Prev: Semilog Bar Plots
Next: sendmail and href links?
From: Ante Lausic on 26 Feb 2010 21:55
I've been having a lot of good experiences with the Weibull function using both the wblfit, wblplot and probplot functions to do a whole bunch of crazy stufff but I have finally run into a wall.
I have a set of wbl data here:
0.65
0.88
1.46
1.69
2.06
2.36
2.59
2.62
3.59
3.91
4.76
4.92
5.23
5.34
5.69
5.86
6.34
6.38
6.53
6.53
6.68
6.82
6.88
6.98
6.99
7.29
7.31
7.48
7.95
9.26
9.83

When wblplot is run, three distinct zones can be seen very well. The most important part now is to get the a and b factors from this data. Using the proper censoring techniques, the first flat region has values of 9.09960, 1.27880 and the second larger sloped region has values of 8.28380, 5.96810. Now the last region is consists of the last 4 points. However, when I run wblfit here with censoring, I get values of 9.02900, 8.34950. This 8.34 value is basically the slope, if I understand this correctly, but when I look at the plot, I feel that it should be more like 1.5 as the first region was and definitely doesn't have a slope larger than the 2nd region!

I'm thoroughly stumped. Any help, please?

Ante Lausic
From: Ante Lausic on 28 Feb 2010 19:17
BUMP!

Any help is really appreciated.

Ante
From: Tom Lane on 2 Mar 2010 12:57

> When wblplot is run, three distinct zones can be seen very well. The most
> important part now is to get the a and b factors from this data. Using the
> proper censoring techniques, the first flat region has values of 9.09960,
> 1.27880 and the second larger sloped region has values of 8.28380,
> 5.96810. Now the last region is consists of the last 4 points. However,
> when I run wblfit here with censoring, I get values of 9.02900, 8.34950.
> This 8.34 value is basically the slope, if I understand this correctly,
> but when I look at the plot, I feel that it should be more like 1.5 as the
> first region was and definitely doesn't have a slope larger than the 2nd
> region!

Ante, I can't follow what you did. I could imagine you might conclude from
the Weibull plot that the values up to some point (maybe 2, 4, or 6) seem to
follow a Weibull distribution. You could censor the values at a point and
estimate parameters from the censored sample:


From: Tom Lane on 2 Mar 2010 12:59
Sorry, I am having a tough time today sending stuff out prematurely.

>> When wblplot is run, three distinct zones can be seen very well. The most
>> important part now is to get the a and b factors from this data. Using
>> the proper censoring techniques, the first flat region has values of
>> 9.09960, 1.27880 and the second larger sloped region has values of
>> 8.28380, 5.96810. Now the last region is consists of the last 4 points.
>> However, when I run wblfit here with censoring, I get values of 9.02900,
>> 8.34950. This 8.34 value is basically the slope, if I understand this
>> correctly, but when I look at the plot, I feel that it should be more
>> like 1.5 as the first region was and definitely doesn't have a slope
>> larger than the 2nd region!
>
> Ante, I can't follow what you did. I could imagine you might conclude from
> the Weibull plot that the values up to some point (maybe 2, 4, or 6) seem
> to follow a Weibull distribution. You could censor the values at a point
> and estimate parameters from the censored sample:

>> fitdist(min(x,6),'weibull','cens',x>6)
ans =
weibull distribution
a = 7.44495
b = 1.53388

But I don't understand how you fit the remaining sections. They would have
to be censored on both the left and the right in order to fit them
separately. There's no built-in support for that. Can you explain what
you're doing?

-- Tom


 | 
Pages: 1
Prev: Semilog Bar Plots
Next: sendmail and href links?