Hi,
Im trying to analyze the relationship between morphological and ecological variables through the phylogeny of the group using BM and OU models. But I don't know how to explain that GLS coefficient values can be as low as 0.009 and still significan (p = 0.004).
I need help!
Thanks, have a great day!
Example 1.
Generalized least squares fit by REML
Model: TPC1 ~ Environ
Data: DF.PGLS_all_data
AIC BIC logLik
-726.3365 -716.4869 366.1682
Correlation Structure: corBrownian
Formula: ~1
Parameter estimate(s):
numeric(0)
Coefficients:
Value Std.Error t-value p-value
(Intercept) 0.00472743 0.028766193 0.164340 0.8696
Environ -0.00919997 0.002538026 -3.624852 0.0004
Correlation:
(Intr)
Environ -0.13
Standardized residuals:
Min Q1 Med Q3 Max
-2.1607098 -0.3774017 0.1315785 0.4589424 1.6401371
Residual standard error: 0.07873684
Degrees of freedom: 199 total; 197 residual
Example 2.
Generalized least squares fit by REML
Model: ABPC2 ~ Environ
Data: DF.PGLS_all_data
AIC BIC logLik
-936.072 -926.2223 471.036
Correlation Structure: corMartins
Formula: ~1
Parameter estimate(s):
alpha
1
Coefficients:
Value Std.Error t-value p-value
(Intercept) 0.008687800 0.003526749 2.463402 0.0146
Environ -0.005631088 0.002449643 -2.298738 0.0226
Correlation:
(Intr)
Environ -0.894
Standardized residuals:
Min Q1 Med Q3 Max
-2.0009436 -0.6660191 -0.0796327 0.6954614 2.6972526
Residual standard error: 0.02178988
Degrees of freedom: 199 total; 197 residual
The magnitude of the coefficient in a GLS regression model depends on the measurement scale of your variables. As far as I can see, your (statistically significant) coefficient of -0.009 is the average decrease in your dependent variable TPC1 for each unit increase in the independent variable Environ, which I guess is a continuous variable. If so, you may want to change the scale of your variable Environ (e.g. by dividing it by a constant) so that the unit increase will be larger. For instance, imagine you want to predict body weight (dependent variable) as a function of the time spent exercising (independent variable) expressed in minutes per week, and you see that, in fact, each minute increase in weekly exercise leads to only a minimal change in body weight, you can always express your predictor as, e.g., hours or fractions of days of exercise per week to catch the "larger", yet equally (in)significant, effect of exercising on body weight. Just keep in mind the scale of your predictor!
From your question it seemed to me that you are puzzled that a statistically significant effect was so small -- so seemingly "insignificant." Prof. Mughini-Gras's response supplies a possible reason for this: the scales of the variables (e.g., a minute vs. an hour). There are at least two additional possibilities. First, the effect you found may be real but frankly not very strong. In social science, at least, it is common to use even very weak correlations or regression effects as conclusive evidence for the importance of some theorized process. This is not necessarily wise. though there are reasons to consider doing it. We have to make use of the data at hand, even if it is not as solid as we would like. The third possibility to keep in mind -- not to be too negative, and I add it only to be thorough -- is that your result could be a false positive.
- Badges
- Science topic
- Similar topics
- Medicine
- Counseling Psychology