Analysing change in proportion of species over time

[bmnice]

Dear all,

  I have data on number of certain species in a rotational grazed plot and also the number of same species in a continuosly grazed one collected for 8 years. The idea is to check whether there is a change in their proportion as year progressed. I have tried repeated measures, but was discouraged. A suggestion to compare proportions was put forward, and i don`t know how to test significant differences in proportions between the two plots. Part of the data looks like this:
                               year  Species  Continuous  Rotational  Prop_Cont  Prop_Rot    1992  AB  440  701  0.0484  0.0842    1992  BE  110  165  0.0121  0.0198    1992  BI  308  218  0.0339  0.0262    1992  CP  764  1571  0.0840  0.1886    1992  CV  296  428  0.0325  0.0514    1992  DM  927  681  0.1019  0.0818    1992  ES  114  131  0.0125  0.0157    1992  ESP  531  659  0.0584  0.0791    1992  FB  810  739  0.0891  0.0887    1992  HC  3825  1624  0.4206  0.1950    1992  PN  42  37  0.0046  0.0044    1992  RN  135  47  0.0148  0.0056    1992  SI  177  73  0.0195  0.0088    1992  TB  247  559  0.0272  0.0671    1992  TT  348  694  0.0383  0.0833    1992  UM  21  3  0.0023  0.0004    1992  Total  9095  8330        1993  AB  1031  1270  0.0328  0.0464    1993  BE  8049  4673  0.2557  0.1706    1993  BI  184  318  0.0058  0.0116    1993  CP  680  1333  0.0216  0.0487    1993  CV  2163  1679  0.0687  0.0613    1993  DM  372  444  0.0118  0.0162    1993  ES  71  128
 0.0023  0.0047    1993  ESP  709  474  0.0225  0.0173    1993  FB  3913  3946  0.1243  0.1441    1993  HC  4063  1925  0.1291  0.0703    1993  PN  4558  4891  0.1448  0.1786    1993  RN  408  367  0.0130  0.0134    1993  SI  171  520  0.0054  0.0190    1993  TB  4481  4803  0.1423  0.1754    1993  TT  160  406  0.0051  0.0148    1993  UM  467  209  0.0148  0.0076    1993  Total  31480  27386


  How do I show that for example there has been a signicant change in proportion of species AB, in 1993. I want a statistical method of dertemining that, especially using proportions.

  Thanx for the help


  ===================================================================
  It is only with the heart that one can see rightly. What is essential is invisible to the eye.
Mr B Moyo
  Fixed  [phone]
  Mobile [phone]
  Fax     [phone]
  ===================================================================



---------------------------------
Be a better Globetrotter. Get better travel answers from someone who knows.
Yahoo! Answers - Check it out.

[Bob Walker-2]

Hi Bethwell,

Couldn't you use multivariate GLM? Yes, you have repeated measures, but in his case there is a significant intervening variable, which is time. Time should, therefore, be treated as a factor in your design. The data looks to be:

DATA LIST FREE / YEAR (F4) SPECIES (A3) CONT ROT PCT_CONT PCT_ROT.
BEGIN DATA
1992,AB,440,701,0.0484,0.0842
1992,BE,110,165,0.0121,0.0198
1992,BI,308,218,0.0339,0.0262
1992,CP,764,1571,0.084,0.1886
1992,CV,296,428,0.0325,0.0514
1992,DM,927,681,0.1019,0.0818
1992,ES,114,131,0.0125,0.0157
1992,ESP,531,659,0.0584,0.0791
1992,FB,810,739,0.0891,0.0887
1992,HC,3825,1624,0.4206,0.195
1992,PN,42,37,0.0046,0.0044
1992,RN,135,47,0.0148,0.0056
1992,SI,177,73,0.0195,0.0088
1992,TB,247,559,0.0272,0.0671
1992,TT,348,694,0.0383,0.0833
1992,UM,21,3,0.0023,0.0004
1993,AB,1031,1270,0.0328,0.0464
1993,BE,8049,4673,0.2557,0.1706
1993,BI,184,318,0.0058,0.0116
1993,CP,680,1333,0.0216,0.0487
1993,CV,2163,1679,0.0687,0.0613
1993,DM,372,444,0.0118,0.0162
1993,ES,71,128,0.0023,0.0047
1993,ESP,709,474,0.0225,0.0173
1993,FB,3913,3946,0.1243,0.1441
1993,HC,4063,1925,0.1291,0.0703
1993,PN,4558,4891,0.1448,0.1786
1993,RN,408,367,0.013,0.0134
1993,SI,171,520,0.0054,0.019
1993,TB,4481,4803,0.1423,0.1754
1993,TT,160,406,0.0051,0.0148
1993,UM,467,209,0.0148,0.0076
END DATA.
* Why not run GLM / Multivariate as follows.
GLM PCT_CONT PCT_ROT BY SPECIES WITH YEAR
  /METHOD = SSTYPE(3)
  /INTERCEPT = EXCLUDE
  /PLOT = PROFILE( SPECIES )
  /EMMEANS = TABLES(OVERALL) WITH(YEAR=MEAN)
  /EMMEANS = TABLES(SPECIES) WITH(YEAR=MEAN)
  /PRINT = DESCRIPTIVE ETASQ OPOWER PARAMETER LOF
  /CRITERIA = ALPHA(.05)
  /DESIGN = YEAR SPECIES.

This should isolate the main + interaction effects between species and time. HTH,

Bob Walker
Surveys & Forecasts, LLC
www.safllc.com


----- Original Message -----
From: Bethwell Moyo
Date: Monday, May 21, 2007 5:23 pm
Subject: Analysing change in proportion of species over time
To: [hidden email]

> Dear all,
>
> I have data on number of certain species in a rotational
> grazed plot and also the number of same species in a continuosly
> grazed one collected for 8 years. The idea is to check whether
> there is a change in their proportion as year progressed. I have
> tried repeated measures, but was discouraged. A suggestion to
> compare proportions was put forward, and i don`t know how to
> test significant differences in proportions between the two
> plots. Part of the data looks like this:
> year Species Continuous
> Rotational Prop_Cont Prop_Rot 1992 AB 440 701 0.0484
> 0.0842 1992 BE 110 165 0.0121 0.0198 1992 BI 308
> 218 0.0339 0.0262 1992 CP 764 1571 0.0840 0.1886
> 1992 CV 296 428 0.0325 0.0514 1992 DM 927 681 0.1019
> 0.0818 1992 ES 114 131 0.0125 0.0157 1992 ESP 531
> 659 0.0584 0.0791 1992 FB 810 739 0.0891 0.0887
> 1992 HC 3825 1624 0.4206 0.1950 1992 PN 42 37 0.0046
> 0.0044 1992 RN 135 47 0.0148 0.0056 1992 SI 177
> 73 0.0195 0.0088 1992 TB 247 559 0.0272 0.0671 1992
> TT 348 694 0.0383 0.0833 1992 UM 21 3 0.0023 0.0004
> 1992 Total 9095 8330 1993 AB 1031 1270 0.0328
> 0.0464 1993 BE 8049 4673 0.2557 0.1706 1993 BI 184
> 318 0.0058 0.0116 1993 CP 680 1333 0.0216 0.0487
> 1993 CV 2163 1679 0.0687 0.0613 1993 DM 372 444
> 0.0118 0.0162 1993 ES 71 128
> 0.0023 0.0047 1993 ESP 709 474 0.0225 0.0173 1993
> FB 3913 3946 0.1243 0.1441 1993 HC 4063 1925 0.1291
> 0.0703 1993 PN 4558 4891 0.1448 0.1786 1993 RN 408
> 367 0.0130 0.0134 1993 SI 171 520 0.0054 0.0190
> 1993 TB 4481 4803 0.1423 0.1754 1993 TT 160 406
> 0.0051 0.0148 1993 UM 467 209 0.0148 0.0076 1993
> Total 31480 27386
>
>
> How do I show that for example there has been a signicant
> change in proportion of species AB, in 1993. I want a
> statistical method of dertemining that, especially using proportions.
>
> Thanx for the help
>
>
> ===================================================================
> It is only with the heart that one can see rightly. What is
> essential is invisible to the eye.
> Mr B Moyo
> Fixed [phone]
> Mobile [phone]
> Fax [phone]
> ===================================================================
>
>
>
> ---------------------------------
> Be a better Globetrotter. Get better travel answers from someone
> who knows.
> Yahoo! Answers - Check it out.
>

 Bob Walker
Surveys & Forecasts, LLC
www.safllc.com

[Maguin, Eugene]

Bethwell Moyo,

Could you state your main aims more clearly. I ask because you say "The idea
is to check whether there is a change in their proportion as year
progressed." I think you are interested in plotting the proportion of all
species that are species XX on the two types of plots (rotational or
continuous) over the years of your study. So what you will have for each
species is two lines, one for rotational and one for continuous, with
proportion on the y axis and time as year on the x axis. Are you thinking
that for each species the lines diverge or converge over time? So that in a
repeated measures analysis, you would get a species by time interaction?

If all this is so, then I don't think you can do what you want because it
appears that you have only one sample per grazing plot type. An analogous
situation would be to measure the change in height of one boy and one girl
from ages 6 to 16.

Perhaps someone more knowledgable can comment on these next points. I seem
to remember that proportions can be used in between groups anovas but that
some type of transformation is recommended to equalize variance over the
different values of a proportion. I am also unsure how transformed
proportions would work in a repeated measures analysis. Lastly, and software
issues aside, is a normal distribution model optimal for this sort of data?

Gene Maguin