top of page


by John Spritzler

July 24, 2023


The URL of this article for sharing it is

[Also please read "THE 97% CONSENSUS ON GLOBAL WARMING?"]




The predictions of catastrophic global warming are invariably based  on one or more so-called "climate models." What's a climate model?


A climate model is, in essence, a very complicated mathematical formula. This is what it does. It takes, as its input, a very large number of numbers given it as


a) values of the magnitudes of various physical aspects of our planet at a given time (such as the temperature at thousands of different locations**, and also the wind velocities, cloud cover, C02 and water and methane, etc.  concentrations, and on and on, plus numbers to indicate astronomical facts about the sun and the earth's orbit),


and as


b) numerically specified assumptions about how and how much these many physical aspects of our planet at that given time affect the values of all of the above physical aspects of the planet at a fairly close time* in the future. Note that one of these assumptions is about exactly how and how much C02 affects temperature at various places on the planet or in its atmosphere.


The mathematical formula can be used with these numbers inserted into it to produce predicted values for all of these physical aspects of the planet at a fairly close time* in the future.



The way the 'simulation' game works is this:


First start by giving the model (inserting into the formula) the huge number of numbers described above. 


Second, use the model/formula to calculate the new values of all the planet's (and perhaps even the astronomical) physical magnitudes at the next moment shortly in the future.


Third, start over by using these new values of the planet's (and perhaps astronomical) physical magnitudes and the same earlier values that specify assumptions about how these physical magnitudes affect their future magnitudes, and use the model/formula to predict new values for the next moment shortly in the future.


Repeat this for as many steps (LOTS of them!) as one likes, until the new time point is far enough in the future to be of interest, at which time one takes all of the predicted temperatures at the different places on the planet and uses them to calculate a value that will be declared to be "the mean global temperature."

If the mean global temperature that is predicted by this procedure, when it is assumed that C02 levels in the atmosphere are what they will likely be if we don't reduce fossil fuel burning as much as the alarmists tell us we must,  is alarmingly high and your funding is from people who are climate warming alarmists, then publish that result.

If the mean global temperature predicted by this procedure is not alarmingly high and your funding is from people who are climate warming alarmists, then do not publish that result.


Here's how climate models gain whatever credibility they have. The scientist who publishes based on a model argues that the model "works" because when it was given input of a) physical magnitude numbers from a time in the DISTANT past and b) assumption values (specifying how the physical values affect their future values) chosen by the scientist, it did a pretty good job of correctly predicting the physical values (i.e., "observations") at a time in the more RECENT past.

Why did the model do such a good job of using the DISTANT past data to predict the more RECENT past observations?

The answer is this. First the scientist used some starting guesses for the assumption values that specify how the physical values affect their future values. Then the scientist ran the model to see how it predicts the recent past observations, and if the predictions are not good (not close to actual observations) then the scientist starts over again using new, changed, assumption values.

The scientist keeps trying new changed assumption values--'tinkering with them'--until he or she finally manages to find some  assumption values that cause the model to make a pretty true prediction from the distant past data to the more recent past time point's  observations. When this happens the scientist says, "See! My model is a very good model that can predict the future accurately."


Here's the problem that the global warming alarmist climate modelers never want to talk about:

“It's tough to make predictions, especially about the future.”

― Yogi Berra

Just because a model has been 'tinkered with' to make it predict the recent past fairly accurately from data about the distant past, does not at all mean that it will be able to predict the future from data about the present even close to accurately. This is just a hard cold fact. It is the reason why climate (or weather) predictions of more than a couple of weeks into the future are notoriously unreliable.

By the way, if a climate model makes predictions about future years, and for some of those years the predictions are very good (and of course make the news) while for other of those years the predictions are just wrong (and of course do not make the news), the very good predictions do not AT ALL indicate that the assumptions (about how the various physical magnitudes, of C02 etc., affect their values in the future) used by the model are correct or even close to correct.


If the model assumptions were correct then ALL of the predictions for all of the future years would be very good. The reason only SOME of the years were predicted very well is likely essentially the same kind of reason why a broken clock gives the PERFECT time twice a day; you don't want to rely on that clock to keep an important appointment, do you? Nor do you want to rely on that model to decide to stop using fossil fuel, do you?


Yes, models are used in medical research. The mantra that we biostatisticians are taught, however, is this: "All models are wrong, but some are more useful than others." (See the source here.)

There are some key differences between medical research models and climate prediction models.

Medical research models are MUCH simpler than climate prediction models. Medical research models are selected that assume (based on past and current observations) some fairly simple general type of relationship holds between a handful*** of physical magnitude numbers (a patient's age, weight, etc., known as the explanatory variables) and some quantity of interest (say, cholesterol level, or probability of dying by a given time, etc.). 

The 'game' in medical research is NOT to run a simulation to predict something far in the future. Rather, the 'game' is to determine from the known covariates and the OBSERVED quantity of interest what is most likely the best numerically specific VERSION of the assumed GENERAL type of relationship (remember, it's wrong, but some assumed general types of relationship are more useful than others!) between the covariates and the quantity of interest.


For example, one might start by assuming that the general type of relationship is that at time t the expected value of a person's cholesterol will be some unknown value plus some other unknown number times t plus another unknown number times the person's age, etc. One then uses the known covariates and known levels of cholesterol in many patients to make a 'best guess" as to what are the values of the above mentioned unknown numbers, which then determine the numerically specific version of the assumed general type of relationship


The question (that is answered by knowing the numerically specific version of the assumed general type of relationship) is typically whether (or how much) some covariate of interest (say getting, or not getting, an experimental treatment) affects the quantity of interest. For example, if the best guess for the unknown number to multiply the person's age by is greater than zero, this would mean that the model-fitting indicates that getting older is associated with having a higher cholesterol level.

If the result of this "model fitting" is to select a specific model that, say, shows the effect of getting the experimental treatment (as opposed to getting a placebo treatment) to be having a better clinical outcome, then there will typically be further clinical trials on an entirely new and larger set of patients to see if this same result is observed. If so, this will be taken as evidence that the experimental treatment is efficacious...until, perhaps, evidence to the contrary emerges.

Consider the contrast between what the climate warming modelers do versus what medical research modelers do


The climate modelers use a climate warming model to make extremely costly, even dangerous, decisions such as to end the use of fossil fuel, based on a model that one has confidence in based only on its ability to predict values correctly because the modelers kept tinkering with it to make sure it would do so, a model that has not been proven to be able to make accurate predictions of the FUTURE--the future being the ONLY thing that the modelers' draconian decision is based upon.

In contrast, medical researchers use models to draw conclusions about presently observed patients, to see if some factor (such as the treatment a patient received) is associated with some clinical outcome of interest. This is not about making a very complicated prediction of the future by simulating with millions of steps from one moment to the predicted next moment, using an assumed model.


Thus medical researches choose a "best fitting" model to draw a tentative answer a question (e.g., Is getting the experimental treatment instead of the placebo treatment associated in the currently observed people with a better clinical outcome?) about presently observed patients, and then keep seeing if that conclusion still holds by seeing if subsequent NEW sets of patients confirm that conclusion. And while doctors' decisions affect their patients for better or for worse, they don't come close to having the draconian consequences for the entire world's population that the climate modelers seek to impose on us, based on models purporting to predict the future, models in which we have little reason to be confident despite how well they may have predicted the already known recent past by being tinkered with until they did so.


* There is a necessary trade-off in deciding how close in the future to make predictions. If one predicts for an extremely short time in the future (say, 30 minutes), then the predictions (if, of course, the assumed values are correct--a BIG IF) will be more accurate than if one predicts for a longer time (say, 3 hours) in the future, because the physical values are constantly changing, not remaining the same even for a short period of time. So, predictions will be more accurate if they are made for very short times in the future. 

But if the predictions are made for an extremely short time in the future, then vastly more of them (in the simulation) must be made to get to a prediction for an interestingly large time in the future. The more intermediate predictions must be made (we're talking maybe millions or billions of them) the longer it will take to run the simulation, even though a fast computer is doing the arithmetic. It could take days, or weeks, or even months of computer running time.

So a trade-off must be made: better (assuming the Big IF) predictions but the need to wait a VERY long time to get them, or vice versa.

** There are an infinite number of different locations on the planet, so no matter how many physical magnitude numbers one uses, it is never enough to even theoretically make a truly correct prediction. And even tiny errors, in a simulation using wrong values to predict wrong values over and over again for millions of steps, can lead to VERY wrong values. The more physical magnitufed numbers one uses, of course, the better (assuming the BIG IF, that the assumptions numbers are correct) the predictions will be, but also the longer it takes for the calculations to happen, and hence the longer a simulation will take. Furthermore, it is just too costly to measure the temperature (etc.) at far more locations than currently is done. It is inevitable that little errors will lead to big errors.

*** Even medical models that work with genetic information, despite dealing with a very large number of genetic data, are simple compared to what a climate model has to handle.


bottom of page