Re: statistical differences between products evaluation using incomplete block design

Posted by Marta Garcia-Granero on Sep 29, 2011; 9:16am
URL: http://spssx-discussion.165.s1.nabble.com/statistical-differences-between-products-evaluation-using-incomplete-block-design-tp4852186p4852402.html

El 29/09/2011 9:53, Alexandra Chirilov escribió:

> Dear list,
>
> I have a problem...one of our client wants to evaluate 6 products in a CLT!
> We propose him a balance incomplete block design (BIBD) (3 products per
> respondents). I generate a quite balance design (D-efficiency = 0.8).
>
> I have never used before BIBD (I used only complete design) and I don't have
> any idea how to statistically compare the products' evaluation? What
> statistical test should I use in order to say that product A is better
> perceived than product B (e.g.: 8.97 significantly higher than 8.80)?
>

Hi Alexandra:

Time ago I managed to replicate the results for this balanced incomplete
block design I found in a book, using SPSS:

Fert. Bl.1  Bl.2  Bl.3  Bl.4  Bl.5
F1    94    96    100   92    -
F2    95    75    76    -     92
F3    76    100    -    97    98
F4    94    -    102    93    96
F5    -     75    91    86    95

* Dataset *
data list list/Block Fert Yield (3 F8.0).
begin data
1 1  94
2 1  96
3 1 100
4 1  92
1 2  95
2 2  75
3 2  76
5 2  92
1 3  76
2 3 100
4 3  97
5 3  98
1 4  94
3 4 102
4 4  93
5 4  96
2 5  75
3 5  91
4 5  86
5 5  95
end data.

*Analysis *.
UNIANOVA
   Yield BY Fert Block
   /RANDOM = Block
   /METHOD = SSTYPE(1)
   /INTERCEPT = INCLUDE
   /DESIGN = Block Fert.

It is important:

- Use SSTYPE(1) instead of the default method - SSTYPE(3)
- The order of both factors in "/DESIGN" (last line) is blocks first,
then treatments

HTH,
Marta GG

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD