Based on the article, these people were not considered "high risk", in which case the Police could have politely knocked on the door, informed the residents what was happening without ever drawing a weapon, or forcing anyone to the ground.
In the same way their decision making is binary, it appears their method of action is also binary - either take no action, or bring the big guns and go all out with a show of force and violence.
... but then they wouldn't get to stage a raid that justified their purchase of military surplus gear.
If they actually had probable cause to believe that it was a major criminal marijuana grow operation, that itself would probably be nearly sufficient reason to expect that the risk of armed resistance would be high enough to warrant the kind of raid actually conducted.
The real reason for the SWAT raid is because SWAT raids are cool. Look at the gear. Look at the guns. Looks like fun. In fact, it looks like more fun than anything in the average policeman's day job. Ever seen a coder pick the more fun, cooler approach rather than the more appropriate, boring tool that gets the job done with a minimum of fuss?
1) Some percentage of people > 0% who buy hydroponic equipment are using it to illegally grow marijuana. Depending on what percentage you assign to this, this is reason enough to go to 2.
2) Some percentage of substances > 0% that test positive for THC are marijuana. Again, depending on what percentage you assign to this, this is enough reason to go to 3.
3) The house is searched, and the suspects are determined not to be growing marijuana.
The huge problem with Bayesian statistics, is that incredibly difficult to assign these probabilities objectively for many real world scenarios. How many people who purchase hydroponic equipment in Johnson County, Kentucky are growing marijuana? 1%? 50%? 90%? You could analyze this to death, and you'd never be able to put an exact number on it.
1) You realize hydroponics has nothing to do with growing pot. 100 % of pot growers breath air. You don't use that to single them out.
2) You make sure the tested substance is actually pot. No test can be perfect, but it can be a LOT better than "sometimes it works".
3) You don't pretend you are military. Just knock the door, have chat with the white couple instead of pushing their face into the floor, pet the dog instead of shooting it, and have a look around for the pot growing operation instead of searching the entire house for something else.
It is undeniable that police do not behave under the presumption that a person is innocent until proven guilty, or innocent until it's been proven beyond reasonable uncertainty - it is purely for the courts to decide that.
Police, when they suspect a person of wrongdoing will treat that person as guilty as far as the law will allow them to. They can handcuff you, put you in jail, point a gun at you, interrogate you, etc. as long as they believe they have a reasonable suspicion (not certainty, not a great deal of evidence, just reasonable suspicion).
Given the number of false positives I've gotten with Bayesian spam filters, I'll take reasonable doubt over Bayesian justice ;)
Imagine that in Bayes World, the presence of a positive THC test updates, in a way that takes into account the probability that the test gives a false positive, a random variable G|* representing the probability that, given all available information, the suspect is growing marijuana. Similarly, the event of the resident's visit to a garden store updates G|*. [0]
Then, perhaps the Bayes City Police Department would use Bayesian Decision Theory [1] (perhaps with a custom loss function that takes into the account that a false positive for sending a SWAT team can actually cause needless loss of life!) to decide who to raid. The Judge (a trained statistician and data scientist) would supervise the algorithm and make sure that procedures are updated to constantly balance the case-specific "gut feelings" of the investigators with the need for a predictable system.
And let's keep rolling with this! In Bayes World, we never think of "guilty" or "not guilty" as a set fact. If there's any uncertainty in the case, that's taken into account quantitatively in sentencing. A jury of the accused's peers, all educated in statistics (because in Bayes World, the politicians dispute certain assumptions going into macroeconomic models, but they grudgingly agree that investments in basic income and education have positive expected value when weighted by the likelihood of those assumptions... but I digress...) choose certain coefficients for sentencing based on the human factors of the case.
But we don't live in Bayes World, so we get stories like this.
[0] http://lesswrong.com/lw/2b0/bayes_theorem_illustrated_my_way... [1] http://www.cse.buffalo.edu/~jcorso/t/CSE555/files/lecture_ba...