@red

I think I address that question in a subsequent post. If not, I can elaborate.

@ab, by all means elaborate.

Let's look an example.

Say I'm a circuit designer. I have a good understanding of circuits, but I'm designing something very complicated. In the past, I would have no choice but to build the circuit as best I could and run tests. If it was wrong, I may need to rebuild the whole circuit. This takes time and money.

Today, I can model my circuit with CAD. Each part, it's true, have been modeled using experimental tests, but *my* circuit hasn't. I can build my circuit, test it (much easier with CAD than on a breadboard, btw) and see if it works. If not, I make changes (for free because I'm not using parts) and then test again. When it looks like it works, I'll make an actual prototype to confirm, but I've saved myself a lot of time and money.

' abge
I didn't say that modelling was ''useless''. I do, however, say that it is (should be) an adjunct to experimental science, rather than the other way round. In my field modellers have insisted that the physical evidence is wrong when it stares them in the face, because it doesn't conform to the model.
You are right, however, to state that it is considerably cheaper and thus much more popular with the powers that be. See also virtual field trips, virtual microscope and similar 'cheap' abominations. There is a real danger that much vital accumulated practical skill and knowledge will be lost to the next generation.

@ful

What is your field?

I think we agree, though. We should be using both.

@abgemacht
I think that's an example from engineering and not science, but I catch your drift.
One thing I think is important is that both the experimentalist and the modeler be very precise in what the assumptions and inaccuracies are.
The world "model" says it: you're taking away things from reality to make reality simpler, but not necessarily more accurate. Which simplifications cause the deviations from the experimental result.
The same goes for the experimentalist: how often did you repeat an experiment before the value you obtained was calculated / determined. What is the spread in your results?
Then the two I think can communicate, otherwise it's just both people pushing their visions. You'd be the Tettleton of science. It could be dangerous.

@ful

And, in my example, of course if my prototype doesn't work when my model did I either:

1) Built it wrong
2) Have the wrong model

That problem would need to be addressed.

Yes, an engineering example was most readily available to me, as I'm an engineer. But, I think the point still stands.

I agree that models include simplifications, but if you're careful about the simplifications you use, it shouldn't be an issue. In fact, it's sometimes better because in experiments you can't always completely eliminate extraneous noise, but in a model you can. For instance, the Schrodinger equation can be modeled very well, but it is often a pain to make measurements without introducing error.

And of course, like anything else, communication is the key to science.

PS I totally agree with fulhamish by the way. You model an experiment, not the other way around.

Yes, but what is the point of modeling an experiment if you never do anything with that model?

OK, here's a example straight out of science:

While searching for the Higgs boson, experimentalists didn't just look throughout the entire universe. They took what they knew, made models and narrowed down the search. Then, they performed experiments. Turns out the models were wrong, but it's still better than randomly searching, no?

Oh my goodness, a dream come true. Background: prospective physics/music double major in undergrad pursuing a pre-med course of action.
Addressing the topic of computational modeling: The problem lies in that computational modeling relies heavily on human understanding. If there's a slight flaw in the theory, then the data would be wrong (of course). Now take experimentation. Experimentational results are not determined by understanding. Experimentation will always yield the correct results. Not necessarily the intended results, but accurate nonetheless. If at all possible, I think experimentation should always be the choice over computational modeling.

@Situation

That could not be more wrong. Experiments require a high level of understanding. Why? Because if you don't understand what you're doing, you're not going to test for it correctly!

@The Situation
I love the contrast between your nickname and your background, you really caught me off guard there

@ abge. No one said that you should set up an experiment without paying attention to your sampling methodology! :-)
In fact that is the first thing I look for when reading a paper. What effort have these guys made to ensure that their sample reflects the real world (i.e., population)?

@ful

What? I don't understand how that relates to what I said about the Higgs boson.

@abgemacht I believe The Situation (I can't even type that without laughing) is right, within the context of the experiment itself that is. If you measured the speed of light at a little more or less than 3E8 m per second, you obviously made a mistake. But if the dial is where it is, it is the correct outcome of an experiment. A flawed one, but an experiment nonetheless.
The same goes for models, but the difference is that once you control the experiment, you know the correct value. Even if you control the model, your value could still be wrong. (Been there by the way)

@ abge Experiments require a high level of understanding.

Surely over methodology? Heck if they understood the results and discussion before doing the experiment what would be the point? Situation is absolutely correct.

@red Performing the wrong experiment is a waste of time. Yeah, I can measure something uncalibrated, but it gives me no information at best and at worst it confuses me into thinking the wrong thing. If I had modeled the test before hand, I would realize me results was completely wrong, and that may help me debug the problem.

@ful

How do you determine where to search for Higgs Boson without modeling? There's no other way.

The point to modeling is that you can use the results of a simple experiment to give you approximations to the results of a complicated experiment. You can then use those results as justification for performing an expensive and complicated experiment.

I think that you guys are mostly younger than me. I would only advise that when your colleges/supervisors keep recommending modelling over experimentation there may well be other motivations than those based on pure scientific research at work. Calling a spade a dirty great shovel I mean finance! More is the pity because this is a very short-sighted view and in my view is the main ''problem'' with science, as raised in this thread.

Yeah... For those of you wondering, my name is referring to a character in the Jersey Shore. But I assure you I am not an accurate representation of that character.

Anyhow, to the topic at hand. Experimentation merely requires that you be able to interpret the results. (I suppose the word "merely" may not be a good choice to use). However, computational modeling requires that you have a flawless grasp of the theory. Indeed experimentation requires understanding, I'm just saying that computational modeling requires far more.
But I do agree that perhaps computational modeling may be used for guiding the experiments.

@ful

Again, what is your field?

An interesting area in computer modeling is monte-carlo simulations.

In reactor physics, many of the variables are known with quite a high degree of precision so MC simulations are very useful in areas where experimentation is impractical.

In general, I have seen a push to move to first principal simulations that don't rely on experimental correlations. As computers get bigger and better these sort of programs will help bridge the perceived gap between physical experiments and computer simulations.

Wow, every time I type a response, a few of you post while I type. Performing the wrong experiment still yields accurate data. It's the interpretation that would be inaccurate. At that point, it is indeed whether you understand the experiment or not. Modeling, however, will yield inaccurate data.

@Situation

In order to perform a useful experiment, you need to understand *how* to test for it. Being able to interpret results is not all that is important.

And, even if modeling requires a perfect understanding of theory, so what? Aren't we after knowledge. Shouldn't we be aiming towards a perfect understanding of theory?

@ abge
How do you determine where to search for Higgs Boson without modeling? There's no other way.

I do not think that anyone is saying modeling has no place are they?

However, using your H-B example that particular modelling required experimentally determined data inputs and the results of the modelling were subjected to experimental testing (as I understand it). It surely makes the case for preeminence of experimental work better than anything else so far raised?

@Abge. Interpreting the results is typically understanding the factors that lead to them. This is where multiple experiments comes in hand. When you vary different factors, and observe the results, the intended information becomes much easier to extrapolate (correct use of the word? Idunno...).
The problem with requiring a perfect understanding of theory is that you must take into account all factors (I suppose excluding negligible factors such as relativity with regards to macroscopic kinematics) before the model is useful. How would you be able to determine whether you have all the factors taken into account?

Yeah and we're drifting off topic a bit too (at least, from the topic as I intended it :) ). I guess we can all agree that accurate models are needed to fit to accurate experiments.

So, let me rephrase a debating topic and see if anybody has anything to say about this:

***Compared to say ten, twenty years ago, have there been any fundamental differences in the way experimental and theoretical results were checked against each other? If confidence in results has shifted from experimental to theoretical results, what does that say about the State of Science? Is it a desirable development?
Subquestion: Is a Monte Carlo simulation a good compromise between doing an experiment and a calculation, or is it a third method that can operate indepedently from the two?***

I think someone mentioned they were a chemist? There are multitudes of modelling tools available for thermodynamics, but very few for kinetics in comparison. Why the disparity? The truthful answer is that thermodynamics is so much easier to model, but does that mean it can report the real world (population) to the exclusion of kinetics? Or is there something very wrong with the sampling methodology at work here?