In addition, a lot of scientific conclusions are drawn not just on interpretation of data already collected, but also on mathematical simulations and models that are constructed to try to describe how the system of interest behaves. These models are intrinsically even more malleable than data analysis...sometimes the constants that go into the equations (gravity, coefficients of friction, viscosities, densities, etc.) aren't known very precisely, and in many cases the equations governing the behavior of the system are nonlinear, which has important ramifications.
One of the first posts I made on this blog was eponymous- I went on for a bit about chaos. Unstable, nonlinear differential equations can lead to chaotic behavior in a variety of systems. One relevant hallmark of chaotic behavior is that small changes in the structure of the system at one point in time can have dramatic effects at some point later in time (e.g., the 'Butterfly Effect').
This dependence on small variations is, in theory, not all that terrible a problem. The difficulty enters, however, when you consider that these equations are nonlinear, and in general nonlinear equations can't be solved analytically - they have to be handled numerically, which means computers. Now... computers work in binary, and they have a limited number of digits that they can store in a given number at one time. So, to the computer, a curve that starts at x=1.001 is the same as one starting from x=1.0010000000000000000001. But, in a chaotic system, even that slight difference could balloon into a dramatic departure as you go forward in time. Thus, our predictive capabilities in chaotic systems are, in general, quite poor.
Weather is a highly nonlinear and chaotic system. One doesn't have to observe the weatherman's (attempts at) predictions for very long to realize that they're not always right. Storms disperse instead of dumping predicted feet of snow (unless you live in Buffalo), cold snaps come instead of balmy spring days, etc. Climate modeling, which is (to my understanding) somewhat easier than weather prediction, has some of the same problems, though by averaging temperatures (and precipitation and etc.) over longer periods of time you smooth out a lot of the jagged variations that you have to deal with in weather prediction. You still have a number of variables that you have to make guesses about—sometimes those guesses are fairly easy to make accurately, other times there's very little information to work from. And, because the system is sooo sensitive to the variations in these values, your results depend heavily on the values you choose.
So. Global warming. I can get behind the idea that a lot of the climate modeling that's being done predicts massive global temperature increases, oceans rising by feet per year, melting of permafrost, glaciers receding, Antarctica evaporating, etc. However, I would posit (with no way of actually demonstrating it) that if you take those exact same climate models and use different, but still reasonable, values of the input parameters, you could also predict a scenario where human CO2 emissions cause a minor to negligible variation in the natural up-and-down rhythm of global climate conditions.
Basically, don't take either side as gospel. The doomsayers are probably using values at one extreme of the spectrum; the everything-is-finers are probably using those at the other end. I expect the actual impact lies somewhere in the middle: we're affecting things, but not in the apocalyptic fashion that so many people are so hyped up about.
Now, the comparison. I've poked around on the blog of former Harvard geology professor (I think, I couldn't find an actual title on his CV), who recently posted on this same topic, but in regard to the hype twenty years ago about the threat of nuclear winter from hypothesized moderate use of tactical nukes and how it compares to the global warming debate today:
Climate modelers must often rely on educated guesses, but stringing together dozens of ‘what ifs?' as Sagan’s cohort did , runs a fatal risk of the ‘Garbage In, Gospel Out ‘ syndrome In the original ‘nuclear winter’ model this meant glossing over thirty nuclear war ‘scenario’ variables , using ‘worst case ‘ values for the lot of them . In practice, Sagan ended up telling a systems programmer to simply turn off the sun off like a light bulb, and leave it that way for a Biblical forty days and forty nights. Whereupon the temperature of the model’s featureless, ocean-less, 1-dimensional Planet Earth plummeted to forty below zero.
Climate models are what you make of them- they can serve equally as real scientific tools or political toys.” Nuclear winter” began with a premise uncontroversial as CO2’s ability to trap heat, Nobel Laureate Paul Crutzen’s observation that it tends to be cooler in the shade, even of a mushroom cloud. But environmental artists hired by Sagan turned Crutzen’s Ambio article “Twilight at Noon” into a dark vision of a frozen planet. To climatologists who understood the devilish details, Sagan’s model-based-- and biased-- megahype was an unfunny joke, played at the expense of their credibility on the eve of the global warming debate.
So. What it really comes down to is, "[t]he best kept secret of the Science Wars is how little both sides know." Yes, global warming is something worth paying attention to. Whether it's worth imposing emissions restrictions that would make a fair chunk of energy and industrial production totally economically infeasible... that depends on the simulations, and there are as many interpretations of those as there are people to interpret them. Makes it something of a tough call...