Pam_SyracuseVIWorkshops

Motor Skill Scores

SubjectID	W1	W2	W3	W4	W5
100	27	27	NA	NA	NA
103	NA	37	40	NA	43
106	38	36	NA	NA	NA
107	24	25	33	NA	24
122	27	43	NA	37	NA
125	46	NA	45	NA	NA
130	18	NA	NA	NA	NA
135	48	NA	NA	NA	NA
137	28	23	NA	NA	NA
139	NA	37	33	NA	26
143	39	44	40	NA	35
160	38	38	NA	NA	24
172	19	33	31	NA	16

Converting dataframe from wide to long form

Motor Skill Scores for 13 Subjects over 5 Workshops

SubjectID	workshop	score
100	W1	27
103	W1	NA
106	W1	38
107	W1	24
122	W1	27
125	W1	46
130	W1	18
135	W1	48
137	W1	28
139	W1	NA
143	W1	39
160	W1	38
172	W1	19
100	W2	27
103	W2	37
106	W2	36
107	W2	25
122	W2	43
125	W2	NA
130	W2	NA
135	W2	NA
137	W2	23
139	W2	37
143	W2	44
160	W2	38
172	W2	33
100	W3	NA
103	W3	40
106	W3	NA
107	W3	33
122	W3	NA
125	W3	45
130	W3	NA
135	W3	NA
137	W3	NA
139	W3	33
143	W3	40
160	W3	NA
172	W3	31
100	W4	NA
103	W4	NA
106	W4	NA
107	W4	NA
122	W4	37
125	W4	NA
130	W4	NA
135	W4	NA
137	W4	NA
139	W4	NA
143	W4	NA
160	W4	NA
172	W4	NA
100	W5	NA
103	W5	43
106	W5	NA
107	W5	24
122	W5	NA
125	W5	NA
130	W5	NA
135	W5	NA
137	W5	NA
139	W5	26
143	W5	35
160	W5	24
172	W5	16

Missing Motor Skill Score Summary

SubjectID	missing
100	3
103	2
106	3
107	1
122	2
125	3
130	4
135	4
137	3
139	2
143	1
160	2
172	1

• using smoother to see subject specific trend

p1 + geom_line() + stat_smooth(aes(group = SubjectID)) + stat_summary(aes(group = SubjectID),
    geom = "point", fun.y = mean, shape = 17, size = 3) 

• Calculating Smoother for each subject

p1 + geom_line() + stat_smooth(aes(group = SubjectID), method = "lm", se = FALSE) +
  stat_summary(aes(group = SubjectID), geom = "point", fun.y = mean, shape = 17, size = 3) 

• Now, we might be interested in estimating the overall trend in the data. One option is to add a line using locally weighted regression (lowess) to ???smooth??? over all the variability and give a sense of the overall or average trend. It just takes one short line of code and is automatically calculated

• Note again, we use group = 1, so the smooth is not calculated separately for each id

• geom_smooth: method=“auto” and size of largest group is <1000, so using loess. Use ‘method = x’ to change the smoothing method.

p1 + geom_line() + stat_smooth(aes(group = 1)) + stat_summary(aes(group = 1),
    geom = "point", fun.y = mean, shape = 17, size = 3) 

• Looks like between worskshop 1 and 4, an intervention/ or something happened, and we wish to fit a piecewise linear model rather than an overall smooth. We can do this by creating a dummy variable (pre/post intervention) and its interaction with time. The only change is a slightly more complex formula.

• The default is y ~ x. I(x > 1) creates a dummy (TRUE/FALSE) variable if x (time in this case) is greater than 1. The aestrick in the formula asks for the main effects and the interaction between x and the dummy variable from x.

• Now we can see that the trend line jumps after W4, and the slope is allowed to change and it seems drastic indicating their is an interaction between time and intervention (if the change appears minimal, it is a suggestion there is not an interaction between our hypothetical intervention and time)

p1 + geom_line() + stat_smooth(aes(group = 1), method = "lm", formula = y ~
    x * I(x > 2), se = FALSE) + stat_summary(aes(group = 1), fun.y = mean, geom = "point",
    shape = 17, size = 3) 

Randomisation based tests along with t test to compare Worskshop scores with baseline workshop W1

• Preparing the data

• Test code for extracting Complete Cases

MS comparison W1vs W2

## [1] TRUE

## # A tibble: 8 x 3
##   SubjectID    W1    W2
##       <chr> <dbl> <dbl>
## 1       100    27    27
## 2       106    38    36
## 3       107    24    25
## 4       122    27    43
## 5       137    28    23
## 6       143    39    44
## 7       160    38    38
## 8       172    19    33

Complete Cases between W1 and W2

SubjectID	W1	W2
100	27	27
106	38	36
107	24	25
122	27	43
137	28	23
143	39	44
160	38	38
172	19	33

## Classes 'tbl_df', 'tbl' and 'data.frame':    8 obs. of  3 variables:
##  $ SubjectID: chr  "100" "106" "107" "122" ...
##  $ W1       : num  27 38 24 27 28 39 38 19
##  $ W2       : num  27 36 25 43 23 44 38 33

MS Comparison W1 vs. W3

## # A tibble: 4 x 3
##   SubjectID    W1    W3
##       <chr> <dbl> <dbl>
## 1       107    24    33
## 2       125    46    45
## 3       143    39    40
## 4       172    19    31

##   SubjectID               W1              W3       
##  Length:4           Min.   :19.00   Min.   :31.00  
##  Class :character   1st Qu.:22.75   1st Qu.:32.50  
##  Mode  :character   Median :31.50   Median :36.50  
##                     Mean   :32.00   Mean   :37.25  
##                     3rd Qu.:40.75   3rd Qu.:41.25  
##                     Max.   :46.00   Max.   :45.00

MS Comparison W1 vs. W4, not possible as n=1

## # A tibble: 1 x 3
##   SubjectID    W1    W4
##       <chr> <dbl> <dbl>
## 1       122    27    37

MS Comparison W1W5,

## # A tibble: 4 x 3
##   SubjectID    W1    W5
##       <chr> <dbl> <dbl>
## 1       107    24    24
## 2       143    39    35
## 3       160    38    24
## 4       172    19    16

##   SubjectID               W1              W5       
##  Length:4           Min.   :19.00   Min.   :16.00  
##  Class :character   1st Qu.:22.75   1st Qu.:22.00  
##  Mode  :character   Median :31.00   Median :24.00  
##                     Mean   :30.00   Mean   :24.75  
##                     3rd Qu.:38.25   3rd Qu.:26.75  
##                     Max.   :39.00   Max.   :35.00

Creating Final Table of p-values related to randomisation based test and t-test for paired comparisons

P-values for Randomisation and t-test

	Rand test	ttest
W2 vs. W1	0.156	0.109
W3 vs. W1	0.184	0.095
W4 vs. W1	NA	NA
W5 vs. W1	1.000	0.909