1
$\begingroup$

I'm working with a large data set looking at climate variables to see whether they could be risk factor for a sheep disease in a country with dependent variable being disease count (cases) and also whether the factors could explain regional variations i.e why more cases in some regions. My problem is that the variation between independent variables seem too be too small making it harder to come up with best method of analysis. I have tried approach such as poisson but get error which likely not the best approach.Some suggested ANOVA which cant figure out how this would work. Any advise will be of great help.

Here is part of the data plus results when I tried poisson (whole data as region just brings all results to - NA cames in likely because of small variation on the variables:

    Region  Max temp    Min temp    total rain  Rain days   Cases
        1   13.45       6.05        922.9       143.9       0
        1   13.45       6.05        922.9       143.9       0
        1   13.45       6.05        922.9       143.9       0
        1   13.45       6.05        922.9       143.9       0
        1   13.45       6.05        922.9       143.9       16
        2   15.32       7.13        629.8       112.6       0
        2   15.32       7.13        629.8       112.6       0
        2   15.32       7.13        629.8       112.6       70
        2   15.32       7.13        629.8       112.6       127
        2   15.32       7.13        629.8       112.6       11
        2   15.32       7.13        629.8       112.6       130
        3   14.17       7.06        1068.8      155.7       0
        3   14.17       7.06        1068.8      155.7       5
        3   14.17       7.06        1068.8      155.7       0
        3   14.17       7.06        1068.8      155.7       0
        3   14.17       7.06        1068.8      155.7       35
        3   14.17       7.06        1068.8      155.7       0
        3   14.17       7.06        1068.8      155.7       0
        4   15.41       7.02        453.7       89.9        0
        4   15.41       7.02        453.7       89.9        130
        4   15.41       7.02        453.7       89.9        98
        4   15.41       7.02        453.7       89.9        20
        4   15.41       7.02        453.7       89.9        565
        4   15.41       7.02        453.7       89.9        0




Call:
glm(formula = Cases2011 ~ Winmaxtemp11 + Springmaxtemp11 + Summaxtemp11 + 
    Autummaxtem11 + Anumaxtemp11 + Winmintemp11 + Springmintemp11 + 
    Summintemp11 + Autmintem11 + Anumintemp11 + Winsundays11, 
    family = "poisson", data = orf)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-18.689  -10.242   -8.238   -1.635  124.064  

Coefficients: (7 not defined because of singularities)
                 Estimate Std. Error z value Pr(>|z|)    
(Intercept)     -2.424380   0.453662  -5.344 9.09e-08 ***
Winmaxtemp11    -0.009405   0.080438  -0.117    0.907    
Springmaxtemp11 -4.570125   0.120415 -37.953  < 2e-16 ***
Summaxtemp11     0.636944   0.121192   5.256 1.47e-07 ***
Autummaxtem11    3.900808   0.067515  57.777  < 2e-16 ***
Anumaxtemp11           NA         NA      NA       NA    
Winmintemp11           NA         NA      NA       NA    
Springmintemp11        NA         NA      NA       NA    
Summintemp11           NA         NA      NA       NA    
Autmintem11            NA         NA      NA       NA    
Anumintemp11           NA         NA      NA       NA    
Winsundays11           NA         NA      NA       NA  
$\endgroup$
  • $\begingroup$ So, for region 1, is the data saying that there are 5 countries that comprise it? What is the temporal unit of analysis? Annual? And are all of these countries/regions on a single continent? $\endgroup$ – Mike Hunter Nov 20 '15 at 19:33
  • $\begingroup$ @DJohnson, data from one country (England) further broken down into 4 regions while climate variables are based on annual values. Climate values are regional based on farm post codes $\endgroup$ – Joshua Onyango Nov 20 '15 at 20:49
  • $\begingroup$ If there are 4 regions, why does region 1 have 5 entries, region 2 have 6, region 3 have 7 entries, etc? My suggestion would be to find more disaggregate weather and sheep disease case data such as quarterly or even monthly. Having worked with it, I know that the weather data is readily available from several sources (e.g., ecmwf.int). The problem will be getting the disease count data. $\endgroup$ – Mike Hunter Nov 20 '15 at 21:40
  • $\begingroup$ Thanks @DJohnson . Sorry I forgot to mention that in the data sample I coded regional entities for easy analysis in R with region 1 (North region) having 5 farms reporting disease cases. I got the data from UK met website which only regional climate variables - no luck with post code specific values. I already have disease count data reported by farmers across farms. $\endgroup$ – Joshua Onyango Nov 20 '15 at 22:02
  • $\begingroup$ There are literally thousands of fixed weather stations in the US alone and at least as many in Europe and the UK. I'm sure that if you dig around a little more, you will find regional weather for the month or quarter. Is farmer disease count data available on a more frequent basis than annual? Those will be the tough numbers to get. $\endgroup$ – Mike Hunter Nov 20 '15 at 22:14
1
$\begingroup$

It is hard to give You a simple answer without having seen the whole dataset and beeing more knowledgable about sheeps but I would do the following things

  • filter out all the row where there is no disease
  • calculate ratio of unhealty/total because the higher number of disease count can be related to the total number of sheeps (if 1 out of 10 sheeps is ill it's the same as 10 out of 100 if you take ratios but in total number you have that the second region is like 10 times "worse")
  • if the variance among the variable is very low try to convert it to categorical variable and then use for example ANOVA
$\endgroup$
  • $\begingroup$ thanks for this. Number of sheep did come out as a risk factor. My main obstacle is how to go around looking at the other variables such as temperature, rainfall... as risk factors for the disease in the regions given that the values with and between region have very small variation. $\endgroup$ – Joshua Onyango Nov 20 '15 at 20:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.