Name: ; UTD-ID: tailed) t.inv(0.

05, df) SST 2 SST= SSR + SSE

CONTINUOUS PROBABILITY DISTRIBUTION 0.05 (one-
tailed)
Right NORM.S.INV(0.95)
t.inv(0.95, df)
+1.6449 ∑ ( yi − y )
Normal Distribution
SSR 2 SSR = SSE - SST
> described by 2 parameters: mean and stdev. Increase in mean
shifts the curve to the right.
0.05 (two-
tailed)
Both NORM.S.INV(0.025)
| t.inv(0.95, df) and
±1.96
∑ ( ^y i − y )
> empirical law for normal distribution. 68.26%, 95.44% 99.72% NORM.S.INV(0.975)
SSE 2 SSE = SST – SSR ors
> excel formulas give solution from the negative infinity.
Skewness and Kurtosis Confidenc Tail(s)
| t.inv(0.975, df)
z-score Result ∑ ( y i −^y i ) (Standard Error)2×(n−k−
1)
> mean = mode = median; skewness=0, symmetric around the e level calculation
mean. 90% Both NORM.S.INV(0.05) ±1.6449
(for one tail, 5%) Regression Statistics
> Positive skewness (>0) indicates a long right tail, and negative
skewness (<0)indicates a long left tail. 95% Both NORM.S.INV(0.025) ±1.96 Item description formula
> Kurtosis measures the "tailedness" or peakedness of a (for one tail, 2.5%)

√ R2
Multiple R Coefficient of
distribution, indicating how much data falls in the tails compared to a 4. Calculate test statistic (Zdata or Tdata): Correlation R|
normal distribution. Situation Formula Reasoning
Formula:
> formula for z score: (𝒙̅ – population mean)/stdev x̅ −μ Population stdev is
known.
R square Coefficient of
Determination SSR
R2 or or
> table: Use the row for the first decimal place (e.g., 2.x) and the (variation in
column for the hundredths (e.g., .04). Gives the area from the
mean up to that z-score.
Zdata = σ
Assume data is normally
distributed or n >= 30 dependent variable
explain by variation
SST
> Both bounds below the mean: Subtract the areas between z-
SSE
√n
Population in independent
scores.
mean (µ) / variable) 1−
> One bound below, one above: Add the areas on either side of
the mean.
Sample mean
(𝒙̅) x̅ −μ Population stdev is
unknown & n<30
SST
area from the mean to 𝑋 from 0.5.
> More than the mean (upper bound is infinity): Subtract the Explained
T dist, df = n-1
>For Cut-off Points Identify the area (probability) first, and then
Tdata = s (sample size is
disregarded for actual
variation/total
variation

+ 𝑍 × stdev
use the z-score to calculate the value of X with the formula: = mean
√n application just Adjusted R CoD adjusted for DF
SSE
𝑧𝑑𝑎𝑡𝑎 =
remember for mcq) Square (R2 adjusted for the
Excel usage: Population proportion is no. preductors
Probability | Area under the curve: NORM.DIST(x, mean, stdev,
^p− p known. n−k−1
TRUE) – {in this case, mean and stdev are known. As is for exact. Population
Proportion ¿
Assume np >= 5 and
1−
SST
√
Less than | less than or norm.dist(x, mean, stdev, n(1-p) >= 5
(𝒑) / Sample
equal to
Greater than | greater than
true)
1-norm.dist(x, mean, stdev, Proportion p ( 1− p )
or equal to true) (𝒑̂) n−1
Atleast =1 - NORM.DIST(x, mean,
stdev, TRUE)
n or
5. Calculate pvalue:
Between NORM.DIST(X2, mean, stdev, Through Z score Through T Test ( 1−R 2 ) ( n−
TRUE) - NORM.DIST(X1,
mean, stdev, TRUE)
Two Tailed 2 * (1 – norm.s.dist) T.DIST.2T(x, 1−
Probability when only z score is known: NORM.S.DIST(z, TRUE)
degree of
freedom, n−k−1
Less than | less than or Norm.s.dist(z,true)

√
true/false) Standard Average distance that
equal to
Greater than | greater than 1-norm.s.dist(z,true)
One Tailed norm.s.dist (less than)
or 1 – norm.s.dist
t.dist(x,df,true) or
1-t.dist(x,df,true)
Error the observed values
fall from the
SSE
or equal to
Atleast 1 - NORM.S.DIST(z, TRUE)
(greater than) regression line
n−k −1
Between NORM.S.DIST(z2, TRUE) - 6. conclude whether to reject or accept the null hypothesis: Observations Number of n
NORM.S.DIST(z1, TRUE) Test Type Alternative Reject Null observations k: Represents the
Cut-Off Point: NORM.INV(probability, mean, stdev) Hypothesis (H1) hypothesis when:
Top x% à cut off point (raw x NORM.INV(1-x, mean, K is 1 in simple linear number of predictors
Left-Tailed H1: μ < μ0 zdata ≤ zstat | tstat ≤ tstat regression (independent
score) stdev) Right-Tailed H1: μ > μ0 zdata ≥ zstat | tdata ≥ tstat
Bottom x% NORM.INV(x, mean, stdev) variables) in the
Two-Tailed H1: μ ≠ μ0 |zdata| ≥ |zstat| same for regression model.
Z-score: NORM.S.INV(probability) ; only when prob is known and z t
score is required.
> Type I Error (False Positive) 𝛼: Rejecting a true null
Type of Errors
If no more than x%, then do norm.inv(x,mean,stdev). > In general the higher the value of R2, the better the model fits the
Exponential Distribution
> Type II Error (False Negative) 𝛽: Failing to reject a false null
hypothesis; probability data.
> smaller value of lambda – flatter the curve. R2 = 1: Perfect match between the line and the data points.
> mean = stdev = 1/λ(the λ is rate constant; lamda): R2 = 0: There are no linear relationship between x and y.
hypothesis;
lambda=1/mean
Lower 𝛼 (reduce Type I) → Higher 𝛽 (increase Type II).
Inverse Relationship: >use R to determine the Quality of Regression and R2 to determine
<=, less than expon.dist(X, LAMDA, FALSE)
Higher 𝛼 (increase Type I) → Lower 𝛽 (reduce Type II).
Percentage explained
BETWEEN =EXPON.DIST(b, lambda, TRUE) -
EXPON.DIST(a, lambda, TRUE) > variance of error is supposed to be constant, violation is called
Mitigation: Increasing sample size reduces both errors.
>=, greater than, at 1 – expon.dist( X EXACT, LAMDA, heteroscedasticity.
Probability of a Type II Error
least TRUE) > relationship between type I and type II: use to calculate and > ε (Greek letter epsilon à random term (error variable): diff
CONFIDENCE INTERVAL deduce. between actual and estimated. If std error sε is small, the fit is good.
Statistical inference: 2 types à (1) Estimation and (2) Hypothesis > steps: If sε is large, the model is poor (compare with sample mean of
Testing. (1) lay down hypothesis. dependent variable)
> Estimation: determine the approximate value of a population (2) calculate the Critical value=μ0+(zα×σ/√n)
parameter on the basis of sample statistics—two types: (1) Point and (3) Type II Error calculation à Zscore :
(2) Interval Estimator.
> confidence interval: 95% | alpha = 100-CL | alpha/2 would be
(100-CL)/2 ; (100-95)/2 = 2.5%
critical value−true mean
> confidence interval = sample mean +- margin of error | sample
stdev population ,
mean: (upper number - lower number)/2
standard error ( ) Four conditions

√n
> probability for z score would be cl + alpha/2 = 97.5% in this For these regression methods to be valid the following four
case. | use norm.s.inv to get z score. conditions for the error variable ε must be met:
> interval width: wide interval provides little information. corresponding pvalue through norm.s.dist is the probability of a Type • The probability distribution of ε is normal.
Formulas II error. • The mean of the distribution is 0; that is, E(ε ) = 0.
Item Population Mean Population Proportion T-test • The standard deviation of ε is σε , which is a constant regardless
> n-1 = degree of freedom defines t distribution. ; increase in of the value of x.
σ √ p−¿^ ¿ ¿
Margin of
error Z* (se) degree of freedom the t distribution approaches the standard normal • The value of ε associated with any particular value of y is
z* (se) distribution. independent of ε associated with any other value of y.

√n required conditions: np
>=10, n(1-p)>=10
> Cut-off weight=μ0+(t-critical×sxˉ)
Distribution
Lower (mean of sample) p-hat - z* σp-hat > T.DIST à left tailed (x, degree of freedom, true/false) ; x is the t-
confidence – Zσx (ME) value (less than or equal to) P(t<=x)
limit > T.DIST.2T à Two tailed (x, degree of freedom, true/false);
extreme values at both tail ends
Upper (mean of sample) p-hat + z* σp-hat > T.DIST.RT -à Right tailed (x, degree of freedom); complement
confidence + Zσx (ME) p-hat = % of success of left tailed (greater than specified value) ; P(t >= x)
Limit Critical T Value
Sample size ( > T.INV à LEFT TAILED ; Find the value of t when the probability is

¿^
given. T.inv(p, df)

Z∗σ p− ¿
> T.INV.2T is used for two tailed T value extraction.
Regression and ANOVA
Margin of Error Theory
)2 ANOVA
Sample Size, Confidence Interval, Confidence level
HYPOTHESIS TESTING Source of sum of degrees mean f-test
validatio squares of square statistic A larger sample size narrows the confidence interval, increasing
> null hypothesis H0 = μ n freedom precision, while a higher confidence level widens the interval to
ensure greater certainty that the interval contains the true
> alternative hypothesis à Treatmen SSTR df1 = k-1 MSTR = Fdata =
parameter.
statement opposite to Null
Hypothesis ( >, <, <>)
t
SSTR MSTR Relation:
> A small p value Larger sample size narrows CI.
(<0.05), reject the H0 .
> A large p value
k−1 MSE Higher confidence level widens CI (if sample size is fixed).
Additions after revisions
Error SSE df2 = n1-k MSE=
(>=0.05), not reject H0.
> p-value measures the
strength of evidence against H0
SSE ANOVA.
> Null hypothesis (H0)= everything is equal. | Alternative
hypothesis (H1) = differ (at least 2 – could be more but difference is
STEPS
1. State Hypothesis and define the type of test (1tailed – 2tailed) n−k the focus)
2. Draw graphs (picture above) Total sst > use F-distribution to test hypothesis. Total area under F curve =
3. Determine critical value à zscore of significance level 1
Alpha Tail(s) z-score Result REGRESSION > F is never negative ; starts at 0 and extends indefinitely to the
Level calculation right but never meets the horizontal axis. It is right skewed.
0.05 (one- Left NORM.S.INV(0.05) -1.6449 Actual Formulaic Relation Different F curve based on the degree of freedom(df).
Formula
> Mean Square Treatment (MSTR or MST): measures the variability
in the sample means.
> Mean Square Error (MSE): measures the variability within the

sample. Fdata=
MSTR . Fdata measures the variability among
MSE
the sample means, compared to the variability within the samples.
>Fdata MST compared to MSE  follows an F distribution with df 1 = k-
1 & df2=nt-k (k: number of groups or treatments being compared
| n: total number of observations across all groups. )
> compare Fdata and Fcriticial to decide whether to reject hypothesis.
Can also use p value to draw the same conclusion.

To calculate:
Pvalue: F.dist.rt(fstat, df-treatement, df-error)
Fcriticial: f.inv.rt(significance alpha, df-treatment, df-error)
> Randomized Block Analysis of Variance: The purpose of
designing a randomized block experiment is to reduce the within-
treatments variation to more easily detect differences between
the treatment means.
> Blocking reduces variation in your results. effects of some outside
variables by bringing those variables into the experiment to form the
blocks. — Separate conclusions can be made from each block,
making for more precise conclusions.