See all my videos at www.tilestats.com/ In this video, we will see how we can use an explanatory variable on a categorical scale (nominal scale with two or three categories) and how we can interpret the output of such a model.
Пікірлер: 8
@kiarashriazi51244 ай бұрын
You are the best teacher. Thank you.
@musclesmalone2 жыл бұрын
Your channel is a goldmine. Thank you
@tilestats2 жыл бұрын
Thank you!
@kimiadarbeh7442 жыл бұрын
Thanks
@ammmaj2 жыл бұрын
Great tutorial. In the example above with metastatic lymph nodes, if you use logistic regression instead of poisson to compare the two treatments, would that be wrong and why? What I mean is to model treatment as the outcome (dependant variable) and the number metastatic lymph nodes as an independent variable. I understand the two approaches model different questions, the logistic regression will model the probability of belonging to treatment A vs B based on the number of metastatic lymph nodes, while the poisson will model the probability of having a certain number of metastatic lymph nodes based on received treatment (A or B). But when you try to answer the question of whether the number of metastatic lymph nodes differ between treatment groups (ie if a treatment reduced the risk of metastatic), is one approach better than the other and would they yield different results?
@tilestats2 жыл бұрын
No, it depends on your question and what you want to extract from the model, as well as assumptions. In this case, we compare the mean counts between two groups, by assuming that the counts follow a Poisson distribution. With logistic regression, you would instead compute an odds ratio (OR) and check if one unit increase in the independent variable results in a OR that is significantly different from one. If so, you can conclude that there is an association between the dependent and independent variables. Thus, if you like to compare the mean counts between two groups, use Poisson regression (or ZIP, NB as explained in the next videos).
@aabhaanekar99905 ай бұрын
Great video! Thank you so much for the beautiful explanation. Based on your previous videos, you mentioned that as the mean gets closer to zero, higher the chances of negative values, so better to consider Poisson distribution, which eliminated the negative values. So is it better to choose Poisson regression over t-test if the means are closer to zero? Also, when or how can we assume if the data has a normal distribution or Poisson distribution? Can we plot them in softwares like R studio and see? As a rule of thumb, can we assume Poisson distribution for a count data?@@tilestats
@tilestats5 ай бұрын
Use Poisson only for count data. A t-test works just fine for negative values.