Spatial matrices

[Dario Maimone Ansaldo Patti]

Dear All,

I am constructing some spatial matrices. I have the following coordinates (where _ID identifies the region]:

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte _ID double(_CX _CY)
 1   4.238880806932661   51.0381795797637
 2  4.3705583790493465 50.836004194715464
 3   5.008810757595679 50.314774734858666
 4   9.048183441796569 48.539120289602224
 5   7.562275359577741  51.47982042591398
 6   9.812804864412303  54.18267278074142
 7   9.031162177003864  50.60312939159003
 8   9.159366625166365  52.76822315642757
 9  11.419101982885781 48.946844237635794
10  -7.910502751973481 42.756695468623235
11 -2.6162526114576328   43.0433556472304
12  -.6599335303805115  41.51954561052405
13 -6.1508402965923334  39.19149210853352
14  -.5548950769729665 39.401518880927945
15 -4.5754816643187635  37.46337461821722
16 -1.4853899956423013  38.00235670507524
17   2.912655060412556  39.57419360329185
18 -15.668155079521242  28.34153154991463
19  -4.782005139037015  41.75429637927728
20  -3.716852162382493   40.4950465162602
21 -3.0047329353222882 39.581044916387285
22  1.5289116704222079 41.798474451952394
23   5.289817422604617  43.58430558367101
24   .8387347371287599 44.338785949212514
25   4.542686675219991 45.517655466556306
26   2.715756068693901   50.4723663905693
27   6.347598245580899  48.17247256770211
28 -1.2573091563937455  47.31613047630117
29  2.5381327644238603 48.292024223497016
30  2.5033493032137306  48.70929509840815
31   9.769067362637838 45.620628170558334
32    8.70486347100189 44.264875054299885
33  14.840443948886977  40.85976362267415
34   6.002563143592553 52.289497160963705
35   4.610661457311046 52.097751734967886
36   6.286832650414219  53.05531323841722
37   5.410618051098782  51.45495894313609
38  12.766735405438116  41.98000397761924
39   16.61930407162165 40.984592383451606
40  11.126082947167264  43.45078957457015
41  11.038069574762444   44.5255464813964
42  16.347610749835038  39.06787919421982
43  -7.678641274716713  41.45752851199433
44  -8.131341102441208  37.24373918509147
45  -9.045001782531612 38.737506880267716
46  -8.014034985515947 38.490875924556505
47  -7.993077306851027    40.133696972864
48 -27.282535206945695  38.34807291397858
49 -16.951549883485765 32.751393510571525
50  -4.184661705559883   56.8395664536139
51  -6.694157393386353  54.60974329011341
52 -3.7637283805348782  52.33639746515705
53  -3.132137327606542  51.00022732220863
54  -.8065406679657919 52.926936493964085
55  -1.229212191489712  53.96554242684258
56 -2.2708002631811794 52.480446124299874
57   7.861511916378289 45.137521996830564
58  11.884422926292736  46.03477268799009
59  12.831751280511003   43.1676278995571
60  14.854738234591938  41.45734716847926
61  11.626845963292368 38.819447705708036
62  -5.337484002822747  43.26057900387312
63  -1.929712993209706   42.5393129401837
64   9.556973106152082  53.42845954603848
65   7.392801608484208  49.85361696167447
66  -2.342632760789819  54.83099522558365
67 -2.5550432219906885 53.548812742891116
68    .955481263001568  52.67083597810112
69 -.14736474261673754  51.61526791744084
end

While I could use spmatrix to create my matrix, the user written spwmatrix routine would allow me to generate a matrix choosing the k nearest neighbours. With the data above, I could write:

spwmatrix gecon _CY _CX, wn(myW) knn(5)

However, if I try to use the myW matrix with spregress I get the following error:

weighting matrix myW not found

I presume that this is due to the fact that spregress does not recognize any spatial object which has not been created using spmatrix.

I tried to generate the above matrix and then to import it using spmatrix import:

spwmatrix gecon _CY _CX, wn(W5) knn(5) xport(myW, txt) replace
insheet using "myW.txt", delim(" ") clear
spmatrix import myW2 using myW.txt

The problem is that the matrix myW2 is not normalized (spectral):

. spmatrix summarize my

Weighting matrix  my
---------------------------------------
           Type |               custom
  Normalization |                 none
      Dimension |              69 x 69
Elements        |
   minimum      |                    0
   minimum > 0  |                    1
   mean         |             .0724638
   max          |                    1
---------------------------------------

Any idea on how to normalize the matrix? Thanks in advance.

[Rafal Raciborski (StataCorp)]

Here is one way.

spwmatrix gecon _CY _CX, wn(myW) knn(5)

mata: W = st_matrix("myW")
mata: id = 1::cols(W)

spmatrix spfrommata spW = W id
spmatrix summarize spW

[Dario Maimone Ansaldo Patti]

Thanks a lot Rafal!

[Dario Maimone Ansaldo Patti]

Dear Rafal,

Unfortunately, I have another problem. Basically, I need to do the inverse process, too. Suppose I generate a contiguity matrix with spmatrix, i.e.

spmatrix create contiguity cW, replace

Then, I export it in mata

spmatrix matafromsp W id = cW

Finally, I would like to import it back using spwmatrix. I read the help file for spwmatrix and it should be possible to import a Matrix from mata, if I am correct.

I tried:

spwmatrix import using W, wname(cWspx) mataf

In this case I get:

file W.raw not found

Which is the problem with my code?

Thanks for your help.

Dario

[Rafal Raciborski (StataCorp)]

spwmatrix spfrommata and spwmatrix matafromsp should do the job, for example:

gen id = _n
spset id, coord(_CX _CY)
spmatrix create idistance spW, norm(none)
spmatrix summarize spW
mata: mata des

spmatrix matafromsp W id = spW
mata: mata des

spmatrix spfrommata newW = W id
mata: mata des

spmatrix summarize spW
spmatrix summarize newW

[Dario Maimone Ansaldo Patti]

Thanks Rafal. Do you mean spwmatrix spfrommata (as in the first line of your reply) or spmatrix spfrommata (as in the code lines you included)?

As in my post, I would like to generate a Matrix using spmatrix. Then, I would like to export it in mata and, finally, to import the mata object using spwmatrix.

The reason is that I want to use genmsp programme and spatlsa to plot hot-cold maps. However, spatlsa does not support matrix generated using spmatrix.

[Rafal Raciborski (StataCorp)]

You can move between Sp and Mata using spmatrix matafromsp and spmatrix spfrommata, there is no need to use spmatrix import.

[Dario Maimone Ansaldo Patti]

Thanks Rafal. I was non aware of this option for spwmatrix.

[Rafal Raciborski (StataCorp)]

I meant spmatrix, edited the previous post.

[Dario Maimone Ansaldo Patti]

YThanks. however this does not solve my question. Is it possible to import a mata object using spwmatrix? In fact, I am not using spmatrix import.

The steps I am following are:

a - generate a spatial Matrix using spmatrix

b - exporting the Matrix in mata using spmatrix matafromsp

c - importing the mata object using spwmatrix

[Rafal Raciborski (StataCorp)]

You can copy a Mata matrix into Stata, but you have to rename the rows and columns of that matrix so it looks like a matrix created by spwmatrix. Below I use the first 11 observations from the data you posted to illustrate.

keep in 1/11

set seed 12345
gen y = rnormal()
spwmatrix gecon _CY _CX, wn(myW) knn(2)
mat list myW
spatlsa y, weights(myW) moran

local row : rowfullnames myW
local col : rowfullnames myW

mata:
    W = st_matrix("myW")
    W = 2*W
    st_matrix("newW",W)
end

mat rownames newW = `row'
mat colnames newW = `col'

spatlsa y, weights(newW) moran

[Dario Maimone Ansaldo Patti]

Thanks for your reply.

Ideally, I would like to create the matrix myW using spmatrix. However, your code does not work if I generate the matrix using spmatrix.

keep in 1/11

set seed 12345
gen y = rnormal()
spmatrix create contiguity myW, replace 

local row : rowfullnames myW
local col : rowfullnames myW

mata:
    W = st_matrix("myW")
    W = 2*W
    st_matrix("newW",W)
end

mat rownames newW = `row'
mat colnames newW = `col'

spatlsa y, weights(newW) moran

Any possibility to construct a Matrix with spmatrix, export it in Mata and writing back in Stata using spwmatrix?

[Rafal Raciborski (StataCorp)]

gen id = _n
spset id, coord(_CX _CY)
spmatrix create idistance spW
spmatrix matafromsp W id = spW
clear
getmata r*=W, double
save tmp, replace
spwmatrix import using tmp, wname(WW) dta

[Dario Maimone Ansaldo Patti]

Thank you. It worked.

[harry sharma]

Hi,

I am running spatial fixed effect model (spxtreg) and I have created idistance weight matrix using _CX _CY.

spset PSUNO, coord(_CX _CY)
spset, modify coordsys(latlong, kilometers)
spmatrix clear
spmatrix create idistance M if year==2008
spmatrix summarize M

When I run fixed effect model, coefficient of all the non-lag variables are omitted. But same problem do not arise in the case of random effect.

I have no missing cell in my data set. Can you suggest me what is wrong in my data set and how to solve this problem?

Thanks

Hari