Now that we've seen which features 
distinguish inductive arguments from 
deductive arguments, we want to look at 
the different kinds of inductive 
arguments one by one. And the first two 
we're going to look at are moving up from 
a sample to a generalization, and then 
moving back down from the generalization 
to some kind of prediction in a 
particular case. 
Generalizations are all around us. 
Almost all movies have credits. 
Most popular bands have drummers. 
Many restaurants are closed on Mondays. 
three quarters of police officers are 
very nice people. 
Two thirds of books have chapters in 
them. 
Half of the people I know like to play 
sports. 
And my favorite of all, 
87.2%. 
The statistics are made up on the spot. 
It should come as no surprise at all that 
there's so many generalizations filling 
our lives. 
Because generalizations can be extremely 
useful, especially when you need to make 
a decision. 
So, for example, if you're feeling a 
little nauseous, that is a little sick at 
your stomach, then you need to know 
whether most people who have symptoms of 
your sort have something serious enough 
that they need to go to a doctor. 
And they also need to know whether 
doctors can usually help people who have 
symptoms of that sort. 
All of those generalizations are relevant 
to whether or not you want to go to the 
doctor to see if you can get some help 
with your sick stomach. 
But then the problem is, how can we 
decide which generalizations to believe? 
Since there generalizations, and you pile 
all instances of certain sort, you can't 
check them all out. 
You have to take a sample of some sort. 
I mean, just imagine that your running a 
bakery. 
And, your making jelly doughnuts and you 
want to know whether the jelly doughnuts 
are filled with the right amount of 
jelly. 
You can make hundreds of jelly donuts. 
You're not going to tear them all apart 
and check every one for how much jelly 
they have in the middle, because then 
you'd have no donuts left to sell to your 
customers. 
Or imagine, that you want to buy a car. 
This time you're the customer. 
You want to buy a car and you need to 
know how often these cars had problems in 
the first year. 
Because if they have problems a lot 
during the first year, you don't want to 
buy that kind of car. 
But you don't want every car of that sort 
to been tested for the first year because 
then you can't buy a new car, you're only 
going to have a used car because they've 
all been used in the test. 
Or, imagine that you want to know what 
types of trees grew during a certain 
period of history. 
So, you look at the soil and you dig 
down, and you start looking for how many 
pollen grains there are, and a certain 
level that indicates a certain age. 
Well, you can't check every spot in the 
field because if you did that, you'd just 
be destroying the entire field. 
So, in these cases and many other cases, 
you just have to take samples. 
You can't test the whole class and then 
you have to generalize from the sample to 
the larger class. 
All of these generalizations from samples 
share a certain form. 
First, we look at one instance of the 
sample. 
We say that first F is G. 
That is the thing fats in the, fits in 
the class F, and has the property G. 
And we check another. 
The second F is also G, and the third F 
is also G. 
And all the rest of the Fs in the sample 
are G. 
So, we conclude that all Fs are G. 
Now notice that all in the conclusion 
means everything in the class of Fs. 
It doesn't only mean the ones in our 
sample. 
So we've started from premises that are 
only about the sample, which is only a 
part of the general class. 
And we've reached a conclusion about the 
whole class when we say that all Fs are 
G. 
Now, other arguments of the same general 
type are a little bit different. 
Because you don't always get complete 
uniformity. 
You don't always get all Fs or G. 
Sometimes you get the first F is G, and 
the second F is G, but the third F is not 
G. 
And the fourth F is G, and the fifth F is 
G, but the sixth F is not G. 
So you get, like, two-thirds of the Fs 
are G. 
And then, you reach the conclusion that 
two-thirds of all Fs are G. 
Again though, you've only looked at the 
Fs in the sample. 
You've only observed a small part of that 
total class of all Fs and you draw a 
conclusion that two-thirds of overall 
class, the whole class of Fs has that 
property, G. 
So, that's why you have a general, that's 
why it's called a generalization. 
You start from a smaller sample, and 
reach a conclusion about a much larger 
class. 
You generalize to the whole class. 
The fact that the argument moves from a 
small part of the class to the whole 
class shows that it's inductive. 
And what does that mean? 
Well first, is the argument valid? 
Think back to one of the examples. 
Almost all bands that I know of have 
drummers. Therefore, almost all bands 
have drummers. 
Well, is it possible that the premises 
are true and the conclusion false? 
Of course it is. 
Because maybe I only listen to bands that 
have drummers, but there are a lot of 
bands out there that don't have drummers, 
and I just didn't happen to listen to 
them. 
That is, they weren't part of my sample, 
okay? 
And is the argument defeasible? 
That means, that you can add further 
information to the premises and it'll 
make the argument much less strong. 
It might even undermine the argument 
totally. 
And of course, it could because you could 
give me all kinds of examples of bands 
that don't have drummers. 
And then, I would have to change my 
conclusion that almost all bands have 
drummers. 
So, this argument is defeasible. 
Does that mean that it can't be strong? 
No. 
It could still be strong. 
It could come in different types of 
strength. 
How would you get a stronger argument or 
a weaker argument in this case, 
while you can have a larger or smaller 
sample. 
If I take a very large sample, the 
argument's going to be stronger. 
If I take a very small sample, the 
argument's going to be weaker. 
So it's not like validity, which is on or 
off. 
It's either valid or not. 
Instead, the strength of the argument 
comes in degrees depending on how big the 
sample is. 
Now, does that mean that since it can't 
be valid, it can only be strong to a 
certain degree that it's no good? 
No. Because it's an inductive argument. 
Inductive arguments don't even try to be 
valid, they don't even pretend to be 
valid. 
That's not what they're supposed to be. 
And sob you can't criticize this argument 
by saying it's not valid. 
It's doing everything that it's supposed 
to do if it provides you with a strong 
reason, 
and a strong enough reason for the 
conclusion. 
That's what an inductive argument is 
supposed to do so if this argument does 
it, [SOUND] it's a good argument. 
Then, even if inductive arguments, 
including generalizations from samples, 
don't have to be valid. Tthey still do 
need to be strong. 
And so, we need to figure out how to tell 
when a generalization from a sample is a 
strong argument, when it provides a 
strong reason for the conclusion. 
And to really answer that question, you 
got to turn to statistics an area of 
mathematics. 
And learn how to analyze the data much 
more carefully than we'll be able to do 
here. 
But even mentioning the name statistics 
raises fear in some people. 
Mark Twain is famous for having said, 
there are three kinds of lies. 
There are lies, there are damn lies, and 
there's statistics. 
He actually gives credit to Israeli for 
having said it first. But the point is 
that statistics, are even worse than damn 
lies, you can't trust them at all, or so 
Mark Twain says. 
But on the other hand, some people say, 
statistics, that's math. It's all about 
numbers, you can't question that. 
Here's one example of somebody who 
suggests such a position. 
There's still one thing that's 
irrefutable, 
and that is the numbers. 
Numbers don't lie. 
You can't parse pull numbers. 
They're infallible. 
Okay, fine. 
He knows that statistics aren't 
infallible. 
He's just being sarcastic. 
But there are many people who put a lot 
of faith in the numbers and in 
statistics. 
They seem to think that you always have 
to trust it when a statistician comes up 
with an answer. 
And we're going to learn that neither of 
these views is correct. 
It's not true that statistics are always 
worse than damn lies, and it's also not 
true that the numbers are irrefutable. 
The truth lies somewhere in the middle.