2

I have made the graph below which shows the SE area around lines. However, I can't work out why they are so large for the 3rd graph.

Graph code:

  dg10<-summarySE(vtrs, measurevar="prop.wsls", groupvars =c("t_blocked","attempts","group"))
  t <- ggplot(data=dg10, aes(x=t_blocked, y = prop.wsls,colour=group))+
    theme_minimal()+
    theme(axis.text = element_text(size = 15)) +
    theme(axis.title.y = element_text(face="bold")) + 
    theme(axis.title.x = element_text(face="bold")) +
    theme(axis.text.y = element_text(size=10)) +
    theme(axis.text.x = element_text(size=8,angle=90)) +
    geom_smooth (method= 'lm',se=TRUE,size=0.7)+
    facet_grid(~attempts)+
    xlim(50,1550)

enter image description here

When I calculate the SE using the describe function, it returns the below. As you can see from the last column I expected the se area in the graph to be larger for #3, but I think it seems far too big given it's ratio to the other levels. 

describeBy(vtrs$prop.wsls,group=vtrs$attempts,mat=TRUE,digits=5) # displays SE

    item group1 vars     n    mean      sd median trimmed mad min max range     skew kurtosis      se
X11    1      1    1 15389 0.69556 0.46018      1 0.74442   0   0   1     1 -0.84987 -1.27780 0.00371
X12    2      2    1 10768 0.72901 0.30986      1 0.76619   0   0   1     1 -0.69631 -0.49109 0.00299
X13    3      3    1   873 0.81558 0.23292      1 0.85598   0   0   1     1 -1.00873  0.24935 0.00788
X14    4      4    1  1719 0.84540 0.18898      1 0.87509   0   0   1     1 -1.05780  0.61456 0.00456

What I'm unsure of is whether the graph is correct and therefore to understand why it comes out so large in the graph. Particularly, why is the standard error around the lines so much larger than the values reported in the 'describe' table. OR if I am making some mistake here, either in my code or understanding of SE.

Any help would be massively appreciated!

poppy039
  • 21
  • 2
  • I believe this is related to a point I made at https://stats.stackexchange.com/a/13317/919 (following "an easy way to get a small $R^2$..."); namely, fits look much less significant when they are based on much narrower ranges of the explanatory variable. You therefore might wish to report summary statistics for the explanatory variable as well as the response variable. – whuber May 21 '20 at 14:37
  • The R code appears fine. You should expect wider bands when fitting a model through a more narrow window of x values. The within-group standard errors are more than half that of group (attempt) 3. Is it the visualization that is bothering you? Try arranging all facets into one column. For instance, try faceting in the row dimension via `facet_grid(attempts ~ .)` to adjust the width of the plots. Or, try `scales = "free_x"` so you're not sharing the same x scale across all four of your facets. – Thomas Bilach May 25 '20 at 21:21

0 Answers0