12

Some of my explanatory variables have a few null values for certain polygons.

Can Geographically Weighted Regression Analysis handle these, or should I remove the whole polygon and all data (dependent and independent variables for that particular polygon)?

|improve this question
  • What do these nulls represent? That a value is not applicable or that it does exist but is unknown? If the latter, why is it unknown? (The chief concern is whether the reason for a value being unknown is in any way related to the value itself; for instance, if you are studying hydrology and using land cover as an explanatory variable, then if land cover is unknown due to flooding, that's important information and means much more than a mere missing value.) – whuber♦ Sep 29 '14 at 20:31
  • 1
    Thank you whuber, Some of the nulls represent data that was omitted for confidentiality due to small units of aggregation, others were simply not available but not as the result of the explanatory variables affecting other varibales. I have true zero values whereby, the zero represents no event/rate for that particular polygon and I also have some null values where the data is missing. Since I have about 29 explanatory variables to start off with, if I take out the polygons where for the rows containing nulls, I am losing 8 polygons out of 279 and I was hoping I didn't have to. Thank you! – I Heart Beats Sep 29 '14 at 20:39
  • I am glad you have information and theories about the missingness. You might want to be a little cautious about those small units, because the values of just about any variable could plausibly be related to the size of the region it represents, thereby creating a potential bias in the pattern of missingness. – whuber♦ Sep 29 '14 at 20:45
  • 2
    You can incorporate nulls by introducing variables to indicate their presence and encoding all original null results with a common value (which value you choose doesn't really matter). You will need one separate indicator for each variable that contains any nulls. Whatever you do, don't just replace nulls by zeros (or any other constant)! A null means something entirely different than zero. – whuber♦ Sep 29 '14 at 21:04
  • 1
    @whuber It looks like there may be enough in your comments to write up an answer on this one. – PolyGeo♦ Mar 27 '17 at 5:17
1

You can try to guess the 'null' values from the surrounding data. That would be better than leaving them out, because that way you wont lose data. Of course you have to be carefull in how you guess the values. Because if your guess is biased, your model will also be.

see also: http://handbook.cochrane.org/chapter_16/16_1_2_general_principles_for_dealing_with_missing_data.htm

and:

Caution:

Whenever using shapefiles, keep in mind that they cannot store null values. Tools or other procedures that create shapefiles from nonshapefile inputs may, consequently, store null values as zero or as some very small negative number (-DBL_MAX = -1.7976931348623158e+308). This can lead to unexpected results. See also: Geoprocessing considerations for shapefile output. (http://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-statistics-toolbox/geographically-weighted-regression.htm)

|improve this answer

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.