Scatter plots of factors scores

[andkov]

The latest version of the graph that plots the scatters of factor scores obtained from the .gh5 looks as follows: The graphs could be viewed grouped by process pairs for females and males

[wibeasley]

@andkov, I like the graphs. I have three reactions.

1. Are panels c & d flipped from their intended position? Something doesn't seem balanced to me.

2. Consider forcing the p-values to have ~3 digits after the decimal.

   Shooting from the hip, I recommend replacing

   paste0("R_ICIP: ", dsL$R_IPIC_est[1]," (",dsL$R_IPIC_se[1],"), p = ",dsL$R_IPIC_pval[1] )

   with

   sprintf("R_ICIP: %f (%f), p=%0.3f", dsL$R_IPIC_est[1], dsL$R_IPIC_se[1], dsL$R_IPIC_pval[1])

   If you want to get fancier and remove the leading zero, see our recent table.

[ampiccinin]

Please see HRS_F_grip-pef (probably not a quick Q, so you may not want to look at it until tomorrow): Scatter and line look positive, but R_SCSP is negative. Mismatch? Also – are these the SE from Mplus? I’ve started to mistrust them, as I expect that the SE of r should not exceed 1. I’ve computed them in Excel for the manuscript, but it would be good to eventually figure out what Mplus is doing.

[andkov]

1) I'm sad that symmetry is not apparent. I had the following in mind: Left-right diagonal.

• top lef(a) : intercepts of the two processes.
• bottom right(d): slopes of the two processes.

Right-left diagonal.

• top right (b) : intercepts by slopes of process 1.
• bottom left(c) : intercepts by slopes of process 2.

I might switch the locations of c and b, to have rows on the same metric.

2) didn't attend to cosmetics at all. I want to take a look at what you did for the tables and implement here.

3) Same as 2.

[wibeasley]

• [ ] Oh, don't be sad, that makes me sad too. :cry:

  I see the pattern now, and I like it. I think it needs only more explicit labels, that are possibly redundant. For instance in the top left panel, it would have helped me if the x and y axis titles were "grip intercept" and "gait intercept". Keep the redundant labels in the top left corners.

• [ ] Here's a graph I did a couple of years ago that might have some elements you'd like for the diagonal panels.

  • y=x diagonal line.
  • linear prediction line.
  • x & y means (is the mean of the factors salient for you?)
  • prediction equation and/or correlation, that's programmatically inserted. In your case, I think it would look sharp if R_ICIP were translated to something like r(b0[gait],b0[grip]). And realistically, the code would look more like

  substitute(italic(r)(italic(b)0[ip], italic(b)0[ic]), list(ip=dsL$process1[1], cp=dsL$process2[1]))

  
  

  (If you were taking this test, hope that you get graded by rater 1031 or 1034, and not 1030.)

  ReliabilityPair <- function( dsPlot, xName, yName, jitterAmount=.25, mainLabel=NULL ) {  
  m <- lm(as.formula(paste(yName, "~", xName)), dsPlot)
  eqn <- as.character(as.expression( #See Recipe 5.9 in Chang, 2013
    substitute(italic(y)==a + b * italic(x) * "," ~ ~italic(r)^2 ~ "=" ~ r2,
               list(a=format(coef(m)[1], digits=3),#The intercept
                    b=format(coef(m)[2], digits=3), #The slope
                    r2=format(summary(m)$r.squared, digits=3)))
  ))
  
  g <- ggplot(dsPlot, aes_string(x=xName, y=yName)) +
    geom_abline(color=alpha("turquoise", alpha=.2), size=2) +
    annotate("text", label=eqn, x=Inf, y=Inf, hjust=1.1, vjust=1.5, parse=TRUE, size=6, color="gray60") +
    annotate("text", label="italic(bar(x))", x=mean(dsPlot[, xName], na.rm=T), y=Inf, hjust=.5, vjust=1.5, parse=TRUE, size=7, color="gray60") +
    annotate("text", label="italic(bar(y))", x=Inf, y=mean(dsPlot[, yName], na.rm=T), hjust=1.5, vjust=.5, parse=TRUE, size=7, color="gray60") +
    geom_vline(x=mean(dsPlot[, xName], na.rm=T), color=rgb(.3, .3, .1, .2), size=3) +
    geom_hline(y=mean(dsPlot[, yName], na.rm=T), color=rgb(.3, .3, .1, .2), size=3) +
    geom_smooth(method="lm", color="orange", fill="orange", alpha=.2, na.rm=T) +
    geom_smooth(method="loess", color="purple", fill="purple", alpha=.2,  na.rm=T) +
    geom_point(stat="identity", position = position_jitter(w=jitterAmount, h=jitterAmount), shape=1) +  
    coord_fixed(xlim=c(1-jitterAmount-.1, 5+jitterAmount+.1), ylim=c(1-jitterAmount-.1, 5+jitterAmount+.1)) +
    labs(title=mainLabel) +
    theme_bw() 
  g
  }

• [ ] I love how you used a divergent color scheme and alpha to include age information. One little caution: the last rows of the data.frame are plotted last, which may artificially shift the visible amount of green or purple. With a high degree of overplotting like this, consider randomizing the order of the dataframe before you plot it. I also like how you tucked away the legend.

[andkov]

@wibeasley ,

1. I agree about the labels. I was on the fence between "grip" and "grip intercept" and chose the former precisely to eliminate redundancy. But i more followed the principle than made a decision. Your reaction settles it for me: i'm ok with that redundancy if it helps improve usability.
2. Very timely script!! I was recently trying to figure just that, out that. Serendipitously, this is exactly the topic of my today's lecture in pys532 and I am putting up an example similar to what you've shown.
3. Good thought, it hasn't even occur to me. I'll randomize the order of id before feeding the dataframe to the plotting function. Funny about the legend. I spent about 10 minutes creating the plots and about 3 hours trying to figure out how to tuck away that legend. And this is not the first time things like these happen.