Data Transformations¶
Normalizing and scaling data¶
The rescale() function provides 8 different scaling options and returns the rescaled data along with the location and scale factors.
Method |
Location |
Scale Factor |
|---|---|---|
|
0 |
Euclidean length |
|
median |
Absolute deviation from median |
|
0 |
Maximum absolute value |
|
(Max+Min)/2 |
Range/2 |
|
Minimum |
Range |
|
Mean |
Standard deviation |
|
0 |
Sum |
|
0 |
Standard deviation around origin |
Example: Rescaling with a specified scaling method¶
// Create a column vector
x = { 12.5,
18.2,
10.8,
8.3,
15.4,
21.5,
14.6,
16.7 };
// Standardize 'x' and return the location and scaling factors
{ x_s, location_s, scale_factor_s } = rescale(x, "standardize");
// Rescale the `x` using the median
{ x_m, location_m, scale_factor_m } = rescale(x, "mad");
print "Standardized rescaling:";
print "x_s = " x_s;
print "location_s = " location_s;
print "scale_factor_s = " scale_factor_s;
print "Median rescaling:";
print "x_m = " x_m;
print "location_m = " location_m;
print "scale_factor_m = " scale_factor_m;
After the code above:
Standardized rescaling:
x_s =
-0.53463295
0.81977052
-0.93857785
-1.5326145
0.15444952
1.6038989
-0.035642197
0.46334856
location_s = 14.75
scale_factor_s = 4.21
Median rescaling:
x_m =
-0.87719298
1.1228070
-1.4736842
-2.3508772
0.14035088
2.2807018
-0.14035088
0.59649123
location_m = 15.00
scale_factor_m = 2.85
The rescale() function can also be used with a known location and scale factor to rescale data.
Example: Rescaling using known location and scaling factors¶
// Additional observations
x2 = { 9.3,
10.9,
11.1,
9.1,
14.6,
18.4,
20.2,
18.5 };
// Rescale matrix above using
// location and scale matrix
// from above
x_s2 = rescale(x2, location_s, scale_factor_s);
After the code above x_s2 is equal to:
-1.2949998
-0.91481638
-0.86729345
-1.3425227
-0.035642197
0.86729345
1.2949998
0.89105492
The rescale() function can also be used to rescale multiple columns at time.
Example: Rescaling multiple columns¶
// Create a matrix with 2 columns
x = { 12.5 1088.5,
18.2 879.3,
10.8 1232.0,
8.3 1189.8,
15.4 932.1,
21.5 1009.2,
14.6 656.7,
16.7 1251.5 };
// Standardize 'x' and return the location and scaling factors
{ x_s, location, scale_factor } = rescale(x, "standardize");
print "x_s = " x_s;
print "location = " location;
print "scale_factor = " scale_factor;
x_s =
-0.53463295 0.28751716
0.81977052 -0.73869039
-0.93857785 0.99144060
-1.5326145 0.78443315
0.15444952 -0.47968581
1.6038989 -0.10148025
-0.035642197 -1.8306302
0.46334856 1.0870957
location = 14.750000 1029.8875
scale_factor = 4.2084948 203.85740
Recoding and reclassifying¶
GAUSS provides a variety of tools for recoding and reclassifying data. These functions can be divided into functions for numeric data and functions for categorical data.
Numeric Functions |
Description |
Recoding specifier |
|---|---|---|
Replaces specified values of a matrix, array or string array. |
User-specified values. |
|
Replaces values of a matrix or array within specified ranges. |
User-specified values. |
|
Creates a new matrix based on recoding of an existing numeric vector. |
Based on logical expression. |
|
Recodes the values of an existing vector of numeric data. |
Based on logical expression. |
|
Substitutes new values for old values in a matrix, depending on the outcome of a logical expression. |
Based on logical expression. |
Categorical Functions |
|
|---|---|
Changes relative order of categorical variable. This changes the key values associated with the categorical labels. |
|
Replaces the labels of categorical variables with new labels. |
Recoding and reclassifying non-categorical data
Both the code() and recode() procedures can be used to recode data using conditional expressions.
The code() procedure:
Creates a new matrix which splits existing data into classes.
Uses N logical expressions to determine N+1 classes.
Works for vectors only.
Example: Coding blood pressure data to create a new (binary) class variable¶
// Blood pressure data
x = { 91,
121,
99,
135,
110,
155 };
// Values for the classes
new_val = { 1, 2 };
/*
** Create a vector containing a 1 for every element
** which is less than 120, or a 0 otherwise
*/
logical = x .< 120;
/*
** Create a new vector which contains the class
** assignment for each element in 'x'
*/
x_class = code(logical, new_val);
The code above generates a new vector x_class which splits the original data into two classes based on whether x is less than 120.
x = 91 logical = 1 x_class = 1
121 0 2
99 1 1
135 0 2
110 1 1
155 0 2
Example: Coding blood pressure data to create a new multi-class variable¶
// Blood pressure data
x = { 91,
121,
99,
135,
110,
155 };
// Values for the classes
new_val = { 1,
2,
3 };
// Create a vector containing a 1 for every element
// which is less than 100, or a 0 otherwise
logical_1 = x .<= 100;
// Create a vector containing a 1 for every element
// which is between 100 and 120, or a 0 otherwise
logical_2 = x .> 100 .and x .<= 120;
// Form a 2 column logical vector using
// horizontal concatenation
logical = logical_1 ~ logical_2;
// Create a new vector which contains the class
// assignment for each element in 'x'
x_class = code(logical, new_val);
Now x_class splits the original data into three classes based on whether x is less than or equal to 100, falls between 100 and 120, or is greater 120.
x = 91 logical = 1 0 x_class = 1
121 0 0 3
99 1 0 1
135 0 0 3
110 0 1 2
155 0 0 3
Note
The setColLabels() function can be used to specify x_class as a categorical variable and to assign labels to the classes.
Recoding values of an existing vector
The recode() procedure :
Replaces specific values of an existing vector with new values.
Uses a logical expression to determine where and how to replace values.
Is valid for vectors.
Some notes to remember about recode():
There should be no more than a single 1 in any row of logical expression matrix.
For any given row of a data matrix and logical expression matrix, if a column of the logical expression is 1, the corresponding replacement values with replace the original element of the data matrix.
If every column of logical expression matrix contains a 0, the original value of the data matrix will be unchanged.
Example: Recoding numeric values based on ranges¶
x = { 20,
45,
32,
63,
29 };
// Create 4 column vectors with a 1 where the statement
// evaluates as 'true'
// Check if 20 < x <= 30
e1 = (20 .lt x) .and (x .le 30);
// Check if 30 < x <= 40
e2 = (30 .lt x) .and (x .le 40);
// Check if 40 < x <= 50
e3 = (40 .lt x) .and (x .le 50);
// Check if 50 < x <= 60
e4 = (50 .lt x) .and (x .le 60);
// Horizontally concatenate the column vectors into a 5x4
// matrix
logical = e1~e2~e3~e4;
v = { 1.2,
2.4,
3.1,
4.6 };
// Replace elements of 'x' with elements from 'v' based upon
// the 0's and 1's in 'e'
x_new = recode(x, logical, v);
Note that in this example x_new is as follows:
0 0 0 0
0 0 1 0
logical = 0 1 0 0
0 0 0 0
1 0 0 0
// Since the third column of the second row of 'e' is equal
// to 1, the second row of 'y' is set equal to the third
// element of 'v', etc.
20.000000
3.1000000
x_new = 2.4000000
63.000000
1.2000000
Reclassifying data
The reclassify() and reclassifyCuts() procedures can be used to reclassify existing values to new values.
The reclassify() procedure:
Replaces values in a from input with values specified in a to input.
Works for matrices, arrays, and string arrays.
Can be used to reclassify matrices to string arrays and vice versa.
Note
The reclassify() function can reclassify matrices to string arrays but does not create a dataframe. To create a dataframe with a string labels from an existing matrix see asDF().
Example: Change instances of 1, 2 and 3 to ‘low’, ‘medium’ and ‘high’.¶
// Vector to be changed
x = { 2,
3,
2,
1,
2,
3 };
from = { 1,
2,
3 };
// Create a 3x1 string array using
// string vertical concatenation operator
to = "low" $| "medium" $| "high";
x_new = reclassify(x, from, to);
print x_new;
After the code above, x_new is equal to:
medium
high
medium
low
medium
high
In this case, if the number of specified strings in to is less than the number of unique values in x, the unmapped values will be converted directly into strings.
// Vector to be changed
x = { 2,
3,
2,
1,
2,
3 };
from = { 1,
2};
// Create a 3x1 string array using
// string vertical concatenation operator
to = "low" $| "medium";
x_new = reclassify(x, from, to);
print x_new;
Now x_new is
medium
3
medium
low
medium
3
Example: Change instances of tea types: ‘black’, ‘green’, ‘oolong’ to 9.95, 11.95 and 10.50, respectively.¶
string orders = { "green",
"green",
"oolong",
"green",
"green",
"green",
"black" };
string tea_types = { "black",
"green",
"oolong" };
price = { 9.95, 11.95, 10.50 };
order_prices = reclassify(orders, tea_types, price);
print order_prices;
The vector order_prices is equal to:
11.95
11.95
10.50
11.95
11.95
11.95
9.95
In this case, if the number of specified values in to is less than the number of unique strings in x, unmapped strings will be reclassified as missings:
string orders = { "green",
"green",
"oolong",
"green",
"green",
"green",
"black" };
string tea_types = { "black",
"green" };
price = { 9.95, 11.95 };
order_prices = reclassify(orders, tea_types, price);
print order_prices;
Now order_prices is:
11.950000
11.950000
.
11.950000
11.950000
11.950000
9.9500000
The reclassifyCuts() procedure:
Splits the data in x into classes based on specified cutoff values.
Works for matrices and arrays.
Cutoff points can be used to define the right endpoint of an interval or the starting points of the next interval. The default is to use the cutoff points as starting points of the next interval.
Example: Basic sequence¶
// Create column vector to place in categories
x = { 0,
0.1,
0.2,
0.3,
0.4,
0.5,
0.6,
0.7 };
// Cut points for data in 'x'
cut_pts = { 0.2,
0.5 };
// Class 0: x <= 0.2
// Class 1: 0.2 < x <= 0.5
// Class 2: 0.5 < x
r_open = reclassifyCuts(x, cut_pts);
// Class 0: x < 0.2
// Class 1: 0.2 <= x < 0.5
// Class 2: 0.5 <= x
r_closed = reclassifyCuts(x, cut_pts, 1);
print "x = " x;
print;
print "r_open = " r_open;
print;
print "r_closed = " r_closed;
print;
print "cut_pts = " cut_pts;
This results in:
x =
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
r_open =
0.00
0.00
0.00
1.0
1.0
1.0
2.0
2.0
r_closed =
0.00
0.00
1.0
1.0
1.0
2.0
2.0
2.0
cut_pts =
0.20
0.50
Example: Classifying blood pressure data¶
// Create a column of blood pressure data
bp = { 87,
154,
127,
112,
159,
90,
151,
109,
125,
107 };
// Assign cut points
cut_pts = { 120, 140 };
// Create categorical variable
bp_category = reclassifyCuts(bp, cut_pts);
print "bp = " bp;
print;
print "bp_category = " bp_category;
print;
print "cut_pts = " cut_pts;
This splits the data in bp into three categories: those that fall below 120, those that greater than or equal to 120 but less than 140, and those that are greater than or equal to 140:
87
154
127
112
bp = 159
90
151
109
125
107
0
2
1
0
bp_category = 2
0
2
0
1
0
cut_pts = 120
140
Substituting values¶
The substute() function replaces values in a matrix based on the outcome of a logical expression.
Example: Setting very small values to zero¶
// Create example vector
x = { 3.8e-21,
1.0,
3.5,
2.7e-18,
0.5,
3.0,
1.1e-16,
0.5,
2.2,
4.0 };
// Substitute all values less than 2.2e-16 with a zero
x_new = substute(x, x .< 2.25e-16, 0);
This results in x_new equal to:
0.00000000
1.0000000
3.5000000
0.00000000
0.50000000
3.0000000
0.00000000
0.50000000
2.2000000
4.0000000
Recoding categorical data
The recodeCatLabels() can be use to change the labels on categorical variables in a dataframe.
Example: Recoding categories in yarn dataset¶
// Load data
fname = getGAUSSHome("examples/yarn.xlsx");
yarn = loadd(fname, "cat(yarn_length) + cat(amplitude) + cat(load) + cycles");
// Get column labels for yarn_length
{ labels, keyvalues } = getColLabels(yarn, "yarn_length");
// Print results
sprintf("%11s", "Key"$~"Labels");
sprintf("%10.0f %10s", keyvalues, labels);
// Recode yarn_length variable from
// 'low', 'medium', and 'high'
// to 'sm', 'md', 'lg'
yarn_recoded = recodecatlabels(yarn, "low"$|"med"$|"high", "sm"$|"md"$|"lg", "yarn_length");
// Get column labels for yarn_length
{ labels, keyvalues } = getColLabels(yarn_recoded, "yarn_length");
// Print results
print "Yarn recoded labels";
sprintf("%11s", "Key"$~"Labels");
sprintf("%10.0f %10s", keyvalues, labels);
This prints the following:
Yarn labels
Key Labels
0 high
1 low
2 med
Yarn recoded labels
Key Labels
0 lg
1 sm
2 md
Reordering categorical data
The reorderCatLabels() can be use to change the key values associated with categorical labels.
// Load data
fname = getGAUSSHome("examples/yarn.xlsx");
yarn = loadd(fname, "cat(yarn_length) + cat(amplitude) + cat(load) + cycles");
// Get column labels for yarn_length
{ labels, keyvalues } = getColLabels(yarn, "yarn_length");
// Print results
print "Yarn labels";
sprintf("%11s", "Key"$~"Labels");
sprintf("%10.0f %10s", keyvalues, labels);
// Order labels
yarn_reordered = reordercatlabels(yarn, "med"$|"high"$|"low", "yarn_length");
// Get column labels for yarn_length
{ labels, keyvalues } = getColLabels(yarn_reordered, "yarn_length");
// Print results
print "Reordered yarn labels";
sprintf("%11s", "Key"$~"Labels");
sprintf("%10.0f %10s", keyvalues, labels);
This prints the following:
Yarn labels
Key Labels
0 high
1 low
2 med
Reordered yarn labels
Key Labels
0 med
1 high
2 low
Time Series Transformations¶
While data lags, leads, differences and recursive terms can always be computed using matrix operations, GAUSS also includes built-in tools for these transformations.
Function |
Description |
Format |
|---|---|---|
Lags a matrix by one time period for time series analysis. |
|
|
Lags or leads a matrix a specified number of time periods. Use negative input t to indicate leads. |
|
|
Lags or leads a matrix a specified number of time periods and removes the incomplete rows. Use negative input t to indicate leads. |
|
|
Shifts the columns of a matrix, or dataframe. |
|
|
Computes a vector of autoregressive recursive series. |
|
|
Computes a vector autoregressive recursive (VAR) series. |
|
Lagging data with the lagn or lag1 procedures¶
The lagn() and lag1() procedures are used to lag data without removing or replacing the missing values. These procedures accepts M x T data matrices, x, and the lagn() and procedure accepts an ExE conformable vector of lags.
The ExE conformability requirement means that lagn() can be used to compute:
The same lag of every column of a data matrix.
Specific lags for each column of a data matrix.
Multiple lags of a single vector of data.
Because missing values are not removed by the lagn() and lag1() procedures, the returns from these procedures will always have the same number of rows as the input, x.
Example: Computing a single lag of a matrix with lagn¶
In this example the PPI matrix contains two variables:
A date column named, date
Observed PPI data column named, PPIACO
To compute the same number of lags of each column of the data, a scalar lag input, t can be used:
// Load PPIACO series
// from FRED database
PPI = fred_load("PPIACO");
// Lag the PPI data
// using lagn
PPI_lag_1 = lagn(PPI, 1);
// Preview PPI_lag_1 data
head(PPI_lag_1);
Our preview shows that the first element of the PPI_lag vector is a missing value:
date PPIACO
. .
1913-01-01 12.100000
1913-02-01 12.000000
1913-03-01 12.000000
1913-04-01 12.000000
Example: Computing a different lags of each column of a matrix with lagn¶
To compute different lags of each column of data at the same time, a vector input of lags specifying a separate lag for each column of data can be used. Note that the lag vector must have the same number of elements as the number of columns in the matrix being lagged:
// Load multiple series
// from FRED
data = fred_load("PPIACO + T10Y2Y");
// Lags vectors
lags = 1|2;
// Compute Lags
data_lag_12 = lagn(data["PPIACO" "T10Y2Y"], lags);
// Preview the lagged data
head(data_lag_12)
This computes the first lag of the PPIACO variable and the second lag of the T10Y2Y series:
Example: Computing a different lags of vector of data using lagn¶
// Load PPIACO series
// from FRED database
PPI = fred_load("PPIACO");
// Specify lags vector
lags = 1|2|3;
// Lag just the observations
// of the PPI data
// using lagn
PPI_lag_123 = lagn(PPI[., "PPIACO"], lags);
// Preview PPI_lag_1 data
head(PPI_lag_123);
This computes the first, second, and third lag of the PPIACO variable. Note that in this case, new variables names, PPIACO_2 and PPIACO_3 variables are created for first and second columns.
PPIACO PPIACO_2 PPIACO_3
. . .
12.100000 . .
12.000000 12.100000 .
12.000000 12.000000 12.100000
12.000000 12.000000 12.000000
Lagging data with the lagTrim procedure¶
The lagTrim() procedure removes resulting missing values from lagging the data. Like the lagn() procedure, the lagTrim() procedure accepts a M x T data matrices, x, and an ExE conformable vector of lags.
The return from the lagTrim() procedure will have a number of rows equal to the number of rows of the input x minus the maximum number of lags specified in t.
Example: Computing multiple lags without missing values¶
// Create file name with full path
fname = getGAUSSHome("examples/beef_prices.csv");
// Load all observations of all variables
beef = loadd(fname);
// Create lag vector
lags = 1|2|3;
// Compute lags using lagTrim
beef_lagTrim = lagTrim(beef[., 2], lags);
// Preview lagged data
head(beef_lagTrim);
// Compare number of rows
print "Rows in original data:";
rows(beef);
print "Rows in lagged data:";
rows(beef_lagTrim);
The beef_lagTrim matrix has 282 rows, 3 less than the input data beef:
111.11000 114.49000 116.64000
108.17000 111.11000 114.49000
107.76000 108.17000 111.11000
105.90000 107.76000 108.17000
106.43000 105.90000 107.76000
Rows in original data:
285.00000
Rows in lagged data:
282.00000
Shifting data with the `shiftc procedure¶
The shiftc() procedure shifts columns of a data matrix and requires three inputs:
A N x K matrix of data.
A scalar or 1 x N input specifying the magnitude of the shift.
A scalar or 1 x N input specifying the value to fill in the shifted rows.
The return from the shiftc() procedure will have a number of rows equal to the number of rows of the data input. The shiftc() procedure can be used to fill the shifted rows with values other than missing values.
Example: Shifting columns of a data matrix¶
// Create file name with full path
fname = getGAUSSHome("examples/beef_prices.csv");
// Load all observations of all variables
beef = loadd(fname);
// Trim data to make smaller example set
beef = beef[1:5,.];
// Shift all columns of beef forward 2 rows
// filling the extra rows with a missing value
beef_lag = shiftc(beef, 2, miss());
After the above code:
beef_lag = date beef_price
. .
. .
199201 116.64000
199202 114.49000
199203 111.11000
Using the beef dataframe from the first example:
// Shift all columns of beef forward 2 rows
// filling the extra rows with a missing value
beef_lag_0 = shiftc(beef, 2, 0);
After the above code:
beef_lag_0 = date beef_price
0 0
0 0
199201 116.64000
199202 114.49000
199203 111.11000
Dummy variables¶
Categorical variables in dataframes will automatically be treated as dummy variables in GAUSS estimation routines. This means no extra steps are necessary to include categorical variables in regression.
Example: Include a categorical variable in OLS¶
// Load data
fname = getGAUSSHome("examples/auto2.dta");
// Include the `rep78`
// categorical variable in
// ols estimation
call olsmt(fname, "price ~ mpg + rep78");
The categorical variable rep78 will automatically be included in the OLS regression as a dummy variable with the base case excluded from the regression. In addition, the category labels will be displayed in the printed output table.
Standard Prob Standardized Cor with
Variable Estimate Error t-value >|t| Estimate Dep Var
---------------------------------------------------------------------------------------
CONSTANT 10450 2251.04 4.64229 0.000 --- ---
mpg -280.261 61.5767 -4.55142 0.000 -0.564519 -0.455949
rep78: Fair 877.635 2063.28 0.425358 0.672 0.0971824 -0.0223477
rep78: Average 1425.66 1905.44 0.748204 0.457 0.24444 0.0859051
rep78: Good 1693.84 1942.67 0.871914 0.387 0.257252 -0.015317
rep78: Excellent 3131.98 2041.05 1.5345 0.130 0.396546 -0.035102
The categories of rep78, "Fair", "Average", "Good", and "Excellent", are included as dummy variables in the regression. The "Poor" category is excluded from the regression, as it is the base case.
Example: Including a categorical variable in GLM estimation¶
// Load data
fname = getGAUSSHome("examples/auto2.dta");
// Loadd data and remove missing values
data = packr(loadd(fname, "price + mpg + rep78"));
// Include the `rep78`
// categorical variable in
// linear regression using glm
call glm(data, "price ~ mpg + rep78", "normal");
Standard Prob
Variable Estimate Error t-value >|t|
---------------- ------------ ------------ ------------ ------------
CONSTANT 10450 2251 4.6423 < 0.0001
mpg -280.26 61.577 -4.5514 < 0.0001
rep78: Fair 877.63 2063.3 0.42536 0.672025
rep78: Average 1425.7 1905.4 0.7482 0.457121
rep78: Good 1693.8 1942.7 0.87191 0.386566
rep78: Excellent 3132 2041 1.5345 0.129915
Outside of estimation, dummy variables can be created using a number of procedures:
Functions |
Description |
|---|---|
Creates dummy variables from discrete data that is split into classes. |
|
Creates dummy variables from continuous data based on break points. The highest (rightmost) category is bounded on the right. |
|
Creates dummy variables from continuous data based on break points. The highest (rightmost) category is unbounded on the right, and a specified column of dummies is dropped. |
|
Creates dummy variables from continuous data based on break points. The highest (rightmost) category is unbounded on the right. |
Example: Create dummy variables based on BP classes¶
This example builds on an earlier example, in which BP data was split into 3 classes using reclassify().
// Classified BP data
bp_class = { 1,
3,
1,
3,
2,
3 };
// Create matrix of dummy
// variables using design
dv_bp_classes = design(bp_class);
After this code dv_bp_classes is equal to:
dv_bp_classes;
1 0 0
0 0 1
1 0 0
0 0 1
0 1 0
0 0 1
Example: Create dummy variables from continuous BP data¶
The dummybr() variable can be used to generate dummy variables from the ranges of
original BP data.
// Create a column of blood pressure data
bp = { 91,
121,
99,
135,
110,
155 };
// Create breakpoints
v = { 100, 120 };
// Create dummy variables
dv_bp = dummy(bp, v);
Note that dv_bp is the same as dv_bp_classes from the first example:
1 0 0
0 0 1
1 0 0
0 0 1
0 1 0
0 0 1
Example: Create dummy variables from continuous BP data and drop first column¶
The dummydn() variable can be used to generate dummy variables from the ranges of
original BP data.
// Create a column of blood pressure data
bp = { 91,
121,
99,
135,
110,
155 };
// Create breakpoints
v = { 100, 120 };
// Create dummy variables
dv_bp_drop = dummydn(bp, v, 1);
Now the dv_bp_drop matrix is the same as the second and third columns of dv_bp and dv_bp_classes:
0 0
0 1
0 0
0 1
1 0
0 1