R package for Two-Way Indicator Species Analysis (Hill 1979).
Two-Way Indicator Species Analysis was developed to classify community data tables. It was supposed to work in the same way as a traditional community ecologist in arranging a community data table.
TWINSPAN is available as a self-standing computer program that can be compiled to work on many platforms. This R package uses the same Fortran code, but allows using TWINSPAN from R together with other R functions for community ecology and statistics.
The design philosophy of TWINSPAN is completely different from well-behaved R functions. TWINSPAN is a traditional console program that runs through a process, and prints its results as it advances. In R, the function should work silently, and return the result for further analysis. The code needs a thorough re-design to be used in R.
You can install the current development version from GitHub using devtools package:
devtools::install_github("jarioksa/twinspan")You need development tools to install a source package. In particular, you need C and Fortran compilers.
If you cannot install a source package, you may try to install from my R repository in GitHub:
install.packages("twinspan", repos="https://jarioksa.github.io/drat/")This repository offers binary packages for Windows and MacOS in addition to source package. It is not updated to all development in GitHub, but only contains some snapshots of development. I have no tools to test if these binary versions are virus free.
The basic command to run twinspan is – unsurprisingly – twinspan:
> library(twinspan)
> library(vegan) # for data
> data(varespec)
> tw <- twinspan(varespec)This will run the analysis with default options, including the default cut levels.
TWINSPAN uses basically binary data, and quantitative data are broken into
pseudospecies by species abundances using argument cutlevels. The default levels
are c(0, 2, 5, 10, 20). Any species present in the data will be marked as species1
where the last digit shows the level of pseudospecies level. If that species is present
at abundance 7, TWINSPAN will also generate species2 and species3, because these
limits are exceeded. To see how data are transformed, you use command twinsform
that transforms the data similarly as twinspan:
> twindat <- twinsform(varespec)
> colnames(twindat)
[1] "Callvulg1" "Empenigr1" "Rhodtome1" "Vaccmyrt1" "Vaccviti1" "Pinusylv1"
[7] "Descflex1" "Betupube1" "Vacculig1" "Diphcomp1" "Dicrsp1" "Dicrfusc1"
[13] "Dicrpoly1" "Hylosple1" "Pleuschr1" "Polypili1" "Polyjuni1" "Polycomm1"
[19] "Pohlnuta1" "Ptilcili1" "Barbhatc1" "Cladarbu1" "Cladrang1" "Cladstel1"
[25] "Cladunci1" "Cladcocc1" "Cladcorn1" "Cladgrac1" "Cladfimb1" "Cladcris1"
[31] "Cladchlo1" "Cladbotr1" "Cladamau1" "Cladsp1" "Cetreric1" "Cetrisla1"
[37] "Flavniva1" "Nepharct1" "Stersp1" "Peltapht1" "Icmaeric1" "Cladcerv1"
[43] "Claddefo1" "Cladphyl1" "Callvulg2" "Empenigr2" "Rhodtome2" "Vaccmyrt2"
[49] "Vaccviti2" "Descflex2" "Vacculig2" "Diphcomp2" "Dicrsp2" "Dicrfusc2"
[55] "Dicrpoly2" "Hylosple2" "Pleuschr2" "Polyjuni2" "Ptilcili2" "Barbhatc2"
[61] "Cladarbu2" "Cladrang2" "Cladstel2" "Cladunci2" "Flavniva2" "Nepharct2"
[67] "Stersp2" "Callvulg3" "Empenigr3" "Vaccmyrt3" "Vaccviti3" "Vacculig3"
[73] "Dicrsp3" "Dicrfusc3" "Hylosple3" "Pleuschr3" "Polyjuni3" "Ptilcili3"
[79] "Cladarbu3" "Cladrang3" "Cladstel3" "Cladunci3" "Flavniva3" "Stersp3"
[85] "Callvulg4" "Empenigr4" "Vaccmyrt4" "Vaccviti4" "Dicrsp4" "Dicrfusc4"
[91] "Pleuschr4" "Ptilcili4" "Cladarbu4" "Cladrang4" "Cladstel4" "Cladunci4"
[97] "Flavniva4" "Stersp4" "Callvulg5" "Vaccviti5" "Dicrsp5" "Dicrfusc5"
[103] "Pleuschr5" "Cladarbu5" "Cladrang5" "Cladstel5" "Cladunci5"The original data of 44 species are extended to a matrix of 107 pseudospecies.
People often use the default cut levels which are rather good for data originally
expressed in cover percentages. If you have cover class data and you want to
preserve the original accuracy, you can give your data values. For instance,
for the dune data in vegan you may use twinspan(dune, cutlevels=1:9).
Use summary to see the classification:
> summary(tw)
1) eig=0.179: -Cladrang5 +Pleuschr4 < 1
2) eig=0.147: +Callvulg2 +Pleuschr3 < 2
4) eig=0.163: -Cladarbu3 < 0
8) eig=0.182: -Cetrisla1 < 0
16) N=1: 4
17) eig=0.169: +Callvulg1 < 1
34) N=2: 7 5
35) N=3: 18 6 3
9) N=4: 2 9 12 10
5) N=3: 13 14 11
3) eig=0.203: -Cetreric1 -Cladarbu2 +Cladstel2 < 0
6) eig=0.161: +Dicrsp2 < 1
12) eig=0.206: +Callvulg1 < 1
24) N=1: 23
25) N=4: 15 22 16 20
13) N=2: 24 25
7) N=4: 27 19 28 21 twinspan is divisive: it splits data into two parts at each step, and these steps are
described here. The splits are based on the first correspondence analysis axis of the
current subset which is still further polished to make the dichotomy clearer. The first
split is made at eigenvalue 0.179, and the pseudospecies best indicating this division
are Cladrang5 and Pleuschr4. Cladrang5 is an indicator of "negative" (or left or
first) group and Pleuschr4 an indicator of "positive" (or right or second) group.
These indicator species are summed up for every quadrat (or sampling unit: quadrat is
the term used in TWINSPAN). The last number after < gives the threshold score for
positive group: if the indicator score is less than 1, the quadrat is in the negative
group (2), and if it is 1 (or higher), the quadrat is in the positive group (3).
These groups are again divided with new correspondence analysis, and from group 2 you go
either to 4 (negative) or 5 (positive). With default settings, groups smaller than 5
items or deeper than 7 levels of divisions are not divided. First such final groups are 34
and 35. For these final groups, summary gives the size (N) and lists
the names of the members (in this case the names are numeric).
You can extract the classification of each quadrat with cut:
> cut(tw)
[1] 35 25 13 7 24 7 25 25 7 5 5 25 13 34 34 35 35 16 9 9 9 9 5 7
> cut(tw, level=2) # use classification at second level
[1] 4 6 6 7 6 7 6 6 7 5 5 6 6 4 4 4 4 4 4 4 4 4 5 7You can also predict the membership of quadrats based on the indicator pseudospecies
and threshold score. This can also be done with argument newdata using data set that
contains same species, but is not used in developing the classification.
> predict(tw, level=2)
[1] 4 6 6 7 6 7 6 6 7 5 5 6 6 4 4 4 4 4 4 4 4 4 5 4Please note that the last quadrat was classified to second-level class 7, but it is predicted to be in class 4. This can happen because TWINSPAN classification is based on the polished ordination, and the indicator pseudospecies only indicate this division. The analysis tries to make the concordance as good as possible, but it does not always succeed. Traditionally these quadrats are called misclassifications. In this case, the last quadrat was misclassified on the first step: the ordination put it into group 3, but indicator pseudospecies to group 2.
TWINSPAN stands for two-way indicator species analysis, and in addition to quadrat classification it also classifies the species (not the pseudospecies):
> summary(tw, "species")
1) eig=0.543
2) eig=0.41
4) eig=0.272
8) N=3: Diphcomp Cladamau Flavniva
9) N=4: Callvulg Cladstel Icmaeric Cladphyl
5) eig=0.122
10) N=4: Cladrang Cladchlo Nepharct Stersp
11) eig=0.068
22) N=4: Cladarbu Cladcocc Cetreric Cladcerv
23) eig=0.049
46) eig=0.048
92) N=10: Empenigr Vaccviti Pinusylv Polyjuni Pohlnuta Cladcorn Cladgrac Cladfimb Cladcris Claddefo
93) N=3: Cladunci Cetrisla Peltapht
47) N=3: Vacculig Polypili Cladsp
3) eig=0.405
6) N=3: Dicrfusc Dicrpoly Ptilcili
7) eig=0.231
14) eig=0.254
28) N=4: Descflex Pleuschr Polycomm Barbhatc
29) N=3: Rhodtome Vaccmyrt Hylosple
15) N=3: Betupube Dicrsp Cladbotr Species classification is based on correspondence analysis where species are weighted by their
indicator potential for the quadrat classification. You can extract the classification vector
with cut(tw, "species").
The data can be tabulated with:
> twintable(tw)
000000000000011111111111
000000000011100000001111
0000001111 0000011
011111 01111
00111
1 1111121212222122
475863292034135260457981
000 Diphcomp -1-2-1-1--1--------1----
000 Cladamau --11-1------------------
000 Flavniva 411111-1-1---1-------1--
001 Callvulg 2--1111-11522-12211-----
001 Cladstel 5112455555415111111-25-4
001 Icmaeric -11--------1----1-------
001 Cladphyl -------1-1-11-----------
010 Cladrang 555555524351533244323212
010 Cladchlo -1--1111-11--1-1--1--1-1
010 Nepharct -1-1-1-----------1-2----
010 Stersp 114111---111111-1-11-11-
0110 Cladarbu 454553112154334333231311
0110 Cladcocc 111111-1-111111111-1-11-
0110 Cetreric 111111-1-111--111111----
0110 Cladcerv 1-----1------------1----
011100 Empenigr 131432324413141333134314
011100 Vaccviti 132442354433454424444435
011100 Pinusylv 1-111-11111111111111-111
011100 Polyjuni 11111-11-111121-1113-111
011100 Pohlnuta 11-1-1-111111111111-1111
011100 Cladcorn 111111111111111111111111
011100 Cladgrac 111111111-11111111111111
011100 Cladfimb 111111111111-11111-11111
011100 Cladcris 111111111111111111111111
011100 Claddefo 111111111111-11111111111
011101 Cladunci 111121111115212212311111
011101 Cetrisla 1------111111-1-1-11--11
011101 Peltapht ---1------1-11-----11---
01111 Vacculig 13-1-1---------2-1-11-1-
01111 Polypili --111-------111--1---1--
01111 Cladsp 1--11--1--1----111-11---
10 Dicrfusc 111111111124124551-42111
10 Dicrpoly -11----1-1--111-111--112
10 Ptilcili ---11111--1-11111-111114
1100 Descflex 11-------------1---12111
1100 Pleuschr 111211123133445544555552
1100 Polycomm -----------1-------111-1
1100 Barbhatc ----------1--1------11-2
1101 Rhodtome ----------1----1----2-12
1101 Vaccmyrt -------1--11--132---3244
1101 Hylosple -----------------1-13-3-
111 Betupube ---------------1---1---1
111 Dicrsp --------1-1---111154--11
111 Cladbotr ----------1--1---11-1111
sites species
24 44 The strings of 0 and 1 in front of the species name and above quadrat name (or number)
give the steps of division. The numeric values in the table are the pseudospecies values
of the analysis.
The twinspan classification can be extracted as a standard R dendrogram. The final
units contain many branches (species, quadrats), and it is best to use fan-like trees:
> plot(as.dendrogram(tw, "species"), type = "triangle")
Hill, M.O. (1979) TWINSPAN: A FORTRAN program for arranging multivariate data in an ordered two-way table by classification of the individuals and attributes. Ecology and Systematics, Cornell University, Ithaca, NY.