Generic method for creating an eikosogram

Description

Generic method for creating an eikosogram

Usage

eikos(
  y,
  x = NULL,
  data = NULL,
  marginalize = NULL,
  main = "",
  main_size = 16,
  ylabs = TRUE,
  ylab_rot = 0,
  yname_size = 12,
  yvals_size = 12,
  yaxs = TRUE,
  yprobs = NULL,
  yprobs_size = 8,
  xlabs = TRUE,
  xlab_rot = 0,
  xname_size = 12,
  xvals_size = 12,
  xaxs = TRUE,
  xprobs = NULL,
  xprobs_size = 8,
  vertical_xprobs = TRUE,
  ispace = list(bottom = 8, left = 2, top = 2, right = 5),
  legend = FALSE,
  col = NULL,
  bottomcol = "steelblue",
  topcol = "snow2",
  lcol = "black",
  draw = TRUE,
  newpage = TRUE,
  lock_aspect = TRUE
)

Arguments

See Also

eikos.default

eikos.formula

Examples

eikos("Hair", "Eye", data=HairEyeColor, legend = TRUE)
eikos(gear ~ cyl, data = mtcars)
eikos(Admit ~  Gender + Dept, data = UCBAdmissions,  
      yaxs = FALSE, xaxs = FALSE, 
      lock_aspect = FALSE, 
      xlab_rot = 90, xvals_size = 8,
      ispace = list(bottom = 15))

Create a new eikosogram

Description

Return a grid graphic object (grob) and draw an eikosogram if draw = TRUE.

Usage

## Default S3 method:
eikos(
  y,
  x = NULL,
  data = NULL,
  marginalize = NULL,
  main = "",
  main_size = 16,
  ylabs = TRUE,
  ylab_rot = 0,
  yname_size = 12,
  yvals_size = 12,
  yaxs = TRUE,
  yprobs = NULL,
  yprobs_size = 8,
  xlabs = TRUE,
  xlab_rot = 0,
  xname_size = 12,
  xvals_size = 12,
  xaxs = TRUE,
  xprobs = NULL,
  xprobs_size = 8,
  vertical_xprobs = TRUE,
  ispace = list(bottom = 8, left = 2, top = 2, right = 5),
  legend = FALSE,
  col = NULL,
  bottomcol = "steelblue",
  topcol = "snow2",
  lcol = "black",
  draw = TRUE,
  newpage = TRUE,
  lock_aspect = TRUE
)

Arguments

Examples

eikos("Hair", "Eye", data=HairEyeColor, legend = TRUE)
eikos("Hair", "Eye", data=HairEyeColor, 
      legend = TRUE, ylabs = FALSE, 
      yname_size = 16, yvals_size = 8)
eikos("Hair", "Eye", data=HairEyeColor, 
      legend = TRUE, ylabs = FALSE, 
      yprobs = seq(0.2, 1, .2))
eikos("Eye", "Hair", data=HairEyeColor, yprobs = seq(0,1,0.25),
      yname_size = 20, xname_size = 20,
      col = c("sienna4", "steelblue", "darkkhaki", "springgreen3"),
      lcol = "grey10",
      lock_aspect = FALSE)

Draw eikosogram using a formula to identify response and conditioning variates

Description

Draw eikosogram using a formula to identify response and conditioning variates

Usage

## S3 method for class 'formula'
eikos(
  y,
  x = NULL,
  data = NULL,
  marginalize = NULL,
  main = "",
  main_size = 16,
  ylabs = TRUE,
  ylab_rot = 0,
  yname_size = 12,
  yvals_size = 12,
  yaxs = TRUE,
  yprobs = NULL,
  yprobs_size = 8,
  xlabs = TRUE,
  xlab_rot = 0,
  xname_size = 12,
  xvals_size = 12,
  xaxs = TRUE,
  xprobs = NULL,
  xprobs_size = 8,
  vertical_xprobs = TRUE,
  ispace = list(bottom = 8, left = 2, top = 2, right = 5),
  legend = FALSE,
  col = NULL,
  bottomcol = "steelblue",
  topcol = "snow2",
  lcol = "black",
  draw = TRUE,
  newpage = TRUE,
  lock_aspect = TRUE
)

Arguments

Examples

eikos(Eye ~ Hair + Sex, data=HairEyeColor)
eikos(Hair ~ ., data=HairEyeColor, 
      yaxs = FALSE, ylabs = FALSE,
      legend = TRUE, 
      col = c("black", "sienna4", 
              "orangered", "lightgoldenrod" ))
eikos(Hair ~ ., data=HairEyeColor, xlab_rot = 30,
      yprobs = seq(0.1, 1, 0.1),
      yvals_size = 10,
      xvals_size = 8,
      ispace = list(bottom = 10),
      bottomcol = "grey30", topcol = "grey70",
      lcol = "white")
eikos(Hair ~ ., data=HairEyeColor, xlab_rot = 30,
      marginalize = "Eye",
      yvals_size = 10,
      xvals_size = 8,
      ispace = list(bottom = 10),
      bottomcol = "grey30", topcol = "grey70",
      lcol = "white")
eikos(Hair ~ ., data=HairEyeColor, xlab_rot = 30,
      marginalize = c("Eye", "Sex"),
      yvals_size = 10,
      xvals_size = 8,
      ispace = list(bottom = 10),
      bottomcol = "grey30", topcol = "grey70",
      lcol = "white")

Create eikosogram data frame

Description

Eikos helper function used to convert data.

Usage

eikos_data(y, x, data, marginalize = NULL)

Arguments

Create eikosogram legend

Description

Eikos helper function used to create legend.

Usage

eikos_legend(
  labels,
  title = NULL,
  yname_size = 12,
  yvals_size = 12,
  col,
  margin = unit(2, "points"),
  lcol = "black"
)

Arguments

eikos helper function. Returns grob with x axis labels.

Description

eikos helper function. Returns grob with x axis labels.

Usage

eikos_x_labels(
  x,
  data,
  margin = unit(10, "points"),
  xname_size = 12,
  xvals_size = 10,
  lab_rot = 0
)

Arguments

Value

gList with x labels and x-axis names as grob frames.

Create grob with eikosogram x-axis probabilities

Description

Creates x axis grob to be placed on eikosogram. Called by eikos functions.

Usage

eikos_x_probs(
  data,
  xprobs = NULL,
  xprobs_size = 8,
  margin = unit(2, "points"),
  rotate = TRUE
)

Arguments

Value

textGrob with x-axis probabilities.

eikos helper function. Returns grob with y axis labels.

Description

eikos helper function. Returns grob with y axis labels.

Usage

eikos_y_labels(
  y,
  data,
  margin = unit(2, "points"),
  yname_size = 12,
  yvals_size = 10,
  lab_rot = 0
)

Arguments

Value

grobFrame with response variable labels and axis text

Create grob with eikosogram y-axis probabilities

Description

Creates y axis grob to be placed on eikosogram. Called by eikos functions.

Usage

eikos_y_probs(data, yprobs, yprobs_size = 8, margin = unit(2, "points"))

Arguments

Value

textGrob with y-axis probabilities.

Mining medical records (fictional)

Description

An entirely artificially constructed data set and context designed for classroom discussion and analysis.

A medical data mining context is given in detail below. In light of the context, interesting scientific questions will arise as to the data collection, and how the results should, or should not, be interpreted. It should also raise questions on what might be done in any follow up studies.

Instructors might choose to invent their own context.

Format

A data frame with 16 rows and 5 variables (providing the counts for a 2x2x2x2 contingency table).

Age

A two level factor recording one of two age groups: "20-39" or "40-59".

Sex

A two level factor recording sex: "Male" or "Female".

Treatment

A two level factor recording the treatment received: "A" or "B".

Outcome

A two level factor recording patient outcome after treatment: "Recovered" or "Died".

Freq

The frequency count of patients having that combination of factors.

Details

One fictional context (constructed in March 2020) for this data set is given below (in the PPDAC style of Mackay and Oldford (2000)).

Problem:

A disease epidemic has broken out in the population of some country. It is thought that adults under the age of 60 appear to be particularly vulnerable. Both men and women contract the disease and need to be treated. Those who go untreated die within 5 days of contracting the disease.

The medical community has tried two quite different approaches to treat patients having the disease – call these 'Treatment A' and 'Treatment B'. For the health of the country, it is important to determine which of these two treatments is more effective.

Plan:

To investigate which is the better treatment, it was decided to mine the medical records from another country of those who had contracted the disease and had been treated with one of the two treatments. Patients treated with either A or B survive the disease and recover fully; some however still die.

Electronic medical records available from several of the more populous districts are accessible. These can be searched to provide records from patients that have received treatment. It was decided that there should be the same number of records drawn for each treatment.

Moreover, concern was raised that the investigation have gender balance (i.e. equal numbers of males and females). So, to make sure that both sexes were equally represented, it was also decided that the number of female patients would be the same as the number of male patients.

Finally, it was desirable to detect even small differences in success rates of the two treatments since small differences could mean many more lives being saved. A sample size of about n = 3,000 was decided on.

Records would be collected until 3,000 were found, 1500 of which were treated with 'A', 1500 with 'B', and there were equal numbers of males and females in the study.

Data:

In this stage, the plan is executed. Instead of 1500 records of treatment 'A' and 'B', 1600 of each were found. The number of males and females was kept equal (now 1600 of each sex).

The process was to search the records in order, selecting those first encountered to get 1600 for each treatment and 1600 of each sex. Many records might be discarded whenever one quota was met and the search continued to meet the other quotas. It was also noticed that the patient's age was available for each record, so that the effect of treatment on younger and older adults might also be considered.

The counts which fell into the various categories were assembled into the data presented here.

Author(s)

R.W. Oldford

References

R.J. MacKay and R.W. Oldford 2000, 'Scientific Method, Statistical Method, and the Speed of Light', Statistical Science, Volume 15, No. 3, pp. 254-278. <doi:10.1214/ss/1009212817>

Tuberculosis 1910 death rates in New York and in Richmond

Description

Tuberculosis 1910 death rates in New York and in Richmond

Format

A 2 x 2 x 2 table of counts involving three binary variables

Group

One of two racial groups: "White" or "Colored"

City

One of two U.S. cities: "New York" or "Richmond"

Total

Either total deaths from tuberculosis or total population in 1910: "Deaths" or "Population"

Details

These are historical data taken from page 449 of Cohen and Nagel's 1934 "Introduction to Logic and Scientific Method". For this reason, the original names of the racial groups have been retained.

The data are of special historical interest in Statistics because they are one of the earliest recorded instances of a real Simpson's paradox (Simpson 1951) occurring in practice (see Blyth 1971). Preserving this historical context, the questions posed by Cohen and Nagel (1934) are also recorded here using their own words. The data and questions appear at the back of their book as exercises on "Chapter XVI: Statistical Methods".

In their table, Cohen and Nagel (1934, p. 449) include the "death rates from tuberculosis in Richmond, Virginia, and in New York City in 1910". These rates (in number per 100,000) are easily calculated and so have been excluded from the table given here.

In their words, Cohen and Nagel (1934, p. 449) pose the following two questions as exercise (*emphasis* is theirs):

"a. Does it follow that tuberculosis caused a greater mortality in Richmond than in New York?

b. Notice that the death rate for whites and that for Negroes were *lower* in Richmond than in New York, although the *total* death rate was *higher*. Are the two populations compared really *comparable*, that is, homogeneous?"

Author(s)

R.W. Oldford.

Source

"An Introduction to Logic and Scientific Method" by Morris R. Cohen and Ernest Nagel, (1934), Harcourt, Brace and Company, New York.

References

Blyth, Colin R. 1972. On Simpson's Paradox and the Sure-Thing Principle. Journal of the American Statistical Association, 67, pp.364-366.

Cohen, Morris R.; Nagel, Ernest. 1934. An Introduction to Logic and Scientific Method. Harcourt, Brace and Company. New York.

Simpson, E.H. 1951. The Interpretation of Interaction in Contingency Tables. Journal of the Royal Statistical Society, Series B, 13, pp. 238-241.