, that the data we present are exhaustive, that is, that we have completed the
exploration on this topic. Scientists often find their work summarized in the MSM in
descriptions that ignore the fundamental tentativeness of all scientific findings. Good science
always has explicit limitations and unveils the next looming unknown.
The most unsettling thing about datism, I think, is the assumption that by measuring something,
we actually capture its essence. Yet, what counts about it may not be something we can
count.
Easy Test: Can you measure your love for your child?
Download