You are on page 1of 2

Hello everyone, we're now going to look at a two-way anova and we're going to use

this to evaluate the significance of our effects that we calculated in the design
of experiments. So, we've discussed the one-way ANOVA and now in this session we
want to extend this to a two-way ANOVA. And we want you then to be able to apply
this in order to determine the significance of the effects that you calculated from
our designed experiments. So let's just review here, as we learned the ANOVA
procedure's used to decide if a significant difference exists among the means of
populations with outcomes generated by different treatment combinations in an
experiment. And we use the ANOVA when we have one factor but more than two
different treatment levels. For our DOE's we had two factors and we only had two
different treatment levels. But two different treatment levels for two different
factors, that gave us four different treatment combinations. Now this makes our
analysis computationally very intensive, the ANOVA method is extensively used for
analyzing data from experiments and normally you take advantage of a computer
program to do these computations. What we're going to do is to go through our ANOVA
method and we're going to apply the results we have from our experiments so you can
see actually how simple the calculations are.
But in this course you're not going to be expected to do a lot of cell averages and
row calculations, that information would be given to you. But we would expect you
to draw the correct conclusion, be able to draw the conclusion if the effects are
significant. So just a reminder, our DOE example, we were designing lawn mowers, we
were looking at the response.
The response, the y-value, was the time to mow the grass and I think the yard was
10,000 square feet. We had two different factors, we had factor A blade angle 12
degrees and 16 degrees, in factor B was deck height 5 inches and 7 inches.
So each of these had a low and a high level.
And again these came from past models of lawn mowers, past experience. The
designers were looking for sensible values that would then allow them to make a
decision on what would be the best combination for an optimal mowing speed. So
we're looking again the best combination of blade angle and deck height in this
experiment. And what we found was, if you remember we had from when they're both at
the low level - level 1, so we indicated by 1 both the low level. Two different
replicates here and the response was 73 minutes. Both at the high level, called
that ab. It's plus sign, we ended up here, we didn't have any variation between our
replicates and we got 83. We noticed that in this experiment, from these four
different corners the best value here was the, shortest mowing time, was 48. So
here this was treatment combination a or treatment combination 2 where we have the
lowest mowing time. And what we're interested in then is, you know further, can we
use the effects of these experiments. So the effect of a we calculated as -1, so
what that would mean is when 'a' went from a low to a high level, the mowing speed
might drop by one minute. And b with, when the height was adjusted from a low to a
high level, so 5-7 inches, we would have a mowing speed increase of 11. So we also
had something new though and this was this the idea of interaction. So the
interaction of the blade angle and the deck height on the mowing speed. And we had
here a number 24. Now we want to know if these values or significance. And what's
important here we learned that if they are significant and either these factor
levels are significant, then it means we can use them to do our designs. However if
the interaction is significant then we have to disregard all of the results. If
there is an interaction we can't use our effects meaningfully. So we're going to do
the significance test with an ANOVA. Now we're going to go back down look at our
one-way ANOVA table where we summarized all the data. So again, this is from the
one-way ANOVA, we had different observations for different, here's our factor
levels one through A, different factor levels. And then we had different
observations, one through n, within each of those factor levels. And we would
calculate an average value for each of those factors and we would calculate an
overall average for all the data. And again the idea for the ANOVA is we were
looking for what we call the sum of squares, it's a representation of the
variability sum of squares treatment, so a representation of the variability due to
the different treatments. And we compare that to a sum of squares error which we
could calculate for each of the different treatments, the error within these
values.
It should all be the same right? And then we will some those all up to give the sum
of squares error. And we would compare those with the F statistic and if the sum of
squares treatment was significantly greater than the sum of the squares error, we
would then say that the means of the treatments were definitely different. For a
two-way ANOVA we also need to have our data and here we have it organized into a
very similar table. And please don't get worried about this table, we're going to
look at an example to just get the general idea. We have factor a, still over here
in the column. Factor a we have level 1, level 2, level 3 down to level a,
different levels. Factor B now we put across the top, level 1..... to level B. Now
in our experiment we only had two levels, one and two, for both of them. This is a
generalized table and like before we can take row averages, which will then relate
all the values we have when factor a is at level one, and column averages, which is
new. For example all the values averaged together when factor B is at level 1 or at
level j, whichever level you're at. And we have again as before an overall average
which takes into account every single point that we've measured. Okay so this is
the same organizations as before, this is just a larger generalized table. Cell
averages can also be calculated, here we see then for all of these values at a
certain factor a level and factor B level, we can get an average for factor A at 1,
factor B at 1.
Here we have for example, average when factor A is at level a and factor B is at
level j. So overall average aj. for that cell. Those are the cell averages, we also
can calculate row averages. Here we see this is a typo in our books, I fixed that
there. If you average all of these values, so overall one level A over all the
different factor B sign over all the replicants, we end up with an overall row
average, y1.. then gives us the overall average value for factor A at level 1.
Column average likewise too.
Here we have then the overall column average for B at a certain level, there j for
example. And you can calculate that, that would be then Y bar with a dot and then j
dot. Or here Y bar, if factor B is at level one, dot one dot. So again you're just
taking row, column averages and then finally our overall average that we have at
the end here is either equal to the average of all the row averages or of all the
column averages, which is the same value and overall value for all the data. Now
obviously we'd like a computer to do this for you but let's look at our example,
we'll do this number country just for our example. Here we are again, and look you
already have a lot of your data that you need. So here's the resulting averages for
each of the different combinations, right? And our two effects on our interaction
effect. Let's put this into our table. So factor A here on the left side, factor B
across the top.
Low level, so level 1, level 2, level 1 and level 2. We had here 72 and 74 at this
level 1 - level 1. And I gave us an overall cell average of 73. Comes directly from
our table, make sure to put the table at the top there. Likewise for them both of
the high level 83 and 83, overall cell average is 83. And so on forth and so forth.
There's our 48, where we have the fastest time, and here's our 60. Now we just add
in the row averages, so you take 73 plus 60 divided by two and you get 66.5. 48
plus 83 and we get 65.5 here. Column average is 60.5 and 71.5.
And an overall overall average then of 66. So much more simple, but then we're only
doing 2x2 experiments. Once you get into something more complicated of course it
really would help to have a computer. But there's no mysteries here, we're taking
the data we already had, in fact we already calculated that, and then just adding
these row and column averages. But let's not forget what we're going to do with the
ANOVA analysis and we'll do that next.

You might also like