Hot answers tagged

8

The bottom line is that (as Jeremy Miles says) the value of the negative log-likelihood doesn't really matter, only differences between the negative log-likelihoods. But you might still wonder why you are getting negative values. Reproducing an answer of mine from here: Technically, a probability cannot be >1, so a log-likelihood cannot be >0, so a ...


4

Debugging strange results from various kinds of software is off-topic here. However, it may help you to figure out what is wrong, if I show output for Fisher's Exact Test as implemented in R. In the process you can try to learn the basics of Fisher's Exact Test. [You may also want to google the Wikipedia page on Fisher's Exact Test. And @whuber's Answer on ...


4

It is not possible to give a definitive answer without accesss to the data itself, and even with access to the data it still might not be possible. However, we can say a few things about this situation. Given that you expect high autocorrelation, I would suggest sticking with the first model. The 2nd model, which has many more parameters, may very well be ...


3

Yes. -2 LL means -2 multiplied by the log likelihood. AIC, BIC etc are (as far as I know) only interpreted in relation to other values from different models. An AIC of -100 doesn't mean anything on its own. It means something when a different model, using the same data, has an AIC of -90, so the difference is 10. The difference is the interesting thing.


3

The research question is: I want to check the relation of the fixed effects to the dependent variable, taking into account the fact that the design is clustered. This is answered by the fixed effects estimates. How do I make interpretation of the random effect Generally there is no requirement to interpret the random effects - you are controlling for ...


2

The difference between those two variance-covariances matrixes is that SPSS assigned PER random effect one Scaled Identity Matrix when you indicate Variance Component. When you indicate Scaled Identity, the covariance structure aplies for each specification of the random effects. Below there is an example; two random effects (Intercept and Time with 2 time ...


2

The proposed model is not an ANCOVA. ANCOVA is a model with a continuous outcome, a categorical independent variable of primary interest (main exposure), and one of more continous variables that are potential confounders or competing exposures. The distinction isn't really important because it's just another (multivariable) regression model. So the model ...


2

Thanks for making a reproducible example. That meant it was possible to give you an answer. I always find manova to be a painful technique, and it's never clear to me what it's actually testing. You've made a mistake in your manova code (I believe) because you've run a model with no intercept. It should be: m <- manova(cbind(PEVOCAB, RAVIN)~ NS + NA. + ...


2

I am not very familiar with SPSS but as far as I see MANOVA in SPSS reports multivariate tests for each predictor separately (just like manova in R). So, it does not report multivariate tests of significance for EFFECT..WITHIN CELLS REGRESSION (maybe there are additional options). However, you can obtain these results using canonical correlation in SPSS and ...


2

You are looking at a cross-sectional regression with non-time related subgroups, say, linear $$y_{is} = \mathbf x_{si}'\beta + \mathbf z_{si} + \mathbf g_s+ v_{si}$$ where $s$ indicates the subgroup (different industry for example) and your individual $z$ variables are not observable as data. Group-wise heterogeneity (different industries, geographical) ...


2

As mediansplit has only two levels you get the outcome of the significance test from the Fixed Effect table and if you request the estimated marginal means tables /EMMEANS=TABLES(mediansplit) you get the same in terms of a t-test. The error occurs as soon as you specify an interaction in the /fixed statement. It should work if you omit the interaction. I ...


2

Both models are identical and will lead to identical predicted probabilities. The apparent discrepancies between SPSS and R are due to different reference levels of the categorical predictors. As I explain here in detail, when using dummy contrasts, the coefficients for categorical predicators are differences between that level of the predictor and the ...


2

I don't mean to dog pile, but you really should not be doing stepwise regression. I won't include reasons why, kjetil has already listed some and you can search for key terms on this site if needed (or just google "Stepwise Regression" and "Frank Harrell" who has already commented here for the definitive list of reasons why not to use ...


2

There are two main considerations when choosing whether to specify random slopes for a variable: Is it biologically / clinically / theoretically possible for each subject (or whatever the grouping variable is) to have their own slope with respect to that variable ? Obviously this also implies that the variable varies within level of the grouping variable. ...


2

The Python coefficients are the logarithms of the differences in the SPSS coefficients, so 0.541+exp(0.743291) is 2.643 0.541+exp(0.743291)+exp(-0.2248) is 3.442 and so on. This parametrisation ensures the thresholds are increasing.


1

Below is an overview of the data that you have collected (or will be collecting): Class Control: Inductive teaching approach (n=14) Student Score Pre Score Post Score Delayed Student 1 x y z ... ... ... ... Student 14 a b c Class Experimentation: Deductive teaching approach (n=16) Student Score Pre Score Post Score Delayed Student 1 x y z ... ... ... ...


1

I assume the software you are using transforms them into Fisher's $z$ (the inverse hyperbolic arctangent). What you suggest is a plausible way forward but note that the value of $z$ is quite sensitive to small changes near unity. For r=0.99 it gives 2.65, for r=0.999 it gives 3.80, and for r=0.9999 it gives 4.95. It might be a good idea to use several values ...


1

Normally data from such Likert scales are interpreted as holding ordinal but not interval scaled information, and the linearity assumption in standard linear regression will be problematic for sure with a response variable that only has 3 output values (it may be OK with Likert explanatory variables but then it may not, depending on the data). There's ...


1

SPSS doesn't have a nonparametric analogue of a 2x2 ANOVA per se, but the GENLIN procedure for generalized linear models offers modeling for variables with other than normal distributions.


1

Paired t tests and Wilcoxon SR tests are inappropriate. If you were to do paired t test on the differences between preference for US and foreign travel, one would wonder about the validity of the results because it is unlikely that fifty small-integer values would be approximately normally distributed. If you were to try to do a Wilcoxon signed-rank test, ...


1

That's not what weighting does. Weighting doesn't apply to variables in the data; it applies to sums in the computation. For example, if you're estimating a mean, the unweighted formula is $$\bar X = \frac{\sum_{i=1}^n X_i}{n}= \frac{\sum_{i=1}^n X_i}{\sum_{i=1}^n1}$$ and the unweighted formula is $$\bar X = \frac{\sum_{i=1}^n X_i}{n}= \frac{\sum_{i=1}^n ...


1

If you compute $\kappa$ for each of the three subsets you can then summarise it either by taking the median or by a weighted mean with weights the inverse of the sampling variances. That gives you the average agreement of a secondary rater with the chosen primary rater. If the primary rater is seen as a gold standard, perhaps because they are highly trained ...


1

The trick is that a paired t-test is just a one-sample test of the pre-post differences, in other words, the null hypothesis is that that difference is 0. If you want to compare two paired differences, you would calculate the differences manually and run the two-sample t-test. Another approach is the ANCOVA. Using a linear regression model, the "pre&...


1

To address your question, there are pros and cons of both approaches. Technically the assumption is normality in each sample, which suggests two tests, however this is multiple testing and therefore has either an increased type I error probability, or bad power when corrected with Bonferroni. A test with both samples taken together (after centering them both ...


1

All you wrote is correct. The total effect is of HAPMEAN on CUSTOM is .1266. The indirect effect of SCMEAN is .1522; and .0997 for SOCFAC. The direct effect is there! The c’ prime is -.1324. It is the effect of CUSTOM on HAPMEAN with all other variables in the models.


1

There is a lot of questions, let break them out. Regarding the title itself, your models imply that the only good predictor is SU_7R_pu. Maybe it is related too much with other VIs so it is the one taking most of the variance in the DV. However, the models, as they stand, imply that your IV and moderator has no effect (statistically) when you control for ...


1

A log-rank test is a test of whether the hazard as a function of time differs between 2 groups. If the log-rank test doesn't show a significant difference between 2 curves in this respect, then you should be very reluctant to evaluate differences at specific survival times. That's a general principle of statistical tests: if the overall test isn't ...


1

I strongly suspect that this particular part of the table, explicitly based on a multi-predictor logistic regression model stratified by sex, has an error in the p-value entry. The odds ratios are presented with respect to a reference category for each of the predictors, so that the p-values should represent the p-value corresponding to a null hypothesis of ...


1

Use a two-sample t test. Use the Welch version of the test unless you have good reason to believe M and F populations have the same variance. Example in R: set.seed(2020) m = rnorm(67, 68, 4) f = rnorm(16, 65, 3) summary(m); length(m); sd(m) Min. 1st Qu. Median Mean 3rd Qu. Max. 55.84 66.52 68.80 68.67 71.49 80.81 [1] 67 # ...


1

As noted in the comments, POSTHOC results are for main effects for between-subjects factors. That averages over the levels of the within-subjects factor, so to see what in that chart is being compared, look at where the lines for each level are in the middle of the chart (halfway between the two plotted values for each group). A and B are indeed much further ...


Only top voted, non community-wiki answers of a minimum length are eligible