lsmeans | R package for least-squares means | Machine Learning library

I would like to test the simetry in the response of an observer to a contrast stimuli with different polarity, positive (white) and negative (black). I took the reaction time (RT) as dependent variable, along four different contrasts. It is known that the response time follows a Pieron curve whose asymptotas are placed (1) at observer threshold (Inf) and (2) at a base RT placed somewere between 250 and 450 msec. The knowledge allows us to linearize the relationship transforming the independent variable (effective contrast EC) as 1/EC^2 (tEC), so the equation linking RT to EC becomes:

RT = m * tEC + RT0

To test the symmetry I established the criteria: same slope and same intercept in the two polarities implies symmetry. To obtain the coefficients I made a linear model with interaction (coding trough a dummy variable for Polarity: Positive or Negative). The output of lm is clear to me, but some colegues prefer somthing more similar to an ANOVA output. So I decided to use emmeans to make the contrasts. With the slope is all right, but when computing the interceps starts the problem. The intercepts computed by lm are very different from the output of emmeans, and the conclusions are also different. In what follows I reproduce the example. The question is two fold: It is possible to use emmeans to solve my problem? If not, it is possible to make the contrasts through other packages (which one)?

Data RT1000 EC tEC Polarity 596.3564 -25 0.001600 Negative 648.2471 -20 0.002500 Negative 770.7602 -17 0.003460 Negative 831.2971 -15 0.004444 Negative 1311.3331 15 0.004444 Positive 1173.8942 17 0.003460 Positive 1113.7240 20 0.002500 Positive 869.3635 25 0.001600 Positive Code ...

Hoping that you can clear some confusion in my head.

Linear mixed model is constructed with lmerTest:

...

I am reviewing one way ANOVAs and trying to integrate least squared means. Here is an example from mtcars.

...

Hi I want to recreate a plot that looks like this?

Here is an example, the closest I can get is a ggplot that looks like this.

...

I am using Proc glimmix to run a random intercept model on a continuous outcome over five time points. Table Type III indicates sex is not significant while the 95% CIs of male and females in table lsmeans have no overlap. I wonder how to explain this discrepancy? When I look at the male and female outcomes over time I also expect a significant higher outcome for males as the CIs in the second table say. Many thanks!

...

I have to analyze an experiment data set to find a most effective combination of a molecular biology reaction.

The experiment has four factors: Temperature, RPM, Time, Catalytic activity. And I am measuring the Efficiency of a reaction (EE). How can I find an effective combination of four factors for the highest efficiency(EE)?

• No repeated measurements. All data are independent experimental data

As I understood - EE is parametric data, factors are categorical data (Fixed combinations). Do I have to go for a Fourway ANOVA?

if so is this model correct for the analysis

...

Brief question

How do I write the model statement in proc mixed to study the interaction between a factor with a second factor nested within a third factor?

• So factor_1 is independent from the others (like year is independent from location)
• but factor_2 is a subdivision of factor_3 (like province is a subdivision of state).

I would expect it to be model factor_1 * factor_2(factor_3); but that does not work.

The real problem and context

I am trying to model this Crossed - Nested Design example from an Analysis of Variance and Design of Experiments course of Eberly College of Science using the techniques from this Nested Treatment Design example and the instructions given in this video.

Loading my data with ...

I am trying to predict the mean abundance of animals sighted during different moon phases (factor), using log-transformed abundance data (better fit) and some other variables. The best model (lowest AIC) turned out to include the interaction of phase and survey duration and the cloud cover (both continuous):

...

I am trying to understand the results I got for a fake dataset. I have two independent variables, hours, type and response pain.

First question: How was 82.46721 calculated as the lsmeans for the first type?

Second question: Why is the standard error exactly the same (8.24003) for both types?

Third question: Why is the degrees of freedom 3 for both types?

...