Other Qs Flashcards
(24 cards)
What distribution might you consider using to model non-negative integers such as count data?
Poisson or negative binomial
If the response variable was assumed to be Poisson distributed, what would you conclude from the output of a generalized linear model that showed the residual deviance was five times the residual degrees of freedom?
The response variable was like likely overdispersed
In no more than 50 words - describe what is meant by a statistical interaction between two variables?
A statistical interaction between two variables occurs when the effects of one explanatory variable on the response variable depend on another explanatory variable.
When is (+1) included in the df for an algebraic formula?
When it is normal or negative binomial distributed - this is an extra df to account for the residual variance
What distribution would you consider to model binary data such as presence/absence data?
Bernoulli
If the response variable was assumed to be normally distributed, what would you conclude from the output of a GLM that showed the residual deviance was six times the residual degrees of freedom?
The model likely explained a modest amount of variation in the data
When conducting a likelihood ratio test to compare two nested models, how is the test statistic calculated? How is it assumed to be distributed?
The test statistic is twice the absolute difference between the log-likelihoods of the simpler and complex model (assuming the simpler model is nested within the more complex one). The test statistic is chi-squared distributed.
What does a graph of probability look like?
S-shaped curve
When is (+1) NOT included when counting up the dfs in a model?
When it is poisson or bernoulli distributed
When would a response variable be modelled as a poisson distribution?
When it is a non-negative integer
Why would you choose to model a variable as a random effect rather than a fixed one?
When you aren’t really interested in its effects and it would be very hard for other studies to replicate
What is the probability of drawing a random number between 0.2 and 0.6 from a uniform distribution defined between 0 and 0.8?
0.6-0.2 = 0.4
0.4/0.8 = 0.5
The probability is 0.5
M1 has a log-likelihood of -67893.62 and M2 has a log-likelihood of -67889.13. Which is the more likely model?
M2 as it has a larger log-likelihood (closer to 0)
A researcher fits a model to data assuming the response variable has a binomial distribution. She observes a residual deviance of 52.78 associated with residual degrees of freedom equal to 45. Should she be concerned by this?
No, the residual deviance is broadly in line with the residual degrees of freedom and overdispersion is not evident
A researcher fits a model to data assuming the response variable has a normal distribution. She observes a residual deviance of 158.12 associated with residual degrees of freedom equal to 31. Should she be concerned by this?
No, overdispersal is not a concern for a normal distribution.
A parameter has an estimated value of 0.26 and standard error of 0.12. Assuming the residual degrees of freedom is large, estimate approximate confidence intervals for this parameter. Is the parameter estimate significantly different to zero at a level of significance of 0.05?
0.26-(2x0.12) to 0.26+(2x0.12)
Confidence intervals are +/- 2 x standard error
The parameter is significantly different to zero since the interval does not include zero.
What distribution is gaussian?
Normal distribution
How do you calculate the sample size from an output?
df + 1
What is the probability of drawing a random number between 0.7 and 0.9 from a uniform distribution defined between 0 and 1?
0.9-0.7 = 0.2
0.2/1 = 0.2
The probability is 0.2
What is the likelihood of drawing a random number equal to 19.6364 from a uniform distribution defined between 0 and 100?
Since the area under the pdf must be 1, likelihood is 0.01
A researcher fits a model to data assuming the response variable has a normal distribution. She observes a residual deviance of 258.12 associated with residual degrees of freedom equal to 53. Should she be concerned by this?
No, overdispersal is not an issue with normally distributed data
In what kind of model is overdispersal not a problem?
Normally distributed model
A Poisson distribution has a mean of 9. What is its standard deviation?
3 (square root of 9)
What is the meaning of the dispersion parameter?
The dispersion parameter reflects how spread out your data are relative to the model’s assumptions.
