Q1: What is Conditional Probability?
Answer: Conditional Probability is the probability of an event occurring given that another event has already occurred. Formula: P(A|B) = P(A ∩ B) / P(B). Read as "probability of A given B." Example: P(High Salary | IT Department) — what is the probability of earning high salary given you are in IT? It narrows the sample space to only the condition event. Conditional probability is the foundation of Bayes' Theorem and is used extensively in machine learning classification, risk assessment, and medical diagnosis.
🎯 Explain: Conditional Probability = ek cheez ho chuki hai, ab dusri ki chance kya hai. P(A|B) = P(A and B) / P(B). Example: 8 employees mein 2 IT mein hain. IT mein salary > 60K wale = 0. P(Salary>60K | IT) = 0/2 = 0. Matlab IT department mein koi 60K se upar nahi kamata (sample mein). Real-world: P(Buy | Visited Website) — website visit kiya toh buy ki chance. Spam filter: P(Spam | certain words). Interview mein "Conditional probability narrows the sample space — I use it for conversion rate analysis and risk scoring."

# Conditional Probability
salaries = [55000,72000,65000,58000,80000,48000,70000,62000]
depts = ["IT","HR","Finance","IT","Marketing","HR","Finance","Marketing"]

# P(Salary > 65K | Finance)
finance_sal = [s for s,d in zip(salaries,depts) if d=="Finance"]
high_in_finance = sum(1 for s in finance_sal if s > 65000)
p_cond = high_in_finance / len(finance_sal)

print(f"Finance salaries: {finance_sal}")   # [65000, 70000]
print(f"P(Salary>65K | Finance) = {p_cond:.2f}")  # 0.50

Q2: What is Bayes' Theorem?
Answer: Bayes' Theorem calculates the reverse conditional probability — updating the probability of a hypothesis based on new evidence. Formula: P(A|B) = [P(B|A) × P(A)] / P(B). Components: P(A) = Prior probability (initial belief). P(B|A) = Likelihood (probability of evidence given hypothesis). P(A|B) = Posterior probability (updated belief after evidence). P(B) = Marginal probability (total probability of evidence). Used in spam filters, medical diagnosis, recommendation systems, and Naive Bayes classifiers.
🎯 Explain: Bayes = naye evidence ke baad belief update karo. Example: Email spam detection. P(Spam) = 30% (prior — 30% emails spam hain). P("Free" | Spam) = 80% (likelihood — spam mein "Free" word 80% aata hai). P("Free") = 40% (overall "Free" word kitni emails mein aata hai). P(Spam | "Free") = (0.80 × 0.30) / 0.40 = 0.60 — "Free" word hai toh 60% chance spam hai. Prior → Evidence → Posterior — yeh update process hai. Interview mein "Bayes updates probability with new evidence — I explain it with the spam filter example."

# Bayes' Theorem — Spam Filter Example
p_spam = 0.30              # Prior: P(Spam)
p_free_given_spam = 0.80   # Likelihood: P("Free" | Spam)
p_free_given_ham = 0.10    # P("Free" | Not Spam)
p_ham = 1 - p_spam          # P(Not Spam) = 0.70

# P("Free") = P("Free"|Spam)×P(Spam) + P("Free"|Ham)×P(Ham)
p_free = p_free_given_spam * p_spam + p_free_given_ham * p_ham

# Posterior: P(Spam | "Free")
p_spam_given_free = (p_free_given_spam * p_spam) / p_free
print(f"P(Spam | 'Free') = {p_spam_given_free:.2f}")
# P(Spam | 'Free') = 0.77 — 77% chance it's spam!

Q3: What is the difference between Independent and Dependent Events?
Answer: Independent Events — the occurrence of one event does not affect the probability of the other. P(A and B) = P(A) × P(B). Example: rolling two dice — result of first die does not affect the second. Dependent Events — the occurrence of one event changes the probability of the other. P(A and B) = P(A) × P(B|A). Example: drawing two cards without replacement — first draw changes what's available for the second. Testing independence: if P(A|B) = P(A), events are independent.
🎯 Explain: Independent = ek event dusre ko affect nahi karta. Coin toss 2 baar — pehla head ya tail ho, dusre pe effect nahi. P(Head AND Head) = 0.5 × 0.5 = 0.25. Dependent = ek event dusre ko affect karta hai. Bag mein 3 red, 2 blue balls. Pehli red nikali (bina replacement) → ab 2 red, 2 blue — probabilities change ho gayi. Testing: P(A|B) = P(A) → independent. Interview mein "Independent events have no influence on each other — I verify with the condition P(A|B) = P(A)."

Q4: What is the difference between Mutually Exclusive and Independent Events?
Answer: Mutually Exclusive events cannot occur simultaneously — P(A and B) = 0. Example: rolling a die — getting 3 and 5 on the same roll is impossible. If one occurs, the other cannot. P(A or B) = P(A) + P(B). Independent events can occur simultaneously but don't influence each other — P(A and B) = P(A) × P(B). Key insight: mutually exclusive events are always dependent (if A happens, B definitely doesn't), so they cannot be independent (unless one has probability 0).
🎯 Explain: Mutually Exclusive = dono saath mein nahi ho sakte. Ek coin mein Head aur Tail saath nahi — P(H and T) = 0. Independent = dono saath ho sakte hain but ek dusre ko affect nahi karte. Dice 1 pe 3 aur Dice 2 pe 5 — dono saath, but independent. Common confusion: students sochte hain mutually exclusive = independent — GALAT! Mutually exclusive actually dependent hain — A hua toh B definitely nahi hoga. Interview mein "Mutually exclusive means they can't co-occur (P(A∩B)=0), Independent means they don't influence each other — they're actually opposite concepts."

📊 Comparison:

Q5: What is the Law of Total Probability?
Answer: The Law of Total Probability states that the total probability of an event can be calculated by summing probabilities across all mutually exclusive conditions. Formula: P(B) = Σ P(B|Aᵢ) × P(Aᵢ). Example: P(High Salary) = P(High|IT)×P(IT) + P(High|HR)×P(HR) + P(High|Finance)×P(Finance) + P(High|Marketing)×P(Marketing). It breaks a complex probability into simpler conditional parts. Used as the denominator in Bayes' Theorem.
🎯 Explain: Law of Total Probability = total probability ko parts mein todo. P(High Salary) jaanne ke liye har department mein separately check karo aur combine karo. P(High) = P(High|IT)×P(IT) + P(High|HR)×P(HR) + ... Bayes' Theorem ke denominator mein yahi use hota hai — P(B) calculate karne ke liye. Interview mein "I use the Law of Total Probability to calculate marginal probabilities by summing over all conditions — it's the denominator in Bayes' Theorem."

Q6: What are Permutations and Combinations?
Answer: Permutations count the number of ways to arrange items where order matters. Formula: P(n,r) = n! / (n-r)!. Example: arranging 3 people in a line from 5 = 5!/(5-3)! = 60 ways. Combinations count the number of ways to select items where order does not matter. Formula: C(n,r) = n! / [r! × (n-r)!]. Example: selecting a committee of 3 from 5 = 5!/[3!×2!] = 10 ways. Key: if order matters → Permutation, if order doesn't matter → Combination.
🎯 Explain: Permutation = order matters — ABC aur BAC different arrangements hain. Lock code: 3 digits from 0-9 — order matters. Combination = order doesn't matter — Team of 3 from 5 — {A,B,C} aur {B,A,C} same team hai. Simple rule: "Arrangement" → Permutation. "Selection/Group" → Combination. Permutation hamesha zyada hota hai Combination se (same n,r ke liye). Interview mein "Permutations for ordered arrangements like rankings, Combinations for unordered selections like team formation."

from math import factorial, perm, comb

# Permutation — order matters
print(f"P(5,3) = {perm(5,3)}")  # 60 ways to arrange

# Combination — order doesn't matter
print(f"C(5,3) = {comb(5,3)}")  # 10 ways to select

# Real example: Select 3 employees from 8 for a project
print(f"Ways to select 3 from 8: {comb(8,3)}")  # 56

Q7: What is a Probability Distribution?
Answer: A Probability Distribution describes all possible values a random variable can take and their associated probabilities. Two types: (1) Discrete — countable values with individual probabilities (PMF — Probability Mass Function). Examples: Binomial, Poisson, Bernoulli. (2) Continuous — infinite values within a range with probability density (PDF — Probability Density Function). Examples: Normal, Exponential, Uniform. Total probability always sums/integrates to 1. Understanding distributions is essential for choosing appropriate statistical tests.
🎯 Explain: Probability Distribution = har possible outcome ki probability kya hai. Discrete: dice roll — 1,2,3,4,5,6 har ek ki probability 1/6. Continuous: salary — koi bhi value le sakta hai — probability density use hoti hai. Distribution jaanoge toh correct test choose kar sakte ho. Normal data → z-test/t-test. Count data → Poisson. Yes/No data → Binomial. Interview mein "Distribution determines which statistical test to use — Normal for continuous symmetric, Binomial for binary outcomes, Poisson for count events."

Q8: What is Binomial Distribution?
Answer: Binomial Distribution models the number of successes in a fixed number of independent trials with only two outcomes (success/failure). Parameters: n (number of trials) and p (probability of success). Formula: P(X=k) = C(n,k) × p^k × (1-p)^(n-k). Mean = np, Variance = np(1-p). Example: flip a coin 10 times — probability of getting exactly 7 heads. Requirements: fixed n, independent trials, constant p, binary outcome. Used for A/B testing, conversion rates, quality control.
🎯 Explain: Binomial = fixed trials, 2 outcomes (success/failure), probability constant. 10 customers visit website, conversion rate 20% — kitne buy karenge? n=10, p=0.2. P(exactly 3 buy) = C(10,3) × 0.2³ × 0.8⁷. Mean = np = 10×0.2 = 2 expected purchases. A/B testing mein: 1000 users, 5% conversion — binomial distribution se expected conversions calculate karo. Interview mein "Binomial for binary outcomes with fixed trials — I use it for conversion rate analysis and A/B test sample size calculation."

from scipy import stats

# Binomial: 10 customers, 20% conversion, P(exactly 3 buy)
n, p = 10, 0.20
print(f"P(X=3) = {stats.binom.pmf(3, n, p):.4f}")   # 0.2013
print(f"P(X≤3) = {stats.binom.cdf(3, n, p):.4f}")   # 0.8791
print(f"Expected: {n*p}, Variance: {n*p*(1-p):.1f}")  # 2.0, 1.6

Q9: What is Poisson Distribution?
Answer: Poisson Distribution models the number of events occurring in a fixed interval of time or space when events happen independently at a constant average rate. Parameter: λ (lambda) = average number of events per interval. Formula: P(X=k) = (λ^k × e^(-λ)) / k!. Mean = λ, Variance = λ (mean equals variance — key property). Examples: number of customer calls per hour, website visits per minute, defects per batch, emails per day. Used when counting rare events in a fixed period.
🎯 Explain: Poisson = fixed time/space mein events count karo. Average 5 calls per hour aate hain (λ=5). P(exactly 8 calls) = Poisson formula se calculate. Key property: Mean = Variance = λ. Agar data mein mean ≈ variance ho toh Poisson fit hoga. Website traffic: average 200 visits/hour. Customer support: average 10 tickets/day. Defect analysis: average 2 defects per 100 items. Interview mein "Poisson for count events in fixed intervals — I verify by checking if mean ≈ variance."

# Poisson: Average 5 calls/hour, P(exactly 8 calls)
lam = 5
print(f"P(X=8) = {stats.poisson.pmf(8, lam):.4f}")   # 0.0653
print(f"P(X≤3) = {stats.poisson.cdf(3, lam):.4f}")   # 0.2650
print(f"Mean = Variance = {lam}")

Q10: What is Uniform Distribution?
Answer: Uniform Distribution assigns equal probability to all outcomes within a range. Discrete Uniform: rolling a fair die — each outcome (1-6) has equal probability 1/6. Continuous Uniform: random number between 0 and 1 — any value equally likely. Parameters: a (minimum) and b (maximum). Mean = (a+b)/2. Variance = (b-a)²/12. Used for random sampling, simulation, and as a baseline distribution. Least informative distribution — maximum uncertainty.
🎯 Explain: Uniform = sab outcomes ki equal chance. Fair dice: 1,2,3,4,5,6 — sab ki probability 1/6 — discrete uniform. Random number 0 se 100 ke beech — har value equally likely — continuous uniform. Flat distribution — koi peak nahi. Random sampling mein use hota hai — random number generate karo, employee select karo. Histogram flat dikhega. Interview mein "Uniform is the baseline — equal probability for all outcomes. I use it for random sampling and as a comparison reference."

Q11: What is the Standard Normal Distribution (Z-Distribution)?
Answer: Standard Normal Distribution is a special case of normal distribution with Mean = 0 and Standard Deviation = 1. Any normal distribution can be converted to standard normal using Z-transformation: Z = (X - μ) / σ. The Z-table gives cumulative probabilities for Z values. Key Z-values: Z=1.96 → 97.5% (used for 95% CI), Z=2.576 → 99.5% (used for 99% CI). Standard Normal enables comparing values from different distributions on a common scale and is the basis for Z-tests.
🎯 Explain: Standard Normal = mean 0, SD 1 wali normal distribution. Koi bhi normal distribution ko Z-score se convert kar sakte ho. Salary distribution (mean 63750, SD 9910) → Z = (X-63750)/9910 → Standard Normal mein aa jayegi. Z-table se probability nikal sakte ho — P(Z < 1.96) = 0.975. 95% confidence interval mein ±1.96 use hota hai. Interview mein "Standard Normal is the reference distribution — I convert any normal data using Z-scores to find probabilities from the Z-table."

# Standard Normal — find probability
# P(salary < 70000) with mean=63750, sd=9910
z = (70000 - 63750) / 9910
print(f"Z-score: {z:.2f}")                        # 0.63
print(f"P(Salary < 70K): {stats.norm.cdf(z):.4f}")  # 0.7357 — 73.57%

# Key Z-values
print(f"Z for 95% CI: ±{stats.norm.ppf(0.975):.3f}")  # ±1.960
print(f"Z for 99% CI: ±{stats.norm.ppf(0.995):.3f}")  # ±2.576

Q12: How do you choose the right distribution for your data?
Answer: Distribution selection guide: (1) Binary outcome (yes/no) → Bernoulli (single trial) or Binomial (multiple trials). (2) Count of events in fixed interval → Poisson. (3) Continuous symmetric bell-shape → Normal. (4) Time between events → Exponential. (5) Equal probability outcomes → Uniform. (6) Skewed positive values → Log-Normal. Verification methods: histogram visualization, Q-Q plot (Quantile-Quantile), statistical tests (Shapiro-Wilk for normality, Kolmogorov-Smirnov). Always visualize before assuming.
🎯 Explain: Data dekho aur sochno — kya type ka data hai? Binary (conversion yes/no) → Binomial. Counts (calls per hour) → Poisson (check: mean ≈ variance?). Continuous symmetric → Normal (check: histogram bell-shaped? Shapiro-Wilk test?). Right-skewed → Log-Normal. Steps: (1) Histogram plot karo. (2) Skewness/Kurtosis check karo. (3) Statistical test karo (Shapiro-Wilk for normality). (4) Q-Q plot dekho. Interview mein "I visualize data first with histograms, then verify with Shapiro-Wilk test for normality — distribution determines which statistical test is appropriate."

📊 Distribution Selection Guide:

Q13: What are the different Sampling Methods?
Answer: Probability Sampling (every member has a known chance): (1) Simple Random — every member equally likely (lottery). (2) Stratified — population divided into groups (strata), random sample from each. (3) Cluster — population divided into clusters, entire clusters randomly selected. (4) Systematic — every kth member selected (every 10th person). Non-Probability Sampling: (5) Convenience — whoever is available. (6) Judgmental — researcher selects based on expertise. (7) Snowball — participants recruit others. Probability sampling reduces bias.
🎯 Explain: Sampling = population se representative subset lena. Simple Random = lottery system — sab ki equal chance. Stratified = groups banao (departments), har group se proportional sample — ensures representation. Cluster = geographical clusters (cities) randomly select karo. Systematic = list se har 10th naam pick karo. Convenience = jo available hai use karo — biased but quick. Interview mein "I prefer stratified sampling when I need proportional representation across groups — it ensures each department or region is fairly represented in the sample."

📊 Sampling Methods:

Q14: What is the Central Limit Theorem (CLT)?
Answer: The Central Limit Theorem states that the distribution of sample means approaches a normal distribution as the sample size increases — regardless of the original population's distribution. Key conditions: sample size n ≥ 30 (rule of thumb), independent observations. The mean of sample means equals the population mean (μ). The standard deviation of sample means (Standard Error) = σ/√n. CLT is why we can use normal distribution-based tests (z-test, t-test) even when the population isn't normal — the foundation of inferential statistics.
🎯 Explain: CLT = statistics ka sabse powerful theorem. Population chahe kisi bhi shape ka ho — uniform, skewed, bimodal — agar 30+ samples lo aur unke means plot karo, toh bell curve ban jayega! Isliye z-test aur t-test kaam karte hain — kyunki sample means normally distributed hote hain. Sample size badhao → SE decrease → estimate zyada precise. Interview mein "CLT is why inferential statistics works — regardless of population shape, sample means are normally distributed for n≥30, enabling z-tests and confidence intervals."

# CLT Demonstration
import numpy as np

# Skewed population (exponential)
population = np.random.exponential(scale=50000, size=100000)

# Take 1000 samples of size 30, calculate means
sample_means = [np.random.choice(population, 30).mean() for _ in range(1000)]

print(f"Population mean: {population.mean():,.0f}")
print(f"Mean of sample means: {np.mean(sample_means):,.0f}")
print(f"Population is SKEWED but sample means are NORMAL!")

Q15: What is a Confidence Interval?
Answer: A Confidence Interval (CI) provides a range of values that likely contains the true population parameter with a specified level of confidence. Formula (for mean): CI = x̄ ± Z × (σ/√n). 95% CI means: if we repeated the sampling 100 times, approximately 95 of those intervals would contain the true population mean. Common levels: 90% (Z=1.645), 95% (Z=1.96), 99% (Z=2.576). Wider interval = more confidence but less precision. Larger sample = narrower interval = more precise.
🎯 Explain: Confidence Interval = "true value is range ke andar hai with X% confidence." 95% CI matlab: 100 baar sample lo toh 95 baar true mean is range mein hoga. CI = Mean ± Margin of Error. Margin of Error = Z × SE. Sample salary mean ₹63,750, SD ₹9,910, n=8. SE = 9910/√8 = 3505. 95% CI = 63750 ± 1.96×3505 = 63750 ± 6870 = [₹56,880, ₹70,620]. "True population mean 95% confidence ke saath ₹56,880 - ₹70,620 ke beech hai." Interview mein "CI gives a range with confidence — wider for more confidence, narrower for larger samples."

# 95% Confidence Interval
import numpy as np
from scipy import stats

salaries = [55000,72000,65000,58000,80000,48000,70000,62000]
n = len(salaries)
mean = np.mean(salaries)
se = np.std(salaries, ddof=1) / np.sqrt(n)

# Using t-distribution (small sample)
ci = stats.t.interval(confidence=0.95, df=n-1, loc=mean, scale=se)
print(f"95% CI: ₹{ci[0]:,.0f} to ₹{ci[1]:,.0f}")

# Manual calculation
t_critical = stats.t.ppf(0.975, df=n-1)
moe = t_critical * se
print(f"Margin of Error: ₹{moe:,.0f}")
print(f"CI: ₹{mean-moe:,.0f} to ₹{mean+moe:,.0f}")

Q16: What is Hypothesis Testing?
Answer: Hypothesis Testing is a statistical method to make decisions about population parameters based on sample data. It starts with two competing statements: Null Hypothesis (H₀) — the default assumption, no effect or no difference. Alternative Hypothesis (H₁ or Hₐ) — the research claim, there IS an effect or difference. The process: collect sample data, calculate test statistic, find p-value, compare with significance level (α), make decision — reject or fail to reject H₀. It answers "Is this result statistically significant or just due to chance?"
🎯 Explain: Hypothesis Testing = data se decision lo — result real hai ya coincidence? H₀ = "koi difference nahi hai" (boring default). H₁ = "difference hai" (exciting claim). Example: H₀: New website design ka conversion rate purane se same hai. H₁: New design ka conversion rate better hai. Data collect karo → test karo → p-value dekho → decision lo. P-value chhota (<0.05) → H₀ reject → difference significant hai! Interview mein "Hypothesis testing determines if observed differences are statistically real or just random noise."

Q17: What is the Null Hypothesis and Alternative Hypothesis?
Answer: Null Hypothesis (H₀) represents the status quo — assumes no effect, no difference, no relationship. It is the hypothesis we try to disprove. Alternative Hypothesis (H₁) represents the research claim — there IS an effect, difference, or relationship. Three forms: Two-tailed (H₁: μ ≠ value — different in any direction), Right-tailed (H₁: μ > value — greater than), Left-tailed (H₁: μ < value — less than). We never "accept" H₀ — we either reject it or fail to reject it.
🎯 Explain: H₀ = status quo — "kuch nahi badla". H₁ = claim — "kuch badla hai". Drug testing: H₀ = drug ka koi effect nahi. H₁ = drug effective hai. Two-tailed: H₁: μ ≠ 60000 — salary 60K se DIFFERENT hai (upar ya neeche). One-tailed: H₁: μ > 60000 — salary 60K se ZYADA hai. Important: H₀ "accept" nahi karte — "fail to reject" bolte hain — kyunki evidence nahi mili reject karne ki, but proof nahi ki H₀ sahi hai. Interview mein "We fail to reject H₀ — we don't prove it true, we just don't have enough evidence against it."

Q18: What is a P-value?
Answer: P-value is the probability of observing results as extreme as (or more extreme than) the sample results, assuming the null hypothesis is true. Small p-value (<α) means the observed data is unlikely under H₀ — strong evidence against H₀ — reject H₀. Large p-value (≥α) means the data is consistent with H₀ — insufficient evidence — fail to reject H₀. Common α levels: 0.05 (5%), 0.01 (1%), 0.10 (10%). P-value does NOT tell the probability of H₀ being true — it measures evidence strength.
🎯 Explain: P-value = "agar H₀ sahi hota, toh yeh result aane ki chance kitni thi?" P-value = 0.03 matlab: agar really koi difference nahi hai (H₀ true), toh aisa extreme result sirf 3% chance se aata — bahut unlikely — toh H₀ reject karo. P-value = 0.45 matlab: 45% chance — common hai — H₀ reject nahi kar sakte. Rule: p < 0.05 → significant (reject H₀). p ≥ 0.05 → not significant (fail to reject). Common misconception: p-value H₀ true hone ki probability NAHI hai! Interview mein "P-value measures evidence strength — smaller means stronger evidence against H₀."

📊 P-value Decision Guide:

Q19: What is the difference between Type I and Type II errors?
Answer: Type I Error (False Positive, α) — rejecting H₀ when it is actually true. "Finding a difference that doesn't exist." Probability = α (significance level, typically 0.05). Type II Error (False Negative, β) — failing to reject H₀ when it is actually false. "Missing a real difference." Probability = β. Power = 1-β (probability of correctly detecting a true effect). Trade-off: decreasing α increases β. Increasing sample size reduces both errors. In medical testing: Type I = diagnosing healthy person as sick, Type II = missing a real disease.
🎯 Explain: Type I = false alarm — doctor ne bola bimaar ho but healthy ho. H₀ sahi tha but reject kar diya. Probability = α = 0.05 (5% chance). Type II = missed detection — doctor ne bola healthy but actually bimaar ho. H₀ galat tha but reject nahi kiya. Probability = β. Power = 1-β — real effect detect karne ki ability. Trade-off: α kam karo (strict) → β badh jaayega (real effects miss karoge). Solution: sample size badhao — dono errors kam. Interview mein "Type I is a false positive, Type II is a false negative — I increase sample size to minimize both."

📊 Error Types:

Q20: What is the difference between Z-test and T-test?
Answer: Z-test is used when: population standard deviation (σ) is known AND sample size is large (n ≥ 30). Uses the Z (standard normal) distribution. T-test is used when: population σ is unknown (estimated from sample) OR sample size is small (n < 30). Uses the t-distribution which has heavier tails than normal — accounts for extra uncertainty. As sample size increases, t-distribution approaches Z-distribution. In practice, t-test is used much more often because population σ is rarely known.
🎯 Explain: Z-test = population SD pata hai + large sample. T-test = population SD nahi pata + small sample. Real-world mein population SD rarely pata hota hai — isliye T-test zyada common hai. T-distribution wider hoti hai (heavier tails) — extra uncertainty accommodate karti hai. n badhao toh t-distribution normal ke karib aa jaati hai. Our salary data: n=8 (small), population SD unknown → T-test use karo. Interview mein "I almost always use t-test because population standard deviation is rarely known — t-test accounts for the additional uncertainty."

from scipy import stats
import numpy as np

salaries = [55000,72000,65000,58000,80000,48000,70000,62000]

# One-sample T-test: Is mean salary = 60000?
# H₀: μ = 60000, H₁: μ ≠ 60000
t_stat, p_value = stats.ttest_1samp(salaries, 60000)
print(f"T-statistic: {t_stat:.3f}")
print(f"P-value: {p_value:.4f}")
print(f"Decision: {'Reject H₀' if p_value < 0.05 else 'Fail to Reject H₀'}")

Q21: What are the types of T-tests?
Answer: Three types: (1) One-sample T-test — compares sample mean to a known/hypothesized value. "Is our average salary ₹60,000?" (2) Two-sample Independent T-test — compares means of two independent groups. "Do IT and HR departments have different average salaries?" (3) Paired T-test — compares means of the same group at two different times. "Did training improve employee performance?" Assumptions: normal distribution (or n≥30), independent observations, equal variances (for independent t-test — use Welch's if unequal).
🎯 Explain: One-sample: sample mean vs ek fixed value. "Kya average salary 60K hai?" Two-sample independent: do groups compare karo. "IT vs HR salary different hai?" Paired: same group, 2 measurements. "Training se pehle aur baad mein performance badli?" Welch's t-test: jab variances unequal hon — safer version — scipy mein equal_var=False. Interview mein "One-sample for testing against a benchmark, Independent for comparing two groups, Paired for before-after studies."

# Two-sample Independent T-test
# H₀: IT salary = Finance salary
it_sal = [55000, 58000]
fin_sal = [65000, 70000]

t, p = stats.ttest_ind(it_sal, fin_sal, equal_var=False)  # Welch's
print(f"T-stat: {t:.3f}, P-value: {p:.4f}")

# Paired T-test (before/after)
before = [55, 60, 45, 70, 65]
after = [62, 68, 50, 75, 72]

t, p = stats.ttest_rel(before, after)
print(f"Paired T-stat: {t:.3f}, P-value: {p:.4f}")
print(f"Training {'improved' if p < 0.05 else 'did not improve'} performance")

Q22: What is the Significance Level (α)?
Answer: Significance Level (α) is the threshold probability for rejecting the null hypothesis — the maximum acceptable probability of making a Type I error (false positive). Common values: α = 0.05 (5% — standard), α = 0.01 (1% — strict, medical/safety), α = 0.10 (10% — exploratory). If p-value < α → reject H₀ (significant). If p-value ≥ α → fail to reject H₀ (not significant). α must be set BEFORE conducting the test — setting it after seeing results is p-hacking (bad practice).
🎯 Explain: α = kitna risk accept karo false positive ka. α = 0.05 → 5% chance galat reject karna acceptable hai. Strict fields (medical) mein α = 0.01 — galat diagnosis bahut costly. Exploratory analysis mein α = 0.10 — thoda lenient. Important: α pehle decide karo, phir test karo — baad mein change karna cheating (p-hacking). Interview mein "I set α = 0.05 as standard, 0.01 for high-stakes decisions, and always decide before running the test to avoid p-hacking."

Q23: What are One-tailed and Two-tailed tests?
Answer: Two-tailed test checks for difference in either direction — H₁: μ ≠ value. Significance split between both tails (2.5% each for α=0.05). Used when you want to detect any difference regardless of direction. One-tailed test checks for difference in a specific direction — H₁: μ > value (right-tailed) or H₁: μ < value (left-tailed). All significance in one tail (5% for α=0.05). More powerful for detecting directional effects but misses effects in the opposite direction.
🎯 Explain: Two-tailed: "salary 60K se different hai?" — zyada ya kam dono check karo. α=0.05 dono sides pe 2.5%. One-tailed: "salary 60K se ZYADA hai?" — sirf ek direction check. α=0.05 ek side pe 5% — zyada power. Two-tailed safe hai — dono directions cover. One-tailed powerful but risky — agar opposite direction mein result ho toh miss ho jayega. Default: two-tailed use karo jab tak specific direction ka strong reason na ho. Interview mein "I default to two-tailed unless I have a strong prior reason for a directional hypothesis."

Q24: What are the steps of Hypothesis Testing?
Answer: Complete hypothesis testing procedure: (1) State hypotheses — H₀ and H₁. (2) Choose significance level — α (usually 0.05). (3) Choose the right test — z-test, t-test, chi-square based on data type and conditions. (4) Calculate test statistic from sample data. (5) Find p-value. (6) Compare p-value with α. (7) Make decision — reject or fail to reject H₀. (8) Interpret in business context — "There is/isn't sufficient evidence that..." Always report confidence interval alongside p-value for practical significance.
🎯 Explain: 8 steps yaad rakho: Hypotheses → α → Test choose → Calculate → P-value → Compare → Decide → Interpret. Example: (1) H₀: avg salary = 60K, H₁: ≠ 60K. (2) α = 0.05. (3) One-sample t-test (small n, unknown σ). (4) t = 1.07. (5) p = 0.32. (6) 0.32 > 0.05. (7) Fail to reject H₀. (8) "Insufficient evidence salary differs from ₹60,000." Always business interpretation do — not just "reject/fail." Interview mein poora workflow confidently batao — structured approach dikhata hai.

# Complete Hypothesis Testing — Step by Step

# Step 1: Hypotheses
# H₀: μ = 60000 (avg salary is 60K)
# H₁: μ ≠ 60000 (avg salary is NOT 60K)

# Step 2: α = 0.05
alpha = 0.05

# Step 3: One-sample t-test (small n, unknown σ)
# Step 4-5: Calculate test statistic and p-value
t_stat, p_value = stats.ttest_1samp(salaries, 60000)

# Step 6-7: Decision
print(f"T-statistic: {t_stat:.3f}")
print(f"P-value: {p_value:.4f}")
print(f"α = {alpha}")

if p_value < alpha:
    print("Decision: REJECT H₀")
    print("Conclusion: Salary significantly differs from ₹60,000")
else:
    print("Decision: FAIL TO REJECT H₀")
    print("Conclusion: Insufficient evidence salary differs from ₹60,000")

# Step 8: Also report CI for practical significance
ci = stats.t.interval(0.95, df=len(salaries)-1, 
                       loc=np.mean(salaries), 
                       scale=stats.sem(salaries))
print(f"95% CI: ₹{ci[0]:,.0f} to ₹{ci[1]:,.0f}")

Q25: What is Covariance?
Answer: Covariance measures the direction of the linear relationship between two variables — how they change together. Positive covariance: both increase together. Negative covariance: one increases, other decreases. Zero covariance: no linear relationship. Formula: Cov(X,Y) = Σ[(Xᵢ-X̄)(Yᵢ-Ȳ)] / (n-1). Limitation: covariance is scale-dependent — its magnitude depends on the units of measurement, making it hard to compare across different variable pairs. This is why correlation is preferred.
🎯 Explain: Covariance = do variables saath mein kaise move karti hain. Positive: Salary badhti hai toh Sales bhi badhti hai. Negative: Price badhta hai toh Demand girti hai. Problem: Salary-Sales ka covariance crores mein hoga, Height-Weight ka thousands mein — compare nahi kar sakte. Isliye Correlation use karte hain — normalized version. Interview mein "Covariance tells direction of relationship but its magnitude is scale-dependent — I use correlation for standardized comparison."

salaries = [55000,72000,65000,58000,80000,48000,70000,62000]
sales = [85000,92000,45000,78000,65000,52000,88000,71000]

# Covariance
cov = np.cov(salaries, sales)
print(f"Cov(Salary, Sales) = {cov[0][1]:,.0f}")
# Positive → both tend to increase together

Q26: What is Correlation Coefficient (Pearson's r)?
Answer: Pearson's Correlation Coefficient (r) measures the strength and direction of the linear relationship between two continuous variables. Formula: r = Cov(X,Y) / (SD_X × SD_Y). Range: -1 to +1. r = +1: perfect positive linear relationship. r = -1: perfect negative. r = 0: no linear relationship. Interpretation: |r| < 0.3 weak, 0.3-0.7 moderate, > 0.7 strong. Pearson's r only measures LINEAR relationships — a strong non-linear pattern can have r ≈ 0. Always visualize with scatter plot before interpreting r.
🎯 Explain: Pearson r = standardized covariance — -1 se +1 ke beech. r = 0.85 → strong positive — salary badhti hai toh sales bhi. r = -0.70 → strong negative — price badhta hai toh demand girti hai. r = 0.10 → weak — almost no linear relationship. Warning: r sirf LINEAR relationship measure karta hai — U-shaped pattern ho toh r ≈ 0 dikhayega even though strong pattern hai. Hamesha scatter plot dekho pehle. Interview mein "I always pair correlation coefficient with scatter plot visualization — r can miss non-linear patterns."

# Pearson Correlation
r, p_val = stats.pearsonr(salaries, sales)
print(f"Pearson r: {r:.3f}")
print(f"P-value: {p_val:.4f}")

# Interpretation
if abs(r) > 0.7: strength = "Strong"
elif abs(r) > 0.3: strength = "Moderate"
else: strength = "Weak"

direction = "Positive" if r > 0 else "Negative"
print(f"{strength} {direction} correlation")

# Correlation matrix with Pandas
import pandas as pd
df = pd.DataFrame({'Salary':salaries, 'Sales':sales})
print(df.corr())

Q27: What is the difference between Pearson and Spearman Correlation?
Answer: Pearson measures linear relationship between continuous variables — assumes normal distribution and linear pattern. Spearman measures monotonic relationship using ranked data — does not assume normality, works with ordinal data, handles non-linear monotonic patterns. Spearman converts values to ranks first, then calculates Pearson on ranks. Use Pearson when: both variables continuous, roughly normal, linear relationship expected. Use Spearman when: ordinal data, non-normal distribution, monotonic (not necessarily linear) relationship, or outliers present.
🎯 Explain: Pearson = linear relationship — straight line fit. Spearman = monotonic relationship — consistently increasing/decreasing but straight line nahi zaroor. Spearman ranks pe kaam karta hai — outlier resistant. Rating data (1-5) ke liye Spearman better — ordinal hai. Salary data (continuous, normal) ke liye Pearson better. Scatter plot mein: straight line pattern → Pearson. Curved but consistently increasing → Spearman. Interview mein "Pearson for linear continuous data, Spearman for ordinal data or when I suspect non-linear monotonic relationships."

# Pearson vs Spearman
pearson_r, p1 = stats.pearsonr(salaries, sales)
spearman_r, p2 = stats.spearmanr(salaries, sales)

print(f"Pearson r:  {pearson_r:.3f} (linear)")
print(f"Spearman r: {spearman_r:.3f} (monotonic)")

# For ordinal data (ratings), use Spearman
ratings = [4,5,3,4,4,3,5,4]
r, p = stats.spearmanr(salaries, ratings)
print(f"Salary vs Rating (Spearman): {r:.3f}")

Q28: What is Chi-Square Test?
Answer: Chi-Square (χ²) Test examines the relationship between categorical variables. Two types: (1) Chi-Square Test of Independence — tests if two categorical variables are related. Example: Is there a relationship between Department and Performance Rating? (2) Chi-Square Goodness of Fit — tests if observed frequencies match expected frequencies. Example: Is customer distribution across regions uniform? Uses a contingency table (cross-tabulation). Large χ² value → variables are related (small p-value). Assumption: expected frequency in each cell ≥ 5.
🎯 Explain: Chi-Square = categorical variables ke beech relationship test karo. Department aur Rating related hain? Contingency table banao (Department × Rating), expected frequencies calculate karo, observed vs expected compare karo. χ² bada = difference bada = related hain. P-value < 0.05 → significant relationship. Goodness of Fit: "Kya customers equally distributed hain regions mein?" Expected: 25% each. Observed: 30%, 20%, 35%, 15%. χ² test batayega ki deviation significant hai ya random. Interview mein "Chi-Square for categorical associations — I use it for A/B testing on conversion categories."

# Chi-Square Test of Independence
# Is Department related to Rating category?

data = pd.DataFrame({
    'Dept': ['IT','HR','Finance','IT','Marketing','HR','Finance','Marketing'],
    'RatingCat': ['Good','Excellent','Average','Good','Good','Average','Excellent','Good']
})

# Contingency table
table = pd.crosstab(data['Dept'], data['RatingCat'])
print(table)

# Chi-Square test
chi2, p, dof, expected = stats.chi2_contingency(table)
print(f"\nChi² = {chi2:.3f}, P-value = {p:.4f}, df = {dof}")
print(f"Dept and Rating are {'related' if p < 0.05 else 'independent'}")

Q29: What is the difference between Statistical Significance and Practical Significance?
Answer: Statistical Significance means the result is unlikely due to chance (p-value < α). Practical Significance means the result is large enough to matter in the real world. A large sample size can make trivially small differences statistically significant — ₹100 salary difference can be significant with n=100,000 but is meaningless practically. Always report effect size (Cohen's d) alongside p-value. Small effect: d=0.2, Medium: d=0.5, Large: d=0.8. Confidence intervals also show practical significance — the range tells you the plausible effect size.
🎯 Explain: Statistical significance ≠ important. 1 lakh employees ka sample lo — ₹100 ka salary difference bhi significant dikhega (p<0.05) — but ₹100 meaningless hai practically. Effect size (Cohen's d) batata hai ki difference kitna BADA hai — d=0.2 chhota, d=0.8 bada. Hamesha p-value + effect size + CI report karo. "Statistically significant with p=0.03 and Cohen's d=0.15" — matlab significant but tiny effect. Interview mein "I always report practical significance alongside statistical — a p-value under 0.05 doesn't mean the effect matters in business context."

# Cohen's d — Effect Size
def cohens_d(group1, group2):
    n1, n2 = len(group1), len(group2)
    var1, var2 = np.var(group1, ddof=1), np.var(group2, ddof=1)
    pooled_std = np.sqrt(((n1-1)*var1 + (n2-1)*var2) / (n1+n2-2))
    return (np.mean(group1) - np.mean(group2)) / pooled_std

it_sal = [55000, 58000]
fin_sal = [65000, 70000]

d = cohens_d(fin_sal, it_sal)
print(f"Cohen's d: {d:.2f}")

if abs(d) < 0.2: print("Negligible effect")
elif abs(d) < 0.5: print("Small effect")
elif abs(d) < 0.8: print("Medium effect")
else: print("Large effect")

Q30: How do you choose the right statistical test?
Answer: Test selection depends on: (1) Data type — continuous vs categorical. (2) Number of groups — one sample, two groups, multiple groups. (3) Relationship type — comparison vs association. (4) Distribution — normal vs non-normal. Guide: One group mean vs value → one-sample t-test. Two independent group means → independent t-test. Same group before/after → paired t-test. 3+ group means → ANOVA. Two categorical variables → Chi-Square. Continuous relationship → Pearson/Spearman correlation. Non-normal data → non-parametric alternatives (Mann-Whitney, Wilcoxon, Kruskal-Wallis).
🎯 Explain: Right test choose karna bahut important hai — galat test = galat conclusion. Decision tree: Comparing means? → How many groups? 1 group → one-sample t. 2 groups → independent t (or paired). 3+ groups → ANOVA. Categorical data? → Chi-Square. Relationship? → Correlation. Non-normal? → Non-parametric version use karo. Interview mein "I follow a decision tree — data type determines the test family, number of groups determines the specific test, and normality determines parametric vs non-parametric."

📊 Statistical Test Selection Guide: