How To Apply SMART Goals to Your Experimental Design

Ever feel like an experiment…doesn’t end?

Experiments cost time (and money) to run, whether you’re in an academic lab or in the biotech/pharma sector, and long experiments cost even more. After moving from academia to biotech, I noticed that the studies that ran forever usually started with an unbounded hypothesis.

In other words, because the experimental question didn’t have an endpoint, the study didn’t either! If I wanted to reduce the cost of my work, I needed to reduce the time I spent working on it.

In short, plan better.

This article explains how I use the SMART goal framework to do that, and why having a bounded experimental question can lead to more efficient experimental design.

Why Using SMART Goals Works

I have used the same plea to colleagues ad nauseam: please do not torture yourself by reinventing the wheel.

I keep saying this because people hear it and then ignore it. Most of us tend to apply those
words to kits, reagents, and protocols, but rarely to experimental design. We might even think
that our own design feels like doing the work “properly”.

This is where SMART design can help.

Adapting SMART Goals For the Bench

The SMART acronym comes from the management world, originally meaning “Specific, Measurable, Actionable, Relevant, and Time-bound“. A SMART research question uses the same terms but, it helps to redefine three of those words (measurable, actionable, and time-bound), as shown in the table below:

Specific	Be specific about your analyte and your source material. Not “does this drug affect immune cells”, but which analyte, in which tissue, from which model system. Have a defined (i.e., be specific!) analyte or target(s), sample type(s), and model system (cell lines, mice, etc.) in your question.
Relevant	Ensure the question is relevant to your actual research topic. It doesn’t make sense to test T and B cell responses if you’re investigating early immune responses, for example.
Measurable	In a goal-setting context, measurable usually means you can track progress. But here, it means the question explicitly names what you will measure (i.e., your observable outcome), such as concentration, yield, percent reduction, or time to an event. Sometimes that measurement already exists as a standard assay. Sometimes you have to develop and characterize it first, which makes it its own SMART question. If you cannot name the quantity, you do not yet have a measurable metric.
Actionable	If you have seen the “A” in SMART, it might have meant achievable: can this realistically be done? However, in this context I use it to ask: Does the answer tell you what to run next? A question is actionable when either outcome moves you. If both a yes and a no leave you exactly where you were, the question is not actionable no matter how elegant the experiment.
Time-bound	In research, this is when a result counts as an answer. For example, 14 days post-infection, 24 hours at 4°C, 12 hours post-treatment. Without that boundary, every result becomes a reason to run one more experiment, and this is how projects become endless.

The Audit: Five Filters to Test Your Experimental Question

You do not need a new project to start using this framework. Instead, take the question you are working on now and ask these five questions:

1. Is there a specific analyte, tissue source, tool, or model system in your question?
2. Are your metrics clear: concentration, yield, percent reduction, time?
3. Does the question define an actionable approach?
4. Is the question relevant to your research goal?
5. Is the endpoint defined: two hours, fourteen days, one month?

Here’s a worked example:

• Old question: “Using a mouse model to evaluate what Leptospira interrogans is doing” is open-ended. It passes on specificity but fails on nearly every other term.
• SMART Question: “Do leptospires persist in peripheral mouse tissues at 14 days post-infection?” Same biology, same model, same underlying interest, but with an endpoint.

How This Changes Your Approach to Experimental Design

Your research question generates the requirements, constraints, and dependencies of your lab work downstream, so it’s worth getting it right. Here are two rules that will help you formulate SMARTer questions:

Rule 1: Metrics Before Methods

If you name your method before you name your metric, you have most likely skipped a step. Notice that in the table below, the method sits near the bottom as an output (not a starting point). There may also be more than one method that can answer your question! Multiple roads lead to Rome, and several may satisfy the same experimental requirement.

Also see that a DNA-based measurement quantifies bacterial load; it does not by itself establish that the organisms are viable or actively replicating. If viability is the question, you need a different method. And the immune readouts you choose here (e.g., anti-Leptospira IgM and IgG by ELISA, ELISpot, flow cytometry on splenocytes) are an interesting second question about host response, not a route to bacterial load. They belong in their own SMART question.

Step	The leptospires example
Question	Do leptospires persist in peripheral mouse tissues at 14 days post-infection?
Analyte	Leptospira interrogans genomic DNA
Sample type(s)	Kidney, liver, and lung tissue; bladder
Model system	Mouse
Metric	Bacterial load: genome equivalents per unit of tissue input
Requirements	Quantify a Leptospira-specific sequence against a standard curve, normalized to tissue input
Method	qPCR for a Leptospira 16S rRNA gene target, on DNA extracted from tissue
Endpoint	14 days post-infection
Big picture	Host response to bacterial infection

Rule 2: One Timepoint Per Question

In the table above, you may have noticed that the endpoint row does not say “timepoint” because the question only specifies one timepoint. If you want to measure a trajectory (days 1, 3, and 7 as well), that is a question about kinetics rather than about persistence at a fixed point, and it deserves its own study.

Keep in mind that not every constraint is time-based. A question comparing transcriptomic expression between MS patients and healthy control donors at baseline uses “at baseline” as the endpoint.

Factors such as Institutional Review Board approval and access to samples or an existing dataset are feasibility constraints that belong in the chain as dependencies rather than as the “T”. Confusing these two is how a research question ends up looking bounded when it is actually blocked! So it is worth taking the time to think it through.

“But My Work is Exploratory!” I Hear You Say

The question I hear the most around endpoints is, “If you are doing discovery, how can you be specific about an outcome you do not know?”

The answer is that you separate the outcome from the question.

In discovery, you want to know if there is a difference between your groups, so that answer is conclusive either way. Specificity comes from how you characterize the groups, for example, what sets group A apart from group B.

The requirements follow: expression profiling points you toward transcriptomics and mRNA-seq; genomic differences toward whole-exome or whole-genome; spatiotemporal patterns in tissue toward spatial approaches.

You have not been using a time constraint, but you have made the question closable, which is what the “T” in “SMART” is meant to do.

The Wheel is Already Invented (and Mostly Free!)

Once you know your research requirements, look for existing tools instead of building from scratch. There are roughly three tiers of tools you can use:

Tier 1: In-House

This tier includes your PI’s procedure binder or root directory, previous papers from the lab, kit instructions, or an experiment template. I was fortunate that every lab I worked in had a PI with procedure binders already built, vetted, and published. If you joined a lab without it, you are not doing anything wrong when you have to look elsewhere.

Tier 2: Freely Available Online

This includes published protocols you can cite and follow: protocols.io, Bio-protocol, and Protocol Online. PrimerBank is good if you need validated qPCR primers for human or mouse gene expression. And OSF, from the Center for Open Science, lets you put a study plan on the record before you start.

I am not endorsing any of these; this is simply the material I was glad to have as a graduate student.

Tier 3: Membership or Paywall

Formal standards mostly live here, including resources from associations, organizations, and
vendors. Free copies tend to be older revisions, and you will likely need an account to log in, access, and potentially purchase the new edition(s).

It’s worth asking about these, as some organizations offer discounts for graduate students and postdocs, and some vendors run schemes where a short written application gets you reagents, or a plate’s worth of analysis. I obtained antibodies that way as a student, on the strength of papers that would otherwise have gone unused!

How to Analyze Your Experiment Data (SMARTly)

Here are two decisions that a SMART question will make for you that will affect your experimental design and analysis.

1. Number of Replicates

Keep in mind that repeatability and reproducibility are not synonyms:

• Repeatability is agreement between measurements under the same, tightly controlled conditions, e.g., replicates within one run.
• Reproducibility is whether the finding survives when conditions change: a different day, operator, reagent lot, or lab (Chesher, Clin Biochem Rev, 2008).

So while documenting your protocol keeps differences interpretable, pinning every run to the same operator tests repeatability, not reproducibility. This also means that technical replicates are not independent observations. Three wells from one sample tell you about your pipetting, not about the effect you are observing. Averaging those wells gives you one result.

Your n is the number of independent experimental units: animals, donors, separately prepared cultures, not wells. Triplicates are a common convention for technical measurements. How many independent replicates you need is a separate question, answered by how variable the system is and the effect size you are chasing.

2. Choice of Analysis

Don’t just pick a statistical test at the end of your study. How you designed your experiment already decided it for you, whether you were aware of it or not!

To choose an analysis technique, look at data structure before data distribution. Ask: Are the observations independent, or repeated on the same unit? How many factors are you varying?

More importantly, your data structure informs your data distribution. Your data points are
collected and classified based on your SMART experimental design. And no amount of statistics
will help you through the constraints or dependencies that you miss at the beginning.

You may be tempted to just run a Student’s t-test and treat it like a gate you must pass. But your data
distribution tells you how noisy your data is, and a normality test can help you understand which statistical test is most appropriate downstream. There is no shortage of these (Shapiro- Wilk, KS Test, JB Test, D’Agostino-Pearson, etc.), but there are enough differences that one may work better for you than the others.

This is also a helpful guide to match your experimental set-up to the analysis:

• Two independent groups → t-test (use Welch’s version if the groups differ in spread).
• Same units measured twice → paired t-test.
• One thing being varied, with several levels (e.g., low/medium/high dose) → one-way ANOVA.
• Two things being varied at once → two-way ANOVA. Note: it’s the number of things you’re varying that picks ANOVA, not the number of groups.
• Data is weird, lopsided, or ranked rather than measured (like 1st/2nd/3rd) → Mann-Whitney or Kruskal-Wallis. These ask “does one group tend to give bigger values?” rather than comparing averages.
• If your design has repeated measures or things nested inside other things → go find a statistician!

A confession: I once took a second semester of a computing class specifically to avoid taking statistics. Stats always comes back around whether you like it or not, so try to do a little refresher every few years. To learn more about the different statistical tests used in biology, check out our articles on methods for comparing multiple datasets and methods for comparing two sets of data.

What to Record Before You Start

The table below shows what my SMART experimental design form includes. Want a copy? Download it here.

I include lot numbers because I learned (the hard way) as a master’s student on a Master of Public Health rotation that not every lot is created the same. Two experiments run to the same protocol with different lots are not the same experiment, and if you don’t write the number down, you cannot find it afterward!

It’s also good practice to keep two copies of the data, with the analyzed one marked “_initials_analysed” so anyone can return to the original and see exactly what was done to it.

Experiment number	…in the header or the filename
Start date and completion date
Title	…if you want one
Objective	…your SMART question, your SMART approach, and what you expect to see
Materials	…with lot numbers, expiration dates, equipment, and serial numbers
Protocol reference	…binder ID, folder link, or the PDF itself
On completion	…the unmodified raw exported data, plus a separate copy for analysis
Summary	…two- or three-sentence description of what you observed, one or two sentences of conclusion, and the next SMART question

Write your SMART Question Down Before You Run Anything

Another good practice is to write your hypothesis (and frame it as your SMART question) down before you run the experiment. Why? Because once you’ve seen the result, you start telling yourself “yeah, that’s basically what I expected.”

A hypothesis or question written beforehand protects you from your expected outcomes and biases. If you write your hypothesis in retrospect (or nudge it to match what you got), it’s just describing the result, not showing that you tested anything!

Once you’ve got your outcome, be careful what words you use to describe it. Don’t say the experiment “worked” or “failed,” or was “positive” or “negative.” Instead think:

• Conclusive: the experiment gave you a clear answer. Even if the answer is the opposite of what you predicted, that’s still a win, not a failure. A clear “nope, wrong” often tells you what to do next faster than a boring “yep, as expected” would.
• Inconclusive: a totally different situation. The experiment couldn’t answer the question at all. And that usually means something was off with your setup/design, rather than telling you anything about the biology you were testing.

If you want to find out more about changing your attitude towards experimental outcomes, check out the article, “Stop Blaming Yourself: 3 Troubleshooting Tools for when Experiments Go Wrong“.

Where to put the effort

In summary, decide what outcome you expect before you look, and judge a result by whether it answered your original SMART question, not if it’s what you expected.

This audit takes ten minutes, but the bench work it determines takes months. So before your next set of experiments, run these five questions, find the endpoint, and name the metric. And whatever you do, please-please-please do not torture yourself by reinventing the wheel. Your future self will thank you!

Now that you have the know-how to create SMARTer research questions, download the SMART Experiment Template. It includes every field Priya has discussed in a printable form. Handy!