Chapter 19: Interviewing Techniques in Psychology

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement not replaced the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

Welcome back to The Deep Dive.

Today we're taking on something that feels, well, almost foundational.

A tool so common in psychology and all the helping professions that we barely even question it.

We're talking about the psychological interview.

Exactly.

It's the one technique that you'll find absolutely everywhere.

Psychiatry, social work, counseling, sure.

But also HR, market research.

It's just universal.

It is.

And there's a good reason for that.

It satisfies a very human need for that face -to -face personal connection.

Right.

It feels like the most direct way to understand someone.

Way more personal than a questionnaire

It really does.

But that feeling, that intuition, that it's the best way is exactly what we need to put under the microscope.

Clinicians, researchers, we've all regarded it as this vital tool for understanding, for predicting behavior.

But the claims have to be tested.

Precisely.

If we're using the interview to make huge decisions, a diagnosis, who gets a job, then it has to be held to the same scientific standard as any psychological test.

We have to ask, is it reliable?

Is it valid?

So that's our mission for this deep dive.

We want to build a really structured understanding of why this tool we all rely on is, well, often shockingly unreliable.

And then look at how modern research is trying to put it on a more scientific footing.

Before we get there, though, we have to be really clear about what we mean by interview.

It's a huge term.

We have a very broad term.

It is.

And a good way to split it comes from Harry Stack Sullivan's work back in 1954.

He made this great distinction between two major uses.

Okay, what are they?

First, you have the interview used to produce change.

That's a therapeutic interview.

The goal is treatment, helping the person.

That's not what we're talking about today.

Not at all.

We are focusing exclusively on the other type.

The interview as an assessment tool.

Its purpose is to collect information and to form judgments based on that information.

That's the crucial distinction.

An assessment device, not a treatment.

And even within assessment, the context matters immensely.

You're saying a social research interview is a very different beast from a clinical one?

Oh, entirely.

In social research, the interviewer's main job is just data collection.

They gather the facts, the attitudes, and then that data gets passed on.

It's analyzed later, usually statistically.

So there's a separation.

The person gathering the info isn't necessarily the one making the judgment call.

Exactly.

There's a buffer.

But in a clinical setting, that luxury is gone.

The psychologist or psychiatrist has to do both jobs, often at the same time.

They're gathering the data and trying to form a judgment right there in the room.

Yes.

A diagnosis, a treatment plan.

The stakes are incredibly high.

Which means any small error you might find in a social research interview.

It gets magnified, compounded really by that pressure for an immediate high stakes judgment.

And this leads us to what you call the fundamental scientific problem.

It does.

When you measure the interview against the standards we use for, say, psychological tests, the kind set by the American Psychological Association,

it's just notorious for its problem.

The kind of problems.

A lack of reliability, for one.

That's consistency.

And then very poor validity.

Its ability to actually predict things accurately.

So we need to move past this idea of just being a good interviewer and start looking at the actual sources of error.

That's the only way forward.

We have to systematically catalog what goes wrong.

Perfect.

So let's start that catalog.

Section one.

Sources of error in just the data collection phase.

This is before anyone even tries to make a judgment.

And we're starting with the very basics.

The questions themselves.

Right.

And in this context,

error is anything that gives you biased or distorted information.

Anything that pulls you away from the truth.

The first place this happens is in question formulation.

We can split questions into two big camps, right?

Open and closed.

You can.

A closed question is straightforward.

The answer is basically predetermined.

Did you leave school at 14?

Yes.

A no.

Specific age.

What's the upside of that?

The upside is methodological.

It's much easier to code the answers reliably.

They're perfect for a structured interview where you need to quantify things.

But the error.

The error is that they're restrictive.

What if the options don't apply to the person?

Maybe they left at 13 and went back at 16.

A closed question forces them into a box that isn't true and right away you've introduced distortion.

Okay.

So then you have the open question.

What was your life at school like?

Exactly.

No specific answer is expected.

And these are fantastic for exploratory interviews.

Like a first session with a patient.

You're not forcing your own framework onto their story.

But I'm guessing there's a downside there, too.

The vulnerability is omission.

If the question is too broad, the person might just not volunteer the most important information.

You're sort of leaving it to chance that they'll measure the one critical detail you need.

So it's not that one is good and one is bad.

Not at all.

Any real interview has to use both.

The key is for the interviewer to be aware, moment to moment, of the specific kind of error they might be introducing with each choice.

And then we get to the most notorious type of question.

Ah, the leading question.

The classic example being, when did you stop beating your wife?

Right.

The danger is obvious.

It assumes a premise that you were beating your wife.

And that premise might be completely false.

It contaminates everything that comes after.

But here's where it gets tricky.

Our sources suggest they're not always bad.

It's all about context.

Some researchers,

Richardson's group, for instance, suggested that if the premise is actually correct, a leading question can get you more accurate information faster.

It just cuts to the chase.

It does.

And this brings us to the, well, the infamous example of Kinsey and his research on sexuality.

Right.

He deliberately used leading questions for sensitive topics.

Yes.

Instead of asking, have you ever masturbated, which invites a no out of shame, he'd ask, when did you first masturbate?

The assumption being that it's normal, it happened.

So let's just talk about the details.

Exactly.

Kinsey's claim was that it normalized the behavior and made it easier for people to report accurately.

And he may have been right for getting certain kinds of data.

But in a clinical setting, I mean, that seems incredibly dangerous.

You could just be confirming your own biases about a patient.

It is intensely dangerous.

The clinician risks just getting the answers they expect to hear based on their favorite theory.

They stop learning about the patient and start collecting evidence for their own case.

And the experimental data backs up this danger.

It does.

A study by Marquis and his team in the early seventies found a really crucial distinction.

What was that?

When the information is hard to recall, you know, fuzzy memories from childhood leading questions actually produce more distortion.

People's memories get reshaped by the question itself.

So the final advice on leading questions is avoid them unless you are absolutely certain about the premise and the information is something the person can recall easily and accurately.

The interrogative Kinsey style is a high -risk tool, not really suited for the subtle work of diagnosis.

Okay, so that's the questions.

Let's move from the what to the who.

The next big source of error is the interviewer themselves.

Yes, interviewer error.

The evidence here is just overwhelming.

Who the interviewer is, their demographics, their personality, their beliefs.

It all profoundly shapes the answers they get.

We're talking about things like race, social class.

All of it.

Race, class, gender, but also just their manner,

their appearance, and critically, their expectations about what they're going to hear.

The classic study on this is from all the way back in 1929, the Rice study.

It's almost a perfect, if depressing example.

It really is.

Rice had different interviewers surveying homeless men to find out the cause of their destitution.

One interviewer was a staunch prohibitionist.

He was convinced alcohol was the root of all evil.

And guess what?

He consistently found that alcoholism was the primary cause.

And the other interviewer.

The other was a committed socialist.

And he consistently found that the causes were social and industrial problems.

The same group of men, two completely different sets of facts, just based on who was asking the questions.

The interviewer acts as a filter.

And if that's happening for something as concrete as employment history, you can just imagine the filtering that happens when we're talking about complex mental states.

It's a huge problem.

And it even messes with the one thing clinicians are taught is the absolute bedrock of a good interview.

Rapport.

Ah, yes.

Rapport.

The idea that you have to build this great, trusting relationship to get good information.

Which sounds right.

It makes intuitive sense.

But the science is a bit more complicated.

A lot more complicated.

In fact, research shows that too much rapport, being overly friendly, can actually distort the information.

How?

The person being interviewed starts trying to please you.

They want to give you the answers they think you want to hear.

They see you nodding and smiling at a certain part of their story.

And so they give you more of that story, even if it's not the most important part.

Exactly.

You're unintentionally reinforcing a certain narrative.

And the effect of this isn't even consistent across different people.

You're talking about that other Marquis study, the one on social reinforcers, the nods, the mm -hmm.

That's the one.

They looked at how these little

affected the accuracy of reporting on chronic illnesses.

And they found this fascinating split based on education.

What was the split?

The better educated subjects.

They actually gave less accurate information when the interviewer was using all those social reinforcers.

Why do you think that is?

The thinking is that for them, it came across as maybe patronizing or distracting.

A more professional, to -the -point style worked better.

But it was the complete opposite for the other group.

The complete opposite.

For subjects with less formal education, those same reinforcers, the nods, the encouragement, led to more accurate information.

For that group, the warmth and encouragement seem necessary to build trust and get them to open up.

The implication for training is massive.

There's no single good interviewer style.

There isn't.

A good interviewer has to be flexible.

They have to tailor their approach to the specific person sitting in front of them.

It's a dynamic process, not a one -size -fits -all script.

Okay, so we've got flawed questions and biased interviewers.

The third major contamination source is something we all deal with every day.

Flawed memory.

Recall of information, yes.

This is a huge one.

We rely on the patient's memory to build their entire history.

And memory is, well, it's not a video camera.

It decays over time, for one.

It decays, yes.

But more importantly, it gets distorted by emotion.

Traumatic or anxious events, in particular, are very hard to recall accurately.

And people are just generally worse at reporting feelings and attitudes accurately compared to, say, concrete facts.

Absolutely.

And this is where two powerful psychological forces come in.

Social desirability and ego enhancement.

We want to look good.

We want to look good and we want to hide the things we're ashamed of.

The data on this is so consistent.

Give us some examples.

Okay.

Studies consistently show that things like mental illness and

genetic urinary diseases are systematically underreported.

People don't want to admit to them.

And on the flip side.

On the flip side, people consistently overreport their good deeds.

Things like charitable donations are always overestimated.

We want to see ourselves and be seen as healthier and more generous than we really are.

Which is a fundamental disaster for a clinical interviewer because their job is to dig into exactly those shameful, difficult areas.

It is.

And it leads directly into the next error.

Recording.

If the clinician just relies on their own memory of the session to write notes later.

The notes are going to be flawed.

Deeply flawed.

The sources call it the most prone to error.

You get emissions.

You get condensation.

You get outright distortions.

And this is what the material calls the massive clinical hazard.

Let's break that down.

The hazard is this.

A clinician interviews a patient about very sensitive topics.

Trauma.

Illness.

Childhood.

Where memory is unreliable and social desirability is high.

Then they rely on their own flawed memory to write up the case notes.

Maybe hours later.

So the foundation is shaky from the start?

The foundation is flimsy.

But from that flimsy data, they build what the source calls an elaborate model of the patient's illness.

And once that model is written down, it takes on a life of its own.

It feels true.

And then it becomes a self -fulfilling prophecy.

A total confirmation bias loop.

All new information gets filtered through that initial, possibly wrong, model.

So the safeguard has to be what?

You have to validate.

You check the interview data against other sources.

Psychological tests.

Medical records.

Direct behavioral observation.

And you have to record the data systematically, immediately.

Not from memory.

Okay.

So we've established that the data collection process is just riddled with potential errors.

Let's move into section two and see what happens when we try to use this flawed data for the biggest judgment call of all.

Psychiatric diagnosis.

This is where we see the reliability crisis of the 1940s and 50s.

The core issue is simple.

If two trained clinicians can't agree on a diagnosis for the same patient, then the diagnosis is meaningless.

The whole system falls apart.

The whole system falls apart.

And the research from that era showed it was, in fact, falling apart.

Let's start with that foundational 1949 study by Ash.

What did he do?

It was a simple, elegant, and devastating design.

He had three psychiatrists independently diagnose patients after a joint interview.

They had a list of major and specific diagnostic categories to choose from.

And the result?

The agreement on specific diagnoses was very poor.

But the headline number is this.

All three psychiatrists only agreed on the exact same diagnosis in less than half of the cases.

Less than half.

So more often than not, there was disagreement among the experts.

That's right.

For a tool that's meant to be scientific,

that's a catastrophic failure.

And this wasn't a one -off finding.

Creighton's study in 61 found something similar.

It did.

Creighton's group looked at experienced clinicians and found that while they could agree on really broad categories like is this person experiencing psychosis, yes or no, the agreement just collapsed when you got to specifics.

Specific diagnoses, specific symptoms.

Both.

The average agreement on the presence of individual symptoms was less than 50%.

It ranged from 0 % on some to about 85 % on others with very obvious delusions.

But on average, it was a coin toss.

So Beck's 1962 review of the field basically concluded what?

He concluded that no study at the time could show psychiatrists agreeing on specific diagnoses more than about 30 to 40 % of the time.

30 to 40%.

That's barely better than random chance.

And if those are the categories you use to decide on treatment.

Then your basis for choosing a treatment is fundamentally unsound.

It casts serious doubt on the entire enterprise.

We can actually break this down further using the concepts from Creighton's data, which was organized in a table.

Let's talk through that.

Which diagnoses were the most reliable?

So they found that things with very clear, often observable symptoms did better.

Organic disorders had the highest reliability, around 75 % specific agreement.

Functional psychoses were also relatively high at about 61%.

Because the symptoms are just harder to miss.

Delusions, hallucinations, severe cognitive decline.

Exactly.

They are more pronounced.

But now, contrast that with the low reliability categories.

This is where it gets really bad.

It does.

For neuroses, the specific agreement rate was only 28%.

28%.

So you have three different doctors looking at the same patient, and they all come up with a different specific neurotic diagnosis.

That was the common outcome.

And even if you collapse it all into one big, generic bucket of neurosis, the agreement only went up to 52%.

So half the time, they couldn't even agree if a neurosis was present at all.

Creighton also pointed out a problem with the classification system itself, didn't he?

He did.

He noted that neurosis and personality disorder were often the second -choice diagnoses for each other, suggesting the definitions were blurry and overlapping.

The system itself was making a hard job even harder.

So the unstructured interview was a failure.

The conclusion was unavoidable.

The only way to fix this was to introduce structure.

That was the only path forward.

And the big breakthrough came from a team led by Wing in 1967.

They developed a structured rating scale.

But they didn't just turn the interview into a checklist, did they?

They kept some flexibility.

That was the genius of it.

It wasn't a rigid questionnaire.

They used predetermined screening questions that had to be read verbatim.

These covered all the main areas of symptoms.

Everyone got asked the same core questions.

Right.

But if a patient answered yes to a screening question, the interviewer was then free to use flexible follow -up questions to dig deeper.

It combined systematic coverage with clinical freedom.

And they built in specific ways to reduce all the errors we just talked about.

Yes, two big ones.

First, they restricted all questions to the past month.

That immediately cut down on errors from memory decay.

Simple but effective.

And second, they banned obscure psychiatric jargon.

They focused on clear, observable events and simple, reportable feelings.

But the biggest innovation was what you call the separation principle.

This was crucial.

They formally separated the active data recording from the active judgment.

The interviewer's job during the interview was just to record symptoms systematically.

The diagnosis, the categorization that happened later after all the data was cleanly collected.

And did it work?

Did it solve the reliability crisis?

It worked spectacularly.

They achieved reliability scores of .8 to .9 for most psychotic and neurotic symptoms.

I mean, that's a night and day difference.

From 28 % agreement on neuroses to 80 or 90%.

It was a revolution in methodology.

It became so reliable, in fact, they could be used for cross -national research.

Cooper and his team used it to compare diagnostic habits in the U .S.

and the U .K.

And what did they find?

They found huge differences.

American psychiatrists were diagnosing schizophrenia much more frequently than their British counterparts, who were using diagnoses like depression more often for the same patients.

And because the data collection was now reliable, they knew the problem wasn't in what they were seeing.

The problem was in how they were judging it.

The data was solid.

The diagnostic criteria were being applied differently.

Structure revealed the real source of disagreement.

And this same approach was used in other fields, too.

Child psychiatry, for instance.

Yes, Rutter and his colleagues applied the same principles.

They found structure brought reliability to child assessments.

And then they took it a step further.

Assessing family life, which seems impossibly complex and subjective to measure reliably.

You'd think so.

But in their 1966 study, Rutter and Brown managed it by making a very clever distinction between objective and subjective measures.

What's an objective measure of a marriage?

It's something concrete and quantifiable.

They asked spouses about the frequency of specific behaviors in a recent short time frame.

How many times did you quarrel this week?

How often did you go out together?

And subjective.

That was either self -reports of feelings or, very cleverly, the interviewer rating the spontaneous expressions of emotion they observed during the interview.

And the reliability held up.

It was incredibly high.

The objective measures were up in the 0 .9 to 0 .95 range.

Even the spontaneous feelings ratings were between 0 .8 and 0 .9.

It proved that with careful structured design, you can reliably measure even the most complex parts of human life.

So if we pull all of this together, we get a clear list of modifications needed to fix the clinical interview.

Six key points.

First, you have to use an interview schedule or structure, period.

Second, you have to have a clear purpose for the interview.

Third, no jargon.

Fourth, focus on recent events.

Fifth, proper interviewer training on the new system.

And finally, number six, immediate recording of data separate from later judgment.

It fundamentally transforms the interview from an art into a science.

It has to.

The stakes are too high to rely on a flawed tool just because it's familiar.

Okay, so structure can make the interview reliable, but that brings us to section three.

Is it valid?

Does it actually help us make accurate predictions?

This is the next critical question.

And the best place to find answers is in the world of personnel selection, where the interview has been constantly tested against objective data for decades.

And what have all the big reviews of that research found?

They're almost unanimous.

The unstructured interview is a very poor predictor of future job performance.

This just doesn't work very well.

Compared to things like biographical data or psychometric tests, no.

It adds surprisingly a little.

The landmark study here is from Kelly and Fisk in 1951, looking at selecting psychology students.

They broke down how much each piece of information added to their predictive accuracy.

It's a brilliant study.

They started with all the objective data, credential files, test scores, autobiographies.

They calculated a validity coefficient based on that.

And then they added the interview, a one -hour or even a two -hour intensive interview with these candidates.

And the result, it's just stunning.

The addition of that entire interview process increased the median validity coefficient by only .01.

It's basically nothing.

An hour of a professional's time for an improvement that is statistically meaningless.

It just makes you wonder why we keep doing it.

Well, the research also started to show why it was so bad.

It uncovered these deep, consistent biases in how interviewers make decisions.

The whole idea of the skilled, objective interviewer turned out to be mostly a myth.

What are the big biases?

First, interviewers make up their minds incredibly early.

Usually in the first few minutes, they get a gut feeling, a first impression.

And the rest of the interview is just, what, window dressing?

It's just substantiation.

They spend the rest of the time looking for evidence to support their initial snap judgment.

They stop being investigators and become lawyers for their own hypothesis.

And there's a specific kind of information they latch onto.

Yes, the third bias.

They overwhelmingly rely on negative or unfavorable information.

So a candidate can have a perfect record, glowing references, great test scores.

But if they say one awkward thing in the interview.

That one negative data point can outweigh everything else.

We seem to be wired to give more weight to the negative when making these kinds of judgments.

But just like with reliability, structure can fix this, right?

It can.

The finding is consistent.

The more structured the interview, the more valid the predictions.

There's that Young study from 56.

That's a great counter example.

It is.

Young used a short, super structured interview, only about 15, 20 minutes, with a detailed item sheet for scoring.

With that system, they were able to make substantially valid predictions of job success.

So the lesson from the entire world of HR is, if you're using an interview to predict something, structure is non -negotiable.

Absolutely.

Now let's bring that back to the clinic.

What about the validity of clinical predictions?

Predicting if a treatment will work, for example.

And the research here is...

It's almost non -existent, which is a scandal in itself.

The one area it was studied was in the military, trying to predict who would be discharged for psychiatric reasons.

And what did they find?

Well, one group, Hunt and colleagues, claim to their brief psychiatric screening was valid.

But they had to admit it was a very blunt instrument.

It could tell the difference between normal and abnormal.

But it was useless at distinguishing between mild and moderate problems.

The very distinctions you actually need to make.

Exactly.

And then Plagg, in 1961, delivered the knockout blow.

He showed that the predictions from that psychiatric interview were no better than predictions you could make, just from biographical data alone.

Age, education,

marital status.

That was just as predictive as the psychiatrist's judgment.

Which brings us right back to Paul Meehl's famous challenge from 1954.

It does.

Meehl's argument, based on tons of data, was that clinical judgment, the kind that comes from an unstructured interview, is generally worse than predictions made by a simple statistical formula.

And the evidence seems to bear him out.

So if the unstructured interview is unreliable and invalid, what's the path forward?

Let's move to section four and look at the more experimental work on the interview process.

Right.

These are the analog studies.

They try to quantify what's actually happening in the interaction.

But they have a huge hurdle to overcome.

The generalization problem.

Exactly.

Can what you find in a controlled lab study actually apply to a real messy emotional therapy session?

Often, the answer is no.

The classic example being Matarazzo's work on speech patterns.

Yes.

He used this device, an interaction chronograph, to measure the length of time people spoke without caring about the words.

And he found this amazing synchrony.

The longer I speak, the longer you speak in reply.

He called it the inverted J -curve.

It suggests this deep, unconscious rhythm in conversation.

It does.

And he found it again and again in his lab studies.

But, and this is the crucial but, when he analyzed recordings of seven actual psychotherapy sessions, sessions he himself had conducted.

The effect disappeared.

It disappeared completely.

The beautiful, replicable lab finding just didn't hold up in the real world.

And that calls into question the value of a lot of this process research.

It seems like a lot of it focuses on these non -content variables.

Eye contact, body language,

ums and ahs.

It does.

And while that's interesting, it doesn't tell us much about assessment until we start analyzing what people are actually saying.

There has been some work on content, though.

That Seidman and Pope study on ambiguous questions.

Right.

They tested the common clinical assumption that if you ask vague, open -ended questions, you get better, deeper material from the patient.

And did they?

Well, yes and no.

They found that ambiguous questions did lead to more talking.

The person's responses were longer.

So more productive.

More productive in terms of volume.

But, and here's the catch, when they analyzed the content, the proportion of meaningful information to neutral filler was exactly the same as with specific questions.

You just get more talk, not necessarily more substance.

Precisely.

More talk doesn't equal more useful data.

It's a really important finding for clinicians who favor a very non -directive style.

Okay, so pulling all of this together, the reliability crisis, the validity problems, the limits of process research, our sources lay out four clear paths for developing a better clinical interview.

The first path is specification of objectives.

We have to stop thinking of the interview as one thing.

We need to break it down.

Into what?

Into different types.

You could have an exploratory interview where the only goal is building rapport.

Open questions would be great there.

Then a data gathering interview, which is just for facts, history, medications.

Closed questions would be more efficient there.

And a third type?

A probing interview designed specifically to explore a sensitive problem area using a mix of techniques.

If we define the goal, we can figure out the best way to achieve it.

Okay, path two is about managing interviewer and patient variables.

We have to accept that one size does not fit all.

Clinicians need to be trained on all the sources of error and how to reduce them.

And we need more research on how different types of patients respond to different interview styles.

Like we saw with education levels and rapport.

Exactly.

A severely depressed person might need a highly structured, closed question approach to give any information at all.

A non -directive style might be a complete failure with them.

We need to test these things.

Path three gets to the heart of it.

Clinical judgment.

The solution is what Wing and Rutter did.

Separate data collection from judgment.

And once we have reliable data, we have to validate the judgments made from it.

We have to take Neal's challenge seriously and compare clinical judgment head to head with statistical prediction.

We have to prove the human adds value over the algorithm.

If we can't, then we need to rethink our processes.

And the final path, number four, is about putting the interview in its proper place.

Right.

Interviewing in relation to other techniques.

It cannot be the only tool.

It has to be part of a package alongside psychological tests, medical exams,

and crucially, behavioral analysis.

And it shouldn't try to do what other tools do better.

That's the key.

Behavioral analysis is great for measuring the frequency of a behavior, like a child's pantrums.

The interview can't compete with that for accuracy.

But the interview can measure something else.

It can measure something behavioral analysis can't, like the parent's attitude toward the child having a tantrum.

We need to use the interview for its unique strengths to fill in the gaps left by other more objective measures.

So if we follow these four paths, we can turn the interview from this subjective, faith -based art form into a sharp, reliable, and genuinely useful scientific tool.

That's the goal.

It's not about getting rid of the interview.

It's about making it better, making it worthy of the trust we place in it.

And as we wrap up, I just want to circle back to that incredibly powerful finding from the selection studies, that interviewers are so strongly biased to focus on negative information.

It seems to be a fundamental part of how we assess other people.

We're looking for the flaw, the disqualifier.

So the final thought for you, our listener, is this.

Knowing that the person across the table is likely hardwired to latch on to any negative information you provide, how does that change how you prepare for a high -stakes conversation?

Whether it's a job interview, a doctor's appointment, any situation where you're being assessed.

If you know their bias, how do you manage it?

Something to think about.

Thank you for joining us for this deep dive into the science of the interview.

We'll see you next time.