Q&A

ResearchGate Q&A lets scientists and researchers exchange questions and answers relating to their research expertise, including areas such as techniques and methodologies.

Browse by research topic to find out what others in your field are discussing.

- Burak Alakent added an answer in Normal Distribution:Deviance for binary regression models for expected number of successes smaller than five?
- Roman V. Gulyaev added an answer in XPS:What is the necessity of using Gaussian Lorentzian mixed function while describing the XPS data of any material ?
- Iván Sanz added an answer in Swine Influenza:Can you answer my questions on genetic and antigenic characterisation of the influenza virus?
- Francesc Codony added an answer in Qualitative PCR:Does anyone have experience in validation of qualitative PCR assays?
- Erkki J. Brändas added an answer in General Relativity:Universe is static!!! Yes or no?
- Keith Henry Nicholls added an answer in Devonian:Can anybody help me in identyfing a Devonian fossil?
- Ersin Karataş added an answer in Oxalates:How can I produce recombinant ENZYME ?
- Cristiano Teodoro Russo added an answer in Paraffin Embedding:Has anyone successfully detected HPV16 E7 by using the cervimax kit on formalin-fixed paraffin embedded human tissue samples?
- Temim Deli added an answer in Ear:How can I run a gel on gDNA extracted from rat ear punch?
- Louis Brassard added an answer in Cultural Analysis:Is the `culture´ concept a way of setting distance between Humans and other animals or it can be a bridge for inter-species comunication?
- Pradip Kumar Singh asked a question in Fellowship:Can anyone educate me about "Carl R. Woese Postdoctoral Fellowship Program" ?
- Holly Tran asked a question in HEPES:Where can I order 1x DNA oligonucleotide annealing buffer?
- JC Spender added an answer in Money and Banking:How is the increasing presence of what Marx defined as interest bearing capital distinct from other social forms credit?
- Chris Hsiung added an answer in Lentivirus:Could we use the DH5alpha instead of Stbl3 for lentiviral plasmid transformation and cloning?
- John George Hardy added an answer in RGD:Why can't MSCs attach to an RGD modified PEG system?
- Vincent G Guida added an answer in Bivalvia:Does intracellular digestion of larger food particles (like it can be found in bivalvia) occur in teleost?
- Alexandra Müller added an answer in Spectrophotometer:Can anyone help me with troubleshooting of a Cary 100 UV-Visible spectrophotometer (Varian)?
- Behzad Vaziri Hassas added an answer in Piezoelectricity:How can i adjust ultrasound intensity?
- Atanu Pati added an answer in Prisons:Can anyone share his/her studies on the sleep-wake profile of the prisoners?
- Paul Germann added an answer in Stress:Do you believe all active substances increase in stress conditions?
- Bernard Desroches added an answer in Nietzsche:Is Western nihilism the result of reduction of Western symbols of human dignity and virtue to consumption slogans?
- Nisha Rizvi asked a question in Perforant Pathway:How to best detect pre- or postsynaptic mechanism using paired-pulse protocol in rat hippocampal CA1?
- Sudev Naduvath added an answer in Topological Graph Theory:What is the number of possible non-isomorphic trees for any node?
- Amir Emami asked a question in Magnetics:Are non-integer magnetic moment materials, half metal?
- Marimuthu Sivakumar asked a question in Coating:What will be the effect when coating NiO on LiFePO4, Will it affects the lattice parameters? If so, why?
- Abbas Rahdar added an answer in Magnetization:What are the conditions of magnitude of Ms and out-of-plane crystal anisotropy for a thin film to have no in-plane magnetization?
- Francesc Codony added an answer in Proteolysis:How can I check wild wine-yeasts proteolysis?
- Oliver Serfling added an answer in R:What resources or books (free or otherwise) do you use for R?
- Hamed Haghi added an answer in Flotation:How can we depress kaolinite clay to improve flotation gold recovery?

I have a single response, which denote the number of faulty products/total products (success, or failure ratio) over quite a number of factors (~15). Since these 15 factors are not independent from each other, the number of successes are close to zero in almost half of the cells (combinations of factors). When I perform logit regression, not surprisingly, I find that the expected values of successes in those cells are smaller than five. It is generally known that deviance or Pearson X^2 is close to normal distributed only if E{X} >= 5 (though this is a conservative estimate). However, expected values of quite a number of my cells are smaller than unity! In this case, can anybody recommend me what to do? As far as I can see, there seems to be two methods to deal with the problem:

1. Construct the model with all the data, find the combinations of factors for which E{X} < 5 (or maybe 3), omit these data and remodel. However, I would not be using some of the data points in (re)modeling.

2. Construct the model with all the data, but in deviance calculation use only the cells for which E{X} > 3-5.. And for degrees of freedom, use DOF = remaining cells - number of parameters. Here, the problem is I'm not sure whether this is statistically valid, since I would be systematically discarding some portion of the data in deviance calculation, and this may perhaps cause some bias.

Any help would be appreciated.