Re: [ECOLOG-L] G-test with zero values
Jason, I’m going to have to heartily agree here. My first thought, when you ask how to “analyze” your data, is that this is a fairly broad term, and the answer will depend entirely on what your question is. I would encourage you to think in terms of parameter estimations. At the end, you will have a number (or several) that describes your results. A G statistic (or t, or Chi-sq, or a p-value, for that matter) is not very informative. A slope of a relationship between two variables (or a difference between two means, or a probability of some event occurring), on the other hand, is quite informative. So think in terms of a somewhat skeptical audience, who, if told, “there is a difference,” will immediately respond, “but how large of a difference?” Being able to answer that question should guide you in producing the most appropriate model. Cheers Matt Talluto On Feb 12, 2014, at 21:41, David Schneider david.schnei...@mun.ca wrote: Hello Jason, The 21st century approach to percent and count data is to write the model, not search for the 'right test.' In my experience it is possible for 4th year undergrads and 1st year grad students, with little stats experience, to learn this approach. Statistical analysis based on writing the statistical model can be carried out in almost all stat packages, including SPSS and Minitab. Not to mention SAS and R. Statistically adept readers of Ecolog will recognize problems with zeros when analyzing percent data or count data once one has learned to write the model. These include too many expected values less than zero, or other problems such as zero inflated counts. I trust they will hold off on such problems -- in my view the first and most important step for you is grasping the idea of writing the model that captures your conceptualization of the research question and operating hypotheses, instead of searching for the 'right test.' In the fall term of 2013 a highly motivated grad student with at best a tenuous grasp of algebra learned this approach. If she can learn to write the model, and execute it, and interpret the result, and check the assumptions, then you can. Wishing you the best, David S. http://www.mun.ca/biology/dschneider/ On Wed, Feb 12, 2014 at 12:56 AM, Jason Hernandez jason.hernande...@yahoo.com wrote: Some time ago, I inquired about ways to analyze percent cover data, and one of the suggestions was to test for heterogeneity. The snag, however, is that this requires multiplying each cell value by its natural log. My data set has a lot of zero values, which are important to keep; but of course there is no natural log of zero. Is there a way to adjust the analysis to included these zero values? i have not managed to find anything on this. Jason Hernandez
[ECOLOG-L] G-test with zero values
Some time ago, I inquired about ways to analyze percent cover data, and one of the suggestions was to test for heterogeneity. The snag, however, is that this requires multiplying each cell value by its natural log. My data set has a lot of zero values, which are important to keep; but of course there is no natural log of zero. Is there a way to adjust the analysis to included these zero values? i have not managed to find anything on this. Jason Hernandez
Re: [ECOLOG-L] G-test with zero values
JasonI don't know how it works with a g-test, but the common statistical remedy for this is to add 1 or 0.1 to all numbers which then allows the calculation of a log transform but preserves the relative relationships among data points. cheers, cheers, g2 On Wed, Feb 12, 2014 at 9:56 AM, Gary Grossman gdgross...@gmail.com wrote: I don't know how it works with a g-test, but the common statistical remedy for this is to add 1 or 0.1 to all numbers which then allows the calculation of a log transform but preserves the relative relationships among data points. cheers, cheers, g2 On Wed, Feb 12, 2014 at 12:56 AM, Jason Hernandez jason.hernande...@yahoo.com wrote: Some time ago, I inquired about ways to analyze percent cover data, and one of the suggestions was to test for heterogeneity. The snag, however, is that this requires multiplying each cell value by its natural log. My data set has a lot of zero values, which are important to keep; but of course there is no natural log of zero. Is there a way to adjust the analysis to included these zero values? i have not managed to find anything on this. Jason Hernandez -- Gary D. Grossman, PhD Professor of Animal Ecology Warnell School of Forestry Natural Resources University of Georgia Athens, GA, USA 30602 Research teaching web site - http://grossman.myweb.uga.edu/http://www.arches.uga.edu/%7Egrossman Board of Editors - Animal Biodiversity and Conservation Editorial Board - Freshwater Biology Editorial Board - Ecology Freshwater Fish Sculpture by Gary D. Grossman https://www.facebook.com/pages/Gary-Grossmans-Sculpture-Portfolio/124819124227147?fref=ts Hutson Gallery Provincetown, MA - www.hutsongallery.net/artists.html My ukulele channel - www.youtube.com/user/garydg29 -- Gary D. Grossman, PhD Professor of Animal Ecology Warnell School of Forestry Natural Resources University of Georgia Athens, GA, USA 30602 http://grossman.myweb.uga.edu/ http://www.arches.uga.edu/%7Egrossman Board of Editors - Animal Biodiversity and Conservation Editorial Board - Freshwater Biology Editorial Board - Ecology Freshwater Fish
Re: [ECOLOG-L] G-test with zero values
Hello Jason, The 21st century approach to percent and count data is to write the model, not search for the 'right test.' In my experience it is possible for 4th year undergrads and 1st year grad students, with little stats experience, to learn this approach. Statistical analysis based on writing the statistical model can be carried out in almost all stat packages, including SPSS and Minitab. Not to mention SAS and R. Statistically adept readers of Ecolog will recognize problems with zeros when analyzing percent data or count data once one has learned to write the model. These include too many expected values less than zero, or other problems such as zero inflated counts. I trust they will hold off on such problems -- in my view the first and most important step for you is grasping the idea of writing the model that captures your conceptualization of the research question and operating hypotheses, instead of searching for the 'right test.' In the fall term of 2013 a highly motivated grad student with at best a tenuous grasp of algebra learned this approach. If she can learn to write the model, and execute it, and interpret the result, and check the assumptions, then you can. Wishing you the best, David S. http://www.mun.ca/biology/dschneider/ On Wed, Feb 12, 2014 at 12:56 AM, Jason Hernandez jason.hernande...@yahoo.com wrote: Some time ago, I inquired about ways to analyze percent cover data, and one of the suggestions was to test for heterogeneity. The snag, however, is that this requires multiplying each cell value by its natural log. My data set has a lot of zero values, which are important to keep; but of course there is no natural log of zero. Is there a way to adjust the analysis to included these zero values? i have not managed to find anything on this. Jason Hernandez
Re: [ECOLOG-L] G-test with zero values
Technically the definition of the nat log of zero is actually an asymptote. What does zero mean in your data? That is the key issue you must approach before modifying the information. If zero means too small to measure but not likely zero, then you can do some decimal estimation or I think you can use a tobit model, but I'm not super familiar with those. X+1 or X+0.1 etc. is typically used with count data. However, if the data do not mean that, or are actually some form of ratios, then when y z, the ln (y/z) will be negative, so when you add one, you will have a bunch of zeros that cause you to lose data! I vaguely recall something about right and left censored models or something. However, you better make sure, I'm pulling this out of somewhere smelly because this is something I read once and I don't remember where. But, maybe it will give you a place to start? Malcolm On Tue, Feb 11, 2014 at 11:56 PM, Jason Hernandez jason.hernande...@yahoo.com wrote: Some time ago, I inquired about ways to analyze percent cover data, and one of the suggestions was to test for heterogeneity. The snag, however, is that this requires multiplying each cell value by its natural log. My data set has a lot of zero values, which are important to keep; but of course there is no natural log of zero. Is there a way to adjust the analysis to included these zero values? i have not managed to find anything on this. Jason Hernandez -- Malcolm L. McCallum Department of Environmental Studies University of Illinois at Springfield Managing Editor, Herpetological Conservation and Biology Nothing is more priceless and worthy of preservation than the rich array of animal life with which our country has been blessed. It is a many-faceted treasure, of value to scholars, scientists, and nature lovers alike, and it forms a vital part of the heritage we all share as Americans. -President Richard Nixon upon signing the Endangered Species Act of 1973 into law. Peer pressure is designed to contain anyone with a sense of drive - Allan Nation 1880's: There's lots of good fish in the sea W.S. Gilbert 1990's: Many fish stocks depleted due to overfishing, habitat loss, and pollution. 2000: Marine reserves, ecosystem restoration, and pollution reduction MAY help restore populations. 2022: Soylent Green is People! The Seven Blunders of the World (Mohandas Gandhi) Wealth w/o work Pleasure w/o conscience Knowledge w/o character Commerce w/o morality Science w/o humanity Worship w/o sacrifice Politics w/o principle Confidentiality Notice: This e-mail message, including any attachments, is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply e-mail and destroy all copies of the original message.
